Merge remote-tracking branch 'origin/rustify' into worktree-agent-ace98b0bdaf33ffff

# Conflicts:
#	docs/IRIS.md
#	docs/RUST.md

.cargo/config.toml

@@ -0,0 +1,6 @@
+# xtask convention (https://github.com/matklad/cargo-xtask), without folding
+# every crate in this repo into one workspace -- they are deliberately
+# independent (see run-tests.sh, which cds into each). `cargo xtask apk`
+# from the repo root runs xtask/src/main.rs directly.
+[alias]
+xtask = "run --quiet --manifest-path xtask/Cargo.toml --"

.claude/skills/ai-app-rigs/SKILL.md

@@ -0,0 +1,244 @@
+---
+name: ai-app-rigs
+description: ai-app's test rigs, harness scripts and reference measurements - ui-sandbox.sh, debug-transcript.sh, transcript-bench.sh, stream-bench.sh, trace-draw.sh, the /usage fixture vocabulary, the fake CLI, the rule that no UI-driving script may tap a coordinate, how to test llama.cpp and ssh on this machine, how importing behaves, and the scroll/stream/explorer numbers not worth re-measuring. Read before running or writing a benchmark, driving the app's UI from a script, exercising the session lifecycle, testing a llama or remote session, or touching the import screen.
+---
+
+# ai-app: rigs, harnesses and measurements
+
+Moved out of `AGENTS.md` on 2026-09-04 so it is read when it is relevant
+rather than sent with every request in this repo -- it was 12 KB of the 35 KB
+that file cost on every one. Unchanged in the move, and still the only copy.
+
+## The rigs
+
+Each exists because something was invisible without it.
+
+- **`app/ui-sandbox.sh`** — a second `ai-server` with its own `$HOME`, config
+  and data directory, holding eight invented Claude Code transcripts and a
+  `claude` that is two lines of shell. **That isolation is the point**: the
+  import screen lists whatever is in `~/.claude/projects`, which in this VM is
+  real agent transcripts, so exercising *delete* against the ordinary server
+  deletes somebody's conversation and exercising *import* starts a real
+  `--resume` on the owner's account.
+  Its port and root derive from the checkout's name, so two checkouts'
+  sandboxes cannot reach each other, and its token is generated once into
+  `~/.config/ai-app/sandbox-token` and carried across restarts along with any
+  the enrolment flow appended — so the emulator app is enrolled **once** (the
+  start banner prints the command) and stays enrolled. It shares the real TLS
+  certificates, because the installed APK pins that CA.
+  Driving verbs, so none of this is re-derived per session:
+  `./ui-sandbox.sh spawn [title]` (an echo session, prints its id),
+  `./ui-sandbox.sh send SID text|@file`, and
+  `./ui-sandbox.sh api /path [curl args]`.
+  `./ui-sandbox.sh keep` restarts the server without wiping the sessions and
+  enrolment already there — for when the fixture under test was expensive to
+  build; plain `start` wipes them, which is right for the list-screen
+  fixtures and wrong for that.
+  It passes `--delay` by default, and `AI_SANDBOX_BIG_MB` puts one large
+  transcript among the small ones while `AI_SANDBOX_SPAWN_DELAY` makes the
+  fake CLI slow to start. Both exist because operations that finish in
+  milliseconds have states on the way that nothing can observe, and an
+  unobservable state is one where broken and working look identical.
+  It also builds a fixture tree at the sandbox home's `~/files` for the
+  explorer, holding the states otherwise only reachable by finding a real
+  machine in one: an empty directory, a name with a tab and one with an
+  apostrophe, a binary file, one over `FILE_LIMIT`, one `chmod 000`, a
+  symlink to a directory and a broken one, a source file per language, and
+  the three sizes the limits were measured against (`edit-32k.rs`,
+  `edit-128k.rs`, `big-source.rs`). Point a session at it with
+  `./ui-sandbox.sh api /sessions/<id>/cwd -X POST -H 'content-type: application/json' -d '{"cwd":"~/files"}'`.
+  The explorer's 409 is produced by editing the file on the machine
+  (`printf … > file`) between pressing the pencil and pressing save.
+- **`app/debug-transcript.sh`** — a real conversation on the emulator. The
+  echo driver is the right rig for most things and the wrong one for anything
+  whose cost scales with what was actually written: a real reply is longer,
+  is real markdown, and carries tool calls whose input and output are
+  kilobytes. Two faults were invisible until a real transcript was loaded — a
+  page of history landing mid-fling threw the reader back to the newest end,
+  and parsing one real reply took 51ms against 4.6ms for a synthetic one.
+  `-b` takes the biggest conversation on the machine rather than the newest,
+  which is what a scrolling test wants; `--stop` takes it down.
+  It copies the transcript into `/tmp` and gives the server a `HOME` of its
+  own, so the import can only see the copy — importing spawns `claude
+  --resume`, and against the real file that is a second CLI writing to a
+  conversation somebody may still be in. **A transcript never goes in this
+  repository**: they hold whatever was said, read and written in that
+  session, and `~/repos` is shared with the host besides.
+- **`/usage` in an echo session puts up an invented meter**, which is how the
+  rate-limit screens' states are reached without spending quota: `/usage 42`,
+  `/usage 95 20` (minutes left), `/usage 42 never` (the between-blocks window
+  with no reset time), `/usage 42 unreadable`, `/usage notloggedin`,
+  `/usage unreachable`, `/usage failed`, `/usage off`. The vocabulary is
+  `usage::Fixture`'s, since those are its states. With none set an echo
+  session meters nothing, which is the ordinary case and draws no bar.
+- **A fake CLI exercises the process lifecycle without a token.** Point a
+  `claude_cli` provider's `command` at a two-line script — `#!/bin/sh` and
+  `cat > /dev/null` — and it behaves the way the lifecycle code cares about:
+  it holds the fifo open, records a real pid, writes nothing, and dies on a
+  signal. So adopt, stop, restart and start are all drivable without a real
+  `--resume` and without spending a turn on somebody's account. Reach for
+  this when what is under test is *whether a process is running*, and for
+  `debug-transcript.sh` when it is *what the transcript draws*.
+- **`app/transcript-bench.sh`** is the standard scroll measurement: it opens
+  the first session (or `-k` keeps the current screen), scrolls a fixed
+  gesture loop, and prints the app's render report — the same one the in-app
+  copy button produces, whose `on screen:` line names what the viewport was
+  holding. Compare two runs with the same gestures; the emulator's absolute
+  frame times transfer nothing, the report's accounting does. Run it either
+  side of any change under `Markdown*.kt`, `Transcript*.kt` or
+  `SessionScreen.kt`'s list, and put the report in the commit. The numbers
+  that move first are the worst `record: one block`, the reparse mean while
+  streaming, and the draw phase's accounting line.
+- **`app/stream-bench.sh [-k] FILE`** is that measurement for a reply still
+  arriving. It taps "Jump to latest" so the list is pinned to the newest end,
+  resets the report, sends FILE, waits for the transcript to stop growing,
+  and prints. Both of those are corrections to a first version that measured
+  nothing: a transcript parked further back never redraws while a reply
+  streams into it, and a session is idle at *both* ends of a turn, so polling
+  for idle answers before the turn has started.
+- **`app/trace-draw.sh`** names what a scrolling frame spends inside the
+  framework, from `atrace` text output with no trace processor needed. It is
+  how the cost of a layout node per link was attributed to the framework
+  rather than guessed at.
+
+### Driving the UI
+
+**No script that drives this app's UI presses a coordinate.** Every control
+is found by the name it already carries for assistive technology —
+`ui-trace record --do "tap 'Session settings'"` — which resolves the label
+against the screen at the moment of the gesture and fails the whole run when
+it is not there. `app/bench-lib.sh` is what the bench scripts share for it. A
+coordinate is a position measured once by hand, and anything that moves the
+control makes the tap land on whatever now sits there — the bench then
+reports a number that was never measured, which reads exactly like a result.
+Both bench scripts pressed the render report at `tap 723 205` until that
+button moved into the session settings dialog on 2026-09-03. The check that
+none has crept back:
+
+    grep -n "tap [0-9]" app/*.sh
+
+Swipes are still coordinates, deliberately: a gesture across a scrolling area
+is a distance rather than a control.
+
+**Two traps in the emulator bench loop**, each of which cost a run.
+`adb shell pm clear` removes the enrolment and the notification permission
+along with the saved anchors, so the next run measures a permission dialog —
+re-enrol with the command `ui-sandbox.sh` prints, and
+`pm grant … POST_NOTIFICATIONS`. And a saved scroll anchor is per session id,
+so the only way two builds start a scroll from the same place is a *fresh
+session for each*.
+
+**The emulator is `~/repos/emulator-tools`' business, not this repo's.**
+`emu up` creates and boots the AVD named after this checkout — whatever `emu
+name` prints, never a name typed out here, since this file is the same in
+every clone. `run-android.sh` is that plus a build and an install. The `adb`
+on `PATH` after sourcing `android-env.sh` is that repo's wrapper, which fills
+in `-s` from the same rule. Gradle does not go through it, so a Gradle init
+script from `emulator-tools` runs `emu check` before `installDebug`,
+`uninstallDebug` and `connectedAndroidTest` and fails rather than fanning out
+to every attached device; when it refuses, say which device you mean at the
+moment you use it — `ANDROID_SERIAL=$(emu serial) ./gradlew …`.
+
+### Testing llama.cpp and ssh here
+
+**Both are set up here as of 2026-09-04** and need nothing typed. The
+prebuilt CPU llama.cpp lives outside the repo at `~/.local/opt/llama.cpp`
+(the 15 MB `ubuntu-x64` release asset) and is symlinked as
+`/usr/local/bin/llama-server`, which is what makes **discovery find it over
+ssh**: `~/.local/bin` is not on the PATH a non-interactive ssh session gets.
+It resolves its own libraries through `$ORIGIN`, so no `LD_LIBRARY_PATH` is
+needed. One model is downloaded — `unsloth/Qwen3-0.6B-GGUF/Qwen3-0.6B-Q8_0.gguf`,
+639 MB under `~/.local/share/ai-app/models` — and answers at usable speed on
+this VM's 8 cores. **Do not test with a 2-bit quant**: the
+IQ2_XXS of that model produces fluent nonsense, which reads exactly like a
+broken driver — `llama-cli` produces the same from the file directly, which
+is how to tell the two apart in a hurry.
+
+There is no second machine, so **ssh this VM to itself**. That is set up
+too: the key is `~/.config/ai-app/ssh-self` (its public half is in
+`~/.ssh/authorized_keys`, labelled removable), and the real config carries a
+setup called **"this vm over ssh"** — `bob@127.0.0.1` with that
+`identityFile` plus
+`options: ["StrictHostKeyChecking=no", "UserKnownHostsFile=/tmp/ai-app-known-hosts"]`
+so it touches nothing real — offering `claude-cli` and `llama-cpp`. It is the
+whole rig for "does a remote llama session work", since the far machine is
+this one and the model file is the same file. For a throwaway setup of your
+own, point a provider's `command` at something harmless like `/bin/echo`
+rather than at `claude`: the transport is what is under test, the process
+exiting immediately is the signal, and it costs no tokens. The remote login
+shell here is **fish**; the
+remote script and `ssh.rs`'s POSIX quoting happen to mean the same thing in
+both, but that is luck rather than design, and a shell that is neither is the
+thing to suspect first if a remote spawn ever mangles an argument.
+
+## Importing
+
+The import list reports each session's **size as well as its line count**,
+because the two disagree in the way that matters: these transcripts embed
+screenshots as base64, so one line can be a megabyte. On this machine a 69 MB
+session has 3,427 lines and a 44 MB one has 6,792 — nothing about a line
+count tells you what continuing a session will cost. Shown, not warned about;
+importing a large session is a choice somebody is entitled to make.
+
+**Never import a Claude Code session that is open in a terminal.** The app
+refuses it — see PLAN.md for the incident that made that a refusal rather
+than a warning.
+
+**One Claude Code session id can name two files, and the listing offers it
+once.** Resuming from a different working directory makes the CLI write a
+second transcript with the same id under that directory's project folder — an
+ordinary state of a machine, not corruption. Everything downstream addresses
+a session by id, and the phone keyed its list on it, so two rows sharing one
+**closed the app** on a Compose duplicate-key throw. `parse_listing` keeps
+the copy with the most lines, because the other is usually a few-hundred-byte
+stub and is often the *newer* of the two, so recency is the wrong key.
+Deleting removes every copy rather than the first, or the row came back after
+a delete that reported success. The phone's half is `uniqueItems`, which
+every list keyed on a server-chosen id goes through: a repeat there must
+never be able to close the app, whatever produced it.
+
+**Deleting a session offers to take the machine's own transcript with it** —
+`DELETE /sessions/{id}?deleteForeign=true`, behind a switch in the
+confirmation, and only where the driver keeps a record of its own
+(`keepsOwnTranscript`, which today means Claude Code). Off by default,
+because leaving that copy is what makes an ordinary delete recoverable — and
+the dialog's paragraph is rewritten when it is on rather than appended to,
+since the sentence promising the conversation "should still be there to
+import again" is exactly the one the switch makes false. The server deletes
+the machine's copy *first*, so a machine it cannot reach leaves the session
+where it was instead of half-deleted.
+
+## Measurements worth not re-taking
+
+- **What the transcript screen costs to scroll.** Taken 2026-08-30 on the GPU
+  emulator against a real imported transcript with the server at
+  `--delay 120`. Settled and flinging fast, both into fresh history and back
+  through rows already drawn: **5.2–5.9% janky frames, 99th percentile
+  29–32ms, 0–2 slow UI-thread frames.** The stock Settings app on the same
+  device is 3.3% and 38ms, so this is at the platform floor. The number that
+  is *not* at the floor is the first few seconds after opening a session,
+  where every row on the way is being composed for the first time; that is
+  inherent to a lazy list and it is why a measurement taken before the screen
+  settles reads three times worse. **Settle first, then reset `gfxinfo`.**
+- **The reset path is not reachable by reopening a session.** Measured
+  2026-09-04 against a session streaming at 20 events a second: reopening one
+  with an anchor 1,800 events back connects **87–119 events behind**, well
+  under `CATCH_UP_LIMIT`'s 200, because the restore is two requests — the
+  opening page, then one span covering the whole distance. To exercise the
+  reset at all you have to lower `CATCH_UP_LIMIT` in a throwaway build; at 5
+  the app takes the reset on a live connection, clears, refills and carries
+  on without reconnecting.
+- **The session screen's stream survives backgrounding here** — 20 seconds at
+  the launcher while 415 events were produced brought no reconnect at all,
+  which is not what the comment above that loop expects, and is most likely
+  this emulator being headless rather than the phone's behaviour.
+- **Reopening a cached session costs one request for one event** (the probe),
+  and scrolling the whole conversation back costs nothing more; a cold open
+  of the same 500-event session is two pages, 100 events. Measured
+  2026-09-04 on the emulator against the sandbox.
+- **Reading is cheap and editing is not.** The viewer handles a 1 MiB,
+  28,000-line file because it draws one row per line; the editor is one
+  `BasicTextField`, which costs two seconds a frame at 128 kB and stops the
+  app at 1 MiB, so `EDIT_LIMIT` caps it at 32 kB with the reason said on
+  screen. If you make the editor faster, that number is what to move.
+  EXPLORER.md's "What the measurements said" has the rest.

.dev-updater.ron

@@ -55,4 +55,21 @@ components: [
         // the terminal the QR would be printed on.
         enroll: "server/enroll-link.sh",
     ),
+    // E5 (RUST.md): app/shellApp packaged by the xtask instead of Gradle
+    // (cargo ndk -> javac -> d8 -> aapt2 -> zipalign -> apksigner), signed
+    // with the same release key as "app" above so the two can install
+    // over each other -- a separate component, not a mode of "app" above,
+    // because it is a different applicationId (com.example.aiapp.shell)
+    // built by a different tool from different sources. No `cwd`: it
+    // defaults to this checkout's root, which both the `cargo xtask`
+    // alias (`.cargo/config.toml`, resolved relative to the working
+    // directory cargo is run from) and `cargo xtask apk`'s own publishing
+    // step (`xtask/build/outputs/apk/<mode>/*.apk`, matching discover.rs's
+    // `*/build/outputs/apk/*/*.apk` pattern -- see apk.rs's module doc)
+    // both need.
+    Apk(
+        name: "shell",
+        modes: ["release", "debug"],
+        build: "cargo xtask apk",
+    ),
 ],

.gitignore

@@ -1,6 +1,7 @@
 .gradle/
 build/
 app/androidApp/build/
+app/shellApp/build/
 local.properties
 .kotlin/
 *.iml
@@ -9,6 +10,11 @@ local.properties
 server/target/
 event-model/target/
 client-core/target/
+android-shell/target/
+
+# E3's native library, built by cargo-ndk straight into the Gradle module
+# (RUST.md) -- an artifact, like server/target/ above, not source.
+app/shellApp/src/main/jniLibs/
 
 # Server logs from a development run (ai-server.log by convention,
 # wg-test.log from ./test-wg-tunnel.sh).
@@ -27,3 +33,10 @@ sessions/
 # iris, the in-house UI library, is vendored at iris/ and built by cargo.
 iris/target/
 iris/android-app/target/
+
+# E5's packaging xtask (RUST.md). `build/` above already covers
+# xtask/build/outputs/apk (the published APK, see apk.rs's module doc).
+# The repo root has no Cargo workspace, so this is xtask's own
+# intermediate working files (target/xtask/apk/...), not a shared one.
+xtask/target/
+/target/

AGENTS.md

@@ -5,24 +5,42 @@ replacing the Claude app for daily use. Rust/Axum backend on the desktop,
 Kotlin/Compose Android app, WireGuard + pinned self-signed TLS + bearer token
 between them.
 
-**`PLAN.md` is the design source of truth** — every decision with its date,
-its rationale, and what was rejected. Read it before changing anything
+**`docs/PLAN.md` is the design source of truth** — every decision with its
+date, its rationale, and what was rejected. Read it before changing anything
 structural, and update it in place when a decision changes rather than
 letting this file and the plan become two versions of the truth. This file is
 the working notes layer: layout, commands, rigs, and things that have bitten.
+The design and working documents live under `docs/` — everything except this
+file and `CLAUDE.md`, which stay at the root because that is where Claude
+Code and other agent harnesses look for them.
 
 The central design point, worth not undoing by accident: **a session is a
 child process, translated into one common event model.** A new session type
 is a new driver — never a session-type branch in shared code (routes,
 transcript, app screens).
 
+The second one, for the Rust port on the `rustify` branch: **the phone app
+and a planned desktop app share almost all of their code.** Screens, widgets,
+folding, paging, config and the network client live in the shared crates
+(`iris`, `client-core`, `transcript-ui`, `tabs-ui`); `android-app` and
+`desktop-app` are thin entry points that own only what the platform forces
+(JNI and the IME on one side, winit and argv on the other). The two
+*layouts* will differ, to suit a phone's screen and a finger against a
+desktop's screen and a mouse -- but the widgets a layout is made of (a
+button, a text field, a list, a card) and the styling (colours, spacing,
+type) are one implementation with no per-platform copy. Anything that could
+work on both goes in a shared crate the first time it is written, and a
+platform crate growing a widget or a colour is a defect to move, not a
+convenience to keep. Iris said this on 2026-09-07; docs/RUST.md carries the
+details.
+
 ## Layout
 
 Mirrors `../dev-updater` deliberately: same stack (axum 0.8 +
 axum-server/rustls, tokio, clap; Kotlin 2.4.x + Compose Multiplatform, single
 `:androidApp` module), same cert scheme, same registry pattern. Read
 dev-updater's `README.md` and `AGENTS.md` before diverging from them.
-Module-by-module intent is in PLAN.md's "Backend layout".
+Module-by-module intent is in `docs/PLAN.md`'s "Backend layout".
 
 - `server/` — the Rust backend (`ai-server`). `routes.rs`'s module doc
   comment is the HTTP table and the surface's source of truth.
@@ -40,16 +58,23 @@ Module-by-module intent is in PLAN.md's "Backend layout".
   projects version-locked to the commit this repo pins. What deliberately did
   **not** move is the API surface and the config *schema*: routes, drivers,
   sessions and setups are what makes this project itself.
-- `EXPLORER.md` — the file explorer's design (`server/src/files.rs` and
-  `FilesScreen.kt` / `FileViewer.kt` / `FileEditor.kt`).
-- `TRANSCRIPT_CACHE.md` — the phone's copy of what it has been sent. Read it
-  before touching `TranscriptCache.kt`, `TranscriptSource.kt`, or the opening
-  and stream effects in `SessionScreen.kt`.
-- `TODO.md` — the working list.
-- `RUST.md` — the plan for moving the app to Rust (on the `rustify`
+- `docs/` — every design and working document except this file and
+  `CLAUDE.md`:
+  - `docs/EXPLORER.md` — the file explorer's design (`server/src/files.rs`
+    and `FilesScreen.kt` / `FileViewer.kt` / `FileEditor.kt`).
+  - `docs/TRANSCRIPT_CACHE.md` — the phone's copy of what it has been sent.
+    Read it before touching `TranscriptCache.kt`, `TranscriptSource.kt`, or
+    the opening and stream effects in `SessionScreen.kt`.
+  - `docs/TODO.md` — the working list.
+  - `docs/RUST.md` — the plan for moving the app to Rust (on the `rustify`
     branch of the `ai-app-2` clone): what has to be reproduced, the
     framework decision, and the ordered experiments with their pass
     conditions. Read it before touching anything under that branch.
+  - `docs/IRIS.md`, `docs/IRIS_TODO.md`, `docs/DECISIONS.md`,
+    `docs/LAYOUT.md`, `docs/TEXTURES.md`, `docs/CLIENT_CORE.md` — iris's
+    own public API log, working list, decisions log, layout/render design,
+    and texture-atlas design, and the client-core crate's design,
+    respectively.
 - `.dev-updater.ron` — what Dev Updater builds here: the server (run as
   `service: Managed(…)`, supervised by Dev Updater's own implementation
   rather than a script kept here) and the APK, in parallel. It points at
@@ -86,7 +111,11 @@ two icon buttons the same width without either being given one — and why
   :androidApp:compileDebugKotlin :androidApp:lintDebug
   :androidApp:testDebugUnitTest`. The unit tests are JVM-only and cover the
   syntax highlighter, the ANSI parser and the transcript cache — the app's
-  pure logic with no Android in it.
+  pure logic with no Android in it. Touching anything under `BenchFixture.kt`,
+  `BenchNetwork.kt`, `BenchRun.kt` or the `bench` build type also needs
+  `:androidApp:compileBenchKotlin :androidApp:lintBench` — a second build
+  type compiles separately and lint has caught real bugs debug alone never
+  would (see "Android Lint" below).
 - **Android Lint is not optional and is not run by a build.** It found a
   crash that had been shipping (`java.time` on a minSdk-24 app with
   desugaring off) and later a permission check that silently dropped every
@@ -146,6 +175,27 @@ two icon buttons the same width without either being given one — and why
 
 Each exists because something was invisible without it.
 
+- **The `bench` build type and `app/bench-fixture/`** exist for P0 (RUST.md
+  and DECISIONS.md's 2026-09-05 entries), the phone benchmark gate Iris
+  asked for before porting continues: a deterministic, checked-in synthetic
+  transcript (`app/bench-fixture/generate.py`, never a real one) that both
+  this app and iris open with no server, so a frame-time comparison
+  measures the renderer rather than the data. `./build-apk.sh bench` builds
+  it — own application id (`com.example.aiapp.bench`) and label ("AI
+  Sessions bench") so it installs beside a real enrollment rather than
+  replacing it. Opening it goes straight to a session screen holding the
+  fixture (no enrollment, no permission prompts) with a "Run benchmark"
+  control beside "Copy" in session settings: it drives the same scroll loop
+  and streaming phase `transcript-bench.sh`/`stream-bench.sh` drive over
+  `ui-trace`, but in-process, since a real phone has no usable system
+  tracing and no agent can drive one (this-machine-android's skill).
+  `BenchFixture.kt`/`BenchNetwork.kt` fake the backend by installing a
+  `URLStreamHandlerFactory` that answers `TranscriptSource`/`EventStream`'s
+  requests from an in-memory copy of the fixture instead of opening a
+  socket — so the fold, the paging and `uniqueItems` under test are the
+  screen's real ones, never a shortcut built just for this. The report
+  gains a `bench:` section (process CPU time, peak RSS, battery current) on
+  every build, empty except when `BenchRun.kt` filled it in.
 - **`app/ui-sandbox.sh`** — a second `ai-server` with its own `$HOME`, config
   and data directory, holding eight invented Claude Code transcripts and a
   `claude` that is two lines of shell. **That isolation is the point**: the
@@ -226,6 +276,27 @@ Each exists because something was invisible without it.
   framework, from `atrace` text output with no trace processor needed. It is
   how the cost of a layout node per link was attributed to the framework
   rather than guessed at.
+- **`iris/android-app/build-apk.sh [debug|release] [--abi ...] [--features
+  ...]`** builds iris-android-app's cdylib (`cargo ndk`) and its APK
+  (Gradle) in one step and verifies the result (`aapt2`/`apksigner`), and
+  **`iris/android-app/run-bench.sh [--apk PATH]`** installs it on this
+  checkout's own emulator, taps "Run benchmark" by label, and prints the
+  report -- written so the P0 build/install/tap/read-report cycle stops
+  being retyped by hand each time (docs/RUST.md's P0 box).
+- **iris's three test layers** (docs/RUST.md's "Three test layers" has
+  the commands and what each cannot answer): test at the cheapest one
+  that can answer the question. `cargo test -p transcript-fixture` runs
+  the real transcript screen over the bench fixture with **no window, no
+  compositor and no GPU** (`iris::harness`), on a clock the test owns and
+  a gesture replayed from a `t_ms action x y` file under
+  `iris/transcript-fixture/touch/` -- which is how the batched 120Hz
+  flick a finger actually makes is testable at all, since a `ui-trace`
+  swipe is many evenly-spaced events. `iris/run-headless.sh phone --phone
+  --shot …` opens the same screen in a window at the phone's own size and
+  density for looking at, and `--replay FILE` drives the same recording
+  into it. The emulator is for JNI, the IME, insets, the surface
+  lifecycle and one verification run before a build goes to the phone --
+  not for iterating on layout.
 
 ### Driving the UI
 
@@ -312,7 +383,7 @@ means here:
 ## Sessions outlive the backend
 
 Since 2026-08-29 a session's process is deliberately left running when
-`ai-server` stops, and adopted again when it starts. PLAN.md has the design;
+`ai-server` stops, and adopted again when it starts. docs/PLAN.md has the design;
 day to day:
 
 - **Stopping the server no longer stops the sessions.** After `pkill
@@ -345,7 +416,7 @@ count tells you what continuing a session will cost. Shown, not warned about;
 importing a large session is a choice somebody is entitled to make.
 
 **Never import a Claude Code session that is open in a terminal.** The app
-refuses it — see PLAN.md for the incident that made that a refusal rather
+refuses it — see docs/PLAN.md for the incident that made that a refusal rather
 than a warning.
 
 **One Claude Code session id can name two files, and the listing offers it
@@ -521,4 +592,4 @@ belongs in `~/.claude/TOOLCHAIN.md` or `~/.claude/MACHINE.md` instead.
   `BasicTextField`, which costs two seconds a frame at 128 kB and stops the
   app at 1 MiB, so `EDIT_LIMIT` caps it at 32 kB with the reason said on
   screen. If you make the editor faster, that number is what to move.
-  EXPLORER.md's "What the measurements said" has the rest.
+  docs/EXPLORER.md's "What the measurements said" has the rest.

CLIENT_CORE.md

@@ -1,130 +0,0 @@
-# client-core
-
-`client-core/` is the app's pure logic held once instead of twice, per
-RUST.md's recommendation item 1. It is a plain Rust library crate with no UI
-framework dependency of any kind, so it can outlive whichever one the app
-ends up drawing with (Masonry, iris, or something else -- see RUST.md).
-`event-model/` is its sibling: the wire shape both this crate and `server/`
-share, extracted from `server/src/session/driver.rs` and
-`session/transcript.rs` on 2026-09-04.
-
-Neither crate is wired into anything yet. `server/` re-exports `event-model`
-so its own behaviour is unchanged (`./run-tests.sh` covers it); `client-core`
-has no caller -- it exists for whichever experiment in RUST.md picks it up
-next (a Masonry or iris transcript screen, most likely).
-
-## What's here, and what Kotlin file it replaces
-
-| `client-core/src/…`                     | Kotlin original                          | Status |
-|------------------------------------------|-------------------------------------------|--------|
-| `event-model/src/lib.rs` (shared crate)   | `Events.kt` (the enum mirror)             | Done   |
-| `ansi.rs`                                 | `Ansi.kt`                                 | Done, ported test-for-test |
-| `highlight/mod.rs`, `languages.rs`        | `Highlighter.kt`, `Languages.kt`          | Done, ported test-for-test |
-| `highlight/markdown.rs`                   | `MarkdownSyntax.kt`                       | Done, ported test-for-test |
-| `transcript_cache.rs`                     | `TranscriptCache.kt`                      | Done, ported test-for-test |
-| `sse.rs`                                  | `Sse.kt` (the framing half)               | Done, new tests (Kotlin had none of its own beyond integration) |
-| `api.rs`                                  | `Api.kt`                                  | Partial -- see below |
-| `event_stream.rs`                         | `EventStream.kt`                          | Done |
-| `transcript_fold.rs`                      | `TranscriptItems.kt`, `ToolRows.kt`       | Partial -- see below |
-| *(not started)*                           | `TranscriptSource.kt`                     | Not started |
-| *(not ported, and may never be)*          | `TranscriptUnits.kt`                      | Out of scope -- see below |
-
-Every file above whose Kotlin counterpart had a JVM unit test (`AnsiTest`,
-`HighlighterTest`, `TranscriptCacheTest`) has had every one of those test
-cases ported alongside it, plus new tests for the pieces that had none
-(`sse.rs`, `api.rs`, `event_stream.rs`, `transcript_fold.rs`). Test count by
-crate as of this writing: **85 in `client-core`**, 0 in `event-model` (its
-types carry no logic of their own to test -- `server/`'s own tests exercise
-them via `session::transcript`'s round-trip coverage).
-
-## Correspondence notes worth knowing before touching either side
-
-- **`ansi.rs`'s `StyledText`/`Style`/`Rgb`** stand in for Compose's
-  `AnnotatedString`/`SpanStyle`/`Color`, since this crate has no Compose.
-  `StyledText` is plain text plus a `Vec<(Range<usize>, Style)>` of
-  non-overlapping spans. Whatever UI framework ends up consuming this
-  crate maps `Style` onto its own text-styling type; nothing here should
-  change to accommodate a particular one.
-- **`highlight`'s `Span`/`Kind`** use **char indices, not byte offsets**
-  (`Vec<char>` internally), mirroring the Kotlin original's `Char`-indexed
-  strings. `highlight::span_text` turns a `Span` back into text for a
-  caller working the same way; a caller that wants byte offsets into a
-  `&str` has to convert.
-- **`transcript_cache.rs`'s `SessionCache::guard`** found a real
-  translation bug while it was being written: an early draft let a
-  *damaged* chunk (one file unreadable, discard just this session) and a
-  genuine I/O failure (disk gone, disable the whole cache) both surface as
-  the same `Err` from one closure, which would have disabled every
-  session's cache over a single corrupt chunk. Fixed by checking a
-  thread-local "was this damage" flag before deciding which failure mode
-  it was -- see the comment on `guard` and the commit message for
-  `transcript_cache.rs`.
-
-## What `api.rs` covers, and what it does not yet
-
-`ApiClient` wraps a `Transport` trait (network I/O kept out from behind, so
-`ApiClient` and `event_stream::follow_session_events` are tested with a
-fake transport and no server). `UreqTransport` is the only real
-implementation, backed by `ureq` -- see its Cargo.toml comment for why
-(blocking, already a project dependency, no extra TLS crate needed since
-`ureq::tls::Certificate::from_pem` reads the pinned CA directly).
-
-Covered: session list/read, message send, unqueue, answer, interrupt,
-stop, start, rename, cwd, model, permission-mode, notify, command,
-compact, delete, and one transcript page.
-
-**Not covered, and each is real work rather than a stub to fill in:**
-setups (`/setups*`, machine and provider discovery), the file explorer
-(`/setups/{id}/dir|file`), usage (`/usage`), models
-(`/models*`, HuggingFace browsing and downloads), attachments
-(`/sessions/{id}/attachments`), importing (`/setups/{id}/importable*`),
-and the `/notifications` stream. `server/src/routes.rs`'s module doc is
-the full table to work from when one of these is next.
-
-## What `transcript_fold.rs` covers, and what it does not yet
-
-`fold_event` covers every `Event` variant server/ can produce today,
-including tool-call/question/image attachment and peer-message placement.
-`group_tool_runs` groups adjacent calls into `TranscriptRow::Tools`.
-
-**Not ported:** `TranscriptItems.kt`'s `joinPages` (and its
-`healSplitMessage`/`adoptRun` helpers) -- the page-boundary healing that
-merges a tool call split across two fetched pages and re-merges a run a
-boundary cut through. This matters the moment paging backward through
-history is exercised; it is deliberately left rather than rushed, since
-it is exactly the kind of boundary logic this project's own "things that
-have bitten" section warns reads fine and is wrong at the edges.
-
-**Known gap, and a decision for whoever closes it:** `event_model::Event`
-has no `Unknown`/catch-all variant, unlike `Events.kt`'s hand-kept mirror.
-A server newer than this build that adds an event type will fail to parse
-that line rather than degrading to a placeholder row. Closing this means
-deciding how `event_model` itself represents "a shape I don't recognise"
--- a shared-model decision affecting `server/` too, not a `client-core`-only
-fix, so it is recorded here rather than silently worked around.
-
-## What is not started at all
-
-- **`TranscriptSource.kt`** -- the layer that decides whether a page comes
-  from the transcript cache or the server, and stitches the two. Needs
-  `transcript_cache.rs` and `api.rs`'s transcript-page method, both of
-  which exist now, so this is unblocked whenever picked up.
-- **The markdown *block* model beyond syntax spans** -- `highlight/markdown.rs`
-  colours a `.md` file or fence for the highlighter, but does not build the
-  block tree (headings, lists, tables, fences as distinct nodes) that a
-  renderer walks to lay out prose versus code versus a table.
-  `CodeFence.kt`'s use of `org.intellij.markdown` for that full CommonMark
-  AST is Compose rendering plumbing, not something to port as-is; a Rust
-  UI layer will want its own block parser or a crate for it, decided
-  alongside the framework choice in RUST.md.
-- **`TranscriptUnits.kt`** (see above) -- deliberately out of scope, since
-  it flattens a row into bounded units for a *specific* lazy-list
-  framework's composition cost, which is a fact about that framework
-  rather than about the transcript.
-
-## Verifying
-
-`./run-tests.sh` from the repo root now runs `event-model`, `client-core`
-and `server` in that order (each `cargo test`, forwarding arguments the
-same way it always has). From `client-core/` directly: `cargo test`,
-`cargo clippy --all-targets`, `cargo fmt` -- all clean as of this writing.

DECISIONS.md

@@ -0,0 +1,44 @@
+# Decisions awaiting review
+
+Choices made while working autonomously, for Bryan to keep or change. Each
+says what was picked and why; the detail is in the design doc it names.
+Delete an entry once it has been looked at.
+
+## Subagent views (2026-09-05, `SUBAGENTS.md`)
+
+Made on my own judgement, limited blast radius:
+
+1. **A subagent is a transcript, not a session.** It has no process,
+   controls or settings; it is addressed as `/sessions/{id}/subagents/{sub}`
+   and stored under the session's directory, so deleting the session takes
+   it. Alternative rejected: registering it as a session of its own, which
+   would give it a card in the main list and a driver that can do nothing.
+2. **Read-only view is the session screen minus its controls**, rather than
+   a second, simpler transcript screen. Keeps paging, caching, selection
+   and rendering in one place. Cost: a `readOnly` mode threaded through
+   `SessionScreen`.
+3. **The list only carries a count.** Each session row says how many
+   subagents it has; their titles and statuses are fetched when the card is
+   expanded. Keeps `GET /sessions` from reading every subagent transcript.
+   Consequence: an expanded card's statuses refresh with the list, not live.
+4. **Expanded/collapsed is remembered per session on the phone**, not on
+   the server. Collapsed by default, per the transcript convention that new
+   things arrive collapsed.
+5. **Subagents of imported sessions are not shown.** The import path still
+   skips `isSidechain` records; the CLI's own `subagents/agent-*.jsonl` files
+   are not read. Only subagents run while this backend was watching exist.
+6. **Echo grows `/subagent [n]`** as the test rig, so nothing here needs a
+   paid turn to exercise.
+
+Deferred, because they reach further than this feature:
+
+- **Live status on the list.** Whether the session list should follow a
+  stream at all (it refreshes on demand today) decides whether subagent
+  status can ever be live there. Not changed.
+- **Nested subagents.** A subagent's own Task calls are shown as tool calls
+  in its transcript and are not given transcripts of their own. Supporting
+  that is the same mechanism one level down, but the UI would need nested
+  expanders.
+
+- **The subagent status row says "context unknown".** Nothing measures a
+  subagent's context; the row could leave it out rather than admit it.

IRIS.md

@@ -1,110 +0,0 @@
-# iris: notable public API changes
-
-For Iris to read on her own time. Each entry is a change to iris's public
-surface that a widget author or app author would notice: a trait method
-added, removed or re-shaped; a type that callers construct differently; a
-capability that moved. Small and trivial changes do not go here.
-
-An entry gives the date, what changed, why, and a short before/after where
-it helps judge the change without the session that made it. Newest first.
-
-## 2026-09-05: a second backend (android-view), and what moved to make room for it
-
-RUST.md's I2. Three changes a widget or app author would notice, all in
-service of the same thing: `default` (winit) and the new `android`
-(android-view) backends sharing what does not depend on windowing.
-
-- **`Selector`/`Selectable`'s bound changed from `Rsc::State:
-  HasDefaultUiState` to `Rsc::State: FocusHost`** (new trait, `attr.rs`).
-  `HasDefaultUiState` still exists and still works — `default/attr.rs` now
-  implements `FocusHost` for anything that has it — so a winit app's
-  existing code is unaffected. An Android app implements `FocusHost` via
-  `HasAndroidUiState` instead. Affects only an app that referenced
-  `HasDefaultUiState` directly at a `Selectable`/`Selector` call site
-  rather than through `.attr::<Selectable>(())`, which nothing in-tree
-  does.
-- **`Tasks::init` takes `Arc<dyn RequestRedraw>` instead of
-  `Arc<winit::window::Window>`.** `RequestRedraw` (`task.rs`) is one method,
-  `fn request_redraw(&self)`; `winit::window::Window` implements it
-  (`default/render.rs`), so `Tasks::init(window)` at a call site is
-  unchanged by inference. Only matters if something constructed a `Tasks`
-  directly rather than through `DefaultRsc`/`AndroidRsc`.
-- **`TextEdit::apply_event`/`TextInputResult` are `#[cfg(not(target_os =
-  "android"))]`** — they take a `winit::event::KeyEvent`, which does not
-  exist on Android; `android/input.rs` drives the same primitives
-  (`backspace`/`delete`/`motion`/`insert`, all still unconditional) from
-  `ndk::event::Keycode` directly instead. New unconditional getters on the
-  way: `TextEdit::text()`/`selection_range()`/`caret()`, and
-  `TextEditCtx::delete_byte_range`/`set_cursor_byte` — the primitives
-  `android/ime.rs`'s `InputConnection` bridge needed and that were not
-  previously exposed publicly.
-
-## 2026-09-04: `Widget::draw` reports the size it used; `desired_width`/`desired_height` are gone
-
-A widget used to implement three methods (`draw`, `desired_width`,
-`desired_height`); it now implements one, `fn draw(&mut self, painter: &mut
-Painter) -> Size`, which draws into `painter.region()` and returns how much
-of it was used. Why: the two extra methods routinely re-simulated what
-`draw` was about to do anyway (`Span::desired_ortho` copied its own draw
-loop to get cross-axis sizing right) — one visit per widget per frame
-instead of up to three. A container that needs a child's size before
-placing it (alignment, centering) draws the child once at a provisional
-region, reads the returned `Size`, and calls the new `Painter::reposition`
-to move it into its final spot — an O(1) offset write, not a second draw. A
-widget whose drawn output never depends on the size it's given (a
-fixed-size `Rect`, a decoded `Image`) overrides the new `fn
-is_size_independent(&self) -> bool { false }` to `true`, which skips
-redrawing it when only its offered region changes shape.
-
-```rust
-// before
-fn draw(&mut self, painter: &mut Painter) { /* ... */ }
-fn desired_width(&mut self, ctx: &mut SizeCtx) -> Len { /* ... */ }
-fn desired_height(&mut self, ctx: &mut SizeCtx) -> Len { /* ... */ }
-
-// after
-fn draw(&mut self, painter: &mut Painter) -> Size { /* ... */ }
-```
-
-`SizeCtx` and `Cache` are gone with it — see `LAYOUT.md` for the full
-design, the move-offset mechanism this shipped alongside, and the file
-list.
-
-## 2026-09-04: texture pipeline rebuilt off the binding array
-
-`Textures`/`TextureHandle`, `GlyphPrimitive`, and `UiRenderNode::new` all
-changed shape. Why: the old pipeline bound every texture ever drawn in one
-`binding_array<texture_2d<f32>>` and asked every device, unconditionally,
-for `VK_EXT_descriptor_indexing` — a real share of Android GPUs lack it,
-and it failed outright on the Android emulator's software Vulkan. See
-TEXTURES.md's "Recommended shape" and "Implemented, 2026-09-04".
-
-- **`UiRenderNode::new` drops its `limits: UiLimits` parameter, and
-  `UiLimits` is gone.** Before: `UiRenderNode::new(&device, &queue,
-  &config, UiLimits::default())`. After: `UiRenderNode::new(&device,
-  &queue, &config)`. Nothing replaces it — there are no more
-  binding-array limits to size.
-- **`src/default/render.rs`'s device request asks for no features and no
-  binding-array limits.** Before: `required_features:
-  Features::TEXTURE_BINDING_ARRAY | Features::PARTIALLY_BOUND_BINDING_ARRAY
-  | Features::SAMPLED_TEXTURE_AND_STORAGE_BUFFER_ARRAY_NON_UNIFORM_INDEXING`
-  plus two `max_binding_array_*` limits. After: `Features::empty()` (the
-  `DeviceDescriptor` default) and only `max_buffer_size` set, which was
-  never about the binding array.
-- **`TextureHandle` has no `primitive()` method any more**; a caller
-  outside `iris` shouldn't have been calling it (it fed the old renderer's
-  internals), but if something did: use `image_index()` for a standalone
-  image's bind-group index. There is no equivalent for a page — a page has
-  no bind group of its own now, see below.
-- **`GlyphPrimitive` has no public constructor from a struct literal.**
-  Before: `GlyphPrimitive { uv_min, uv_max, view_idx, sampler_idx, color,
-  flags }`. After: `GlyphPrimitive::new(uv_min, uv_max, layer, color,
-  flags)` — one `layer` (the shared atlas array's layer) instead of a
-  `view_idx`/`sampler_idx` pair, since a page is now a layer of one array
-  texture rather than its own bound texture.
-- **A widget author drawing images is unaffected**: `Painter::texture`/
-  `texture_at`/`texture_within` and `Textures::add` keep their signatures.
-  What changed underneath is that each standalone image now gets its own
-  `wgpu::BindGroup` and draw call instead of a slot in the shared array —
-  invisible from the widget API, visible only in `UiRenderNode`'s internals
-  and in `iris`'s device requirements.

IRIS_TODO.md

@@ -1,66 +0,0 @@
-# iris: known problems and things still to build
-
-Iris's own list for the library, recorded 2026-09-04 in her words where it
-matters, so the agents working through RUST.md pick these up in a sensible
-order rather than rediscovering them. Each item says where it sits in the
-order and what "done" looks like. Tick and date them in place.
-
-## Fix
-
-- [x] **Input does not fall through by input type (2026-09-04).**
-  `SensorUi::run_sensors` (`src/default/sense.rs`) used to set "consumed,
-  stop checking lower layers" from mere hover — a widget registered for
-  nothing but `click()` blocked a `Scroll` meant for whatever was behind
-  it, since "the cursor is over this widget" and "this widget handled the
-  event" were the same check. Fixed by judging consumption per input
-  kind: with no button transition and no scroll happening this frame
-  ("momentary" activity), the topmost hovered widget still wins, same as
-  before; when something momentary *is* happening, only a widget whose
-  registered senses actually include a matching non-hover one (checked
-  via a new `TypeEventManager::registered`, which lists what a widget
-  registered without running anything) consumes it, so a widget with only
-  `Hovering`/click handlers can no longer block a scroll from reaching a
-  list underneath. `iris/src/sense_tests.rs` builds a button-over-a-list
-  `Stack` with a plain `HasEvents` impl (no GPU or window) and checks both
-  directions: a scroll over the button reaches the list, and a real click
-  still reaches the button — confirmed to fail on the pre-fix code and
-  pass after.
-
-## Build
-
-- [ ] **Benchmarks**, not unit tests, run on demand (a `benches/` or a
-  script under `iris/`, never in `cargo test`). The scenario that matters
-  most is a **message list** — chat apps and this app's transcript alike —
-  stressed with many messages and many images. One case in particular:
-  **resizing an input box** (typing enough text to grow it) that pushes a
-  long list of messages above it must stay very fast and recalculate
-  almost nothing — a move of everything above, not a re-layout. That is
-  exactly the O(1) move chain in LAYOUT.md; the benchmark is what proves
-  it. Done when the numbers are in this file with the command, and the
-  input-box case reports draws re-run, not just frame time.
-- [ ] **Masks defined relative to each other.** Wanted: mask A multiplies
-  by something *and also* applies mask B — a mask can reference a parent
-  mask, the way the move chain references a parent offset. Today masks
-  are independent regions. Design it beside the move chain (same shape:
-  a parent index and a bounded walk in the shader); do it when a real
-  widget needs it, not before.
-- [ ] **Positions as a single float per scroll.** Iris raised, and half
-  rejected, letting a scroll update one float rather than positions:
-  input handling cares about most elements in a list, so absolute
-  positions must be computed on the CPU anyway. LAYOUT.md's design
-  already lands here (GPU walks the chain, CPU resolves on demand for
-  hit tests). Keep the CPU resolution lazy and per query; do not
-  materialise every row's absolute position per frame.
-- [ ] **Animations, last.** Cosmetic, so after everything above. Must be
-  **modular — a piece of the library rather than a core part forced into
-  everything, the same way input is**. Whatever the mechanism, a widget
-  that does not animate must pay nothing and import nothing for it.
-
-## Reconsider
-
-- [ ] **`WidgetView`.** Iris is unsure of it: what she wants is an easy way
-  to compose a widget from others (a button is the main case). With
-  sizing folded into `draw`, composing may be easy enough that `View` is
-  redundant. Decide after the layout change lands, by writing a button
-  both ways and keeping the one that is shorter to explain; delete the
-  other rather than keeping two ways.

SUBAGENTS.md

@@ -0,0 +1,141 @@
+# Subagents
+
+A session's subagents -- the helpers a Claude Code session starts through its
+Task tool -- each get a transcript of their own, listed under the session's
+card and readable in the same transcript view the session has. Designed
+2026-09-05; the decisions Bryan has not yet reviewed are in `DECISIONS.md`.
+
+## What a subagent is here
+
+**A subagent is a second transcript owned by a session, in the same event
+model, with no process and no controls.** It is not a session: it cannot be
+messaged, stopped or started, and it has no setup, model or usage of its
+own. Everything it shares with a session -- the transcript file format, the
+paging routes, the SSE stream, the phone's cache and rendering -- is reused
+by addressing, not by copying.
+
+The CLI reports a subagent's messages on the parent's own stream-json
+output, each carrying `parent_tool_use_id` = the id of the Task `tool_use`
+that started it. Before this the translator dropped those lines
+(`subagent_events_are_not_duplicated_into_the_transcript`); now it routes
+them to that subagent's own translator and transcript. The parent's
+transcript still shows only the Task call itself.
+
+## Storage
+
+Under the session directory:
+
+```
+<session>/subagents/<tool_use_id>/meta.json        {title, created}
+<session>/subagents/<tool_use_id>/transcript.jsonl same SeqEvent lines as the session's
+```
+
+The id is the Task tool_use id (`toolu_…`), which is unique, stable across a
+backend restart, and already the key everything on the parent side uses.
+Only ids matching `[A-Za-z0-9_-]+` are ever created or looked up, since the
+id becomes a path.
+
+The transcript's sequence numbers are its own, starting at 1. `Transcript`,
+`read_window`, `catch_up` and `read_after` work on it unchanged.
+
+Its path out: deleting the session deletes its directory, subagents included.
+There is no separate delete.
+
+## Lifecycle, as events in the subagent's transcript
+
+1. Created on the first child line for an unseen parent id (or, when the
+   parent Task call was seen, at that call). First lines written:
+   `Status Running`, then `UserMessage { text: <the Task's prompt> }` when
+   the prompt is known -- it genuinely is the subagent's first user turn.
+2. Every child line is translated by that subagent's own `Translator`
+   (one per subagent: tool ids are unique but streaming deltas are by
+   content-block index, and parallel subagents interleave).
+3. **The parent's `tool_result` never finishes a subagent.** The Task tool
+   runs in the background by default: the `tool_result` -- "Async agent
+   launched..." -- arrives the moment it *starts*, while the subagent goes
+   on working for however long its own turn takes, sometimes minutes. What
+   ends it is its own turn ending: the raw API's `message_delta` on its
+   stream carrying `stop_reason: "end_turn"` (a `stop_reason` of `tool_use`
+   is the model about to call one, not an end), or a `result` line for its
+   own turn if a future CLI version ever sends one. Either maps to
+   `Status Exited`; the subagent's vocabulary has no `Idle`, so the
+   equivalent event `dispatch` produces for an ordinary session is dropped
+   rather than written. A shipped version of this finished on the
+   `tool_result` instead, which read a running background agent as
+   "finished" with its transcript truncated at the moment it launched.
+4. **A child line for a subagent that already finished reopens it**
+   (`Status Running`) rather than being dropped: a background Task can be
+   sent another message long after its first turn ended, and that is
+   exactly what a further line for it means. Same transcript, same child
+   `Translator`, just picking back up.
+5. When the parent session's process exits (`Status Exited` on the
+   session), every subagent still `Running` gets `Status Exited` too: its
+   process was the parent's.
+
+A subagent that was mid-flight when the backend restarted keeps working:
+the registry reopens the existing transcript on the next child line, and
+the file continues its sequence -- the same reopening #4 describes, whether
+what closed it was a restart or its own `end_turn`. If its turn ended while
+the backend was down nothing recorded that until the next line arrives, so
+its last status stays `Running`, which the list reports as **unknown**
+rather than as running (see the wire shape) until then.
+
+Title: the Task call's `description` input, then ` (<subagent_type>)` when
+one is given; falling back to the tool's name when the child arrives before
+(or without) the parent call being seen.
+
+## Server layout
+
+- `session/subagent.rs` -- the registry: `Subagents` (per session, in
+  `Shared`), `Subagent` (its `Transcript` behind a mutex plus a
+  `broadcast::Sender<SeqEvent>`), `record(id, event)`, `start(id, title,
+  prompt)`, `finish(id)`, `reopen(id)`, `finish_all()`, `list()` from disk. Drivers get an
+  `Arc<Subagents>` beside their `EventSink`; llama ignores it.
+- `session/claude/translate.rs` -- routes child lines by parent id, holds
+  one child `Translator` per subagent, remembers pending Task calls'
+  description/prompt/subagent_type.
+- `session/echo.rs` -- `/subagent [n]`: the test rig. Starts *n* (default 1)
+  subagents at once, each named "helper k". Each writes the prompt as its
+  user message, streams a few words of text, runs one `Bash` tool call, then
+  finishes about three seconds after starting, and the parent's Task calls
+  end when their subagent does. Three seconds so the running state can be
+  seen on the phone.
+- `routes.rs` -- three routes, in the doc table.
+
+## Wire shape
+
+```
+GET /sessions/{id}                       SessionInfo gains `subagents: N` (count, 0 when none)
+GET /sessions                            same field on each row
+GET /sessions/{id}/subagents             [{id, title, status, created, lastActivity}], oldest first
+GET /sessions/{id}/subagents/{sub}/transcript   exactly the session transcript's query and answer
+GET /sessions/{id}/subagents/{sub}/events?after=N  exactly the session events stream
+```
+
+`status` is the transcript's last `Status` event, serialised like a session's
+(`running`, `exited`), except that a subagent whose session is not itself
+running cannot be running: the list answers `unknown` for that one. The
+phone words these as *running*, *finished* and *unknown* on the subcard.
+
+The count on `SessionInfo` is a directory listing, so the list stays cheap.
+The per-subagent status is only read when the list route is asked for.
+
+## Phone
+
+- `SessionSummary.subagents: Int`. A card with a non-zero count ends in an
+  expander row -- a full-width `Chevron(Pointing.Down)` row that flips to
+  `Pointing.Up` -- collapsed by default. Expanding fetches
+  `/sessions/{id}/subagents` and draws one `OutlinedCard` per subagent,
+  indented inside the session card, the way dev-updater draws a project's
+  components: title, then the status word and a relative time. The
+  expansion state is per session id and survives a refresh of the list.
+- Tapping a subcard opens `Screen.Subagent`, which is `SessionScreen` in
+  **read-only** form: the same transcript, paging, cache, selection,
+  images and status row, with the composer, the process button, the model
+  picker, the files button, the settings cog and the usage bar left out.
+  The header shows the subagent's title with the session's title beneath
+  it. Back returns to the list.
+- Addressing: `fetchTranscript`, `EventStream`, `TranscriptSource` and the
+  cache take a transcript address rather than a session id --
+  `sessions/{id}` or `sessions/{id}/subagents/{sub}` -- so the cache nests a
+  subagent's copy under its session's and the same code serves both.

android-shell/Cargo.toml

@@ -0,0 +1,36 @@
+[package]
+name = "android-shell"
+version = "0.1.0"
+edition = "2024"
+
+# The JNI bridge behind E3's two Java stub classes (`MainActivity`,
+# `NotificationService` -- see RUST.md's "How much Java is unavoidable" for
+# why those two classes cannot be anything but Java/Kotlin, registered from
+# the manifest by name). Everything they would otherwise have done in
+# Kotlin -- the SSE follow loop, deciding where a notification is shown,
+# picking a session for a share -- is here instead, built on `client-core`
+# so the networking and parsing are not duplicated a third time next to the
+# server and the Kotlin app.
+#
+# `cdylib` for `System.loadLibrary`; `lib` too so `cargo test`/`clippy` run
+# on a normal host target without an Android NDK toolchain, the same
+# posture `client-core` and `server` already have.
+
+[lib]
+name = "android_shell"
+crate-type = ["cdylib", "lib"]
+
+[dependencies]
+client-core = { path = "../client-core" }
+jni = "0.22"
+log = "0.4"
+
+# `LogErrorAndDefault` (the `native_method!` error policy this crate uses
+# throughout, see lib.rs) logs through the `log` facade, which is a no-op
+# without a backend installed -- so without this, every recoverable error
+# at a native entry point would be silently dropped rather than reaching
+# logcat. Android-only: nothing else here needs it, and it does not build
+# off-device (see `notify::ensure_logger`'s call site, the only place this
+# is used).
+[target.'cfg(target_os = "android")'.dependencies]
+android_logger = "0.15"

android-shell/src/jcall.rs

@@ -0,0 +1,152 @@
+//! Thin wrappers around the five `Env` calls this crate makes constantly
+//! (a class name, a method name and a signature, all as plain `&str`).
+//!
+//! `jni` 0.22 wants a class or method *name* as `AsRef<JNIStr>` (its own
+//! modified-UTF-8 type; `JNIString::new` is the runtime conversion, used
+//! here uniformly rather than switching to the compile-time `jni_str!`
+//! literal macro call by call -- these are a handful of short, one-off
+//! lookups, not a hot loop, so the difference is not worth two code paths
+//! for the same thing) and a *signature* as a parsed `MethodSignature`/
+//! `FieldSignature`, which is why those go through
+//! `RuntimeMethodSignature`/`RuntimeFieldSignature::from_str` instead: the
+//! parsed form is what lets these calls skip re-validating the signature
+//! against the arguments on every call, which is the whole reason `jni`
+//! moved to it.
+//!
+//! **The classloader gotcha, found by testing (2026-09-05).** A class
+//! lookup by name (`find_class`, `new_object`, `call_static_method`,
+//! `get_static_field` -- anything that resolves a *class*, as opposed to
+//! `call_method` on an object it already has, which needs no such lookup)
+//! defaults to `FindClass`'s ordinary search when it cannot find the
+//! calling thread a classloader through `Thread.getContextClassLoader()`.
+//! That default is fine on a thread the JVM itself started -- an
+//! `onCreate`/`onStartCommand` callback -- but every one of these calls
+//! from `android-shell`'s own background thread (the notification
+//! follow-loop, the share upload) is running on a thread *Rust* spawned
+//! and attached with `JavaVM::attach_current_thread`, which the platform
+//! never gave an app classloader. Framework classes
+//! (`android.app.Notification$Builder`, ...) still resolve, because they
+//! are reachable from the bootstrap loader `FindClass` falls back to --
+//! `androidx.core.app.NotificationManagerCompat` is not, since it is
+//! packaged inside this app's own APK. The failure was
+//! `Error::NoClassDefFound`, logged by `notify::show`'s `LogErrorAndDefault`
+//! as "failed to resolve Java class ... (class not found or linkage
+//! error)" -- on a real device this reads as "the notification silently
+//! never arrives," since the whole call is inside the follow loop and the
+//! ongoing foreground notification (built on the main thread, in
+//! `try_start`, before the background thread exists) posts fine either
+//! way. `remember_class_loader` caches the app's own `ClassLoader` the
+//! first time any entry point has a `Context` to ask, and every class
+//! lookup below goes through it explicitly via `LoaderContext::Loader`
+//! rather than the thread-dependent default -- so it is correct on the
+//! main thread and on this crate's own background threads alike.
+
+use jni::Env;
+use jni::errors::Result;
+use jni::objects::{JClass, JClassLoader, JObject, JValue, JValueOwned};
+use jni::refs::{Global, LoaderContext};
+use jni::signature::{RuntimeFieldSignature, RuntimeMethodSignature};
+use jni::strings::JNIString;
+use std::sync::OnceLock;
+
+static CLASS_LOADER: OnceLock<Global<JClassLoader<'static>>> = OnceLock::new();
+
+/// Caches `context`'s own `ClassLoader`, the first time this is called.
+/// Cheap to call from every entry point that has a `Context` on hand
+/// (`MainActivity`'s and `NotificationService`'s all do): later calls are
+/// a `OnceLock::get` and nothing else.
+pub fn remember_class_loader(env: &mut Env, context: &JObject) -> Result<()> {
+    if CLASS_LOADER.get().is_some() {
+        return Ok(());
+    }
+    // context.getClass().getClassLoader() -- resolved via `call_method` on
+    // real objects throughout, so this needs no class-name lookup of its
+    // own and has nothing to bootstrap.
+    let class_obj = call_method(env, context, "getClass", "()Ljava/lang/Class;", &[])?.l()?;
+    let loader_obj = call_method(
+        env,
+        &class_obj,
+        "getClassLoader",
+        "()Ljava/lang/ClassLoader;",
+        &[],
+    )?
+    .l()?;
+    let loader = env.cast_local::<JClassLoader>(loader_obj)?;
+    let global = env.new_global_ref(&loader)?;
+    // Lost the race with another entry point calling this concurrently --
+    // both loaders name the same app, so either one is fine and there is
+    // nothing to reconcile.
+    let _ = CLASS_LOADER.set(global);
+    Ok(())
+}
+
+/// Resolves `name` (slash-separated, e.g. `androidx/core/app/NotificationCompat`)
+/// through the cached app classloader when one has been remembered, and
+/// through the ordinary default otherwise -- which is every call made
+/// before any entry point has run, and is also correct for a main-thread
+/// caller, so there is no case this makes worse.
+fn resolve_class<'local>(env: &mut Env<'local>, name: &str) -> Result<JClass<'local>> {
+    match CLASS_LOADER.get() {
+        Some(loader) => {
+            let binary_name = name.replace('/', ".");
+            LoaderContext::Loader(loader).load_class(env, JNIString::new(&binary_name), true)
+        }
+        None => env.find_class(JNIString::new(name)),
+    }
+}
+
+pub fn find_class<'local>(env: &mut Env<'local>, name: &str) -> Result<JClass<'local>> {
+    resolve_class(env, name)
+}
+
+/// A new Java string as a plain `JObject` -- what every call site here
+/// wants it as (`JValue::Object` takes `&JObject`, not `&JString`, and
+/// `JString: Into<JObject>` is the documented way across).
+pub fn jstr_obj<'local>(env: &mut Env<'local>, text: impl AsRef<str>) -> Result<JObject<'local>> {
+    Ok(env.new_string(text)?.into())
+}
+
+pub fn new_object<'local>(
+    env: &mut Env<'local>,
+    class: &str,
+    sig: &str,
+    args: &[JValue],
+) -> Result<JObject<'local>> {
+    let sig = RuntimeMethodSignature::from_str(sig)?;
+    let class = resolve_class(env, class)?;
+    env.new_object(class, sig.method_signature(), args)
+}
+
+pub fn call_method<'local>(
+    env: &mut Env<'local>,
+    obj: &JObject,
+    method: &str,
+    sig: &str,
+    args: &[JValue],
+) -> Result<JValueOwned<'local>> {
+    let sig = RuntimeMethodSignature::from_str(sig)?;
+    env.call_method(obj, JNIString::new(method), sig.method_signature(), args)
+}
+
+pub fn call_static_method<'local>(
+    env: &mut Env<'local>,
+    class: &str,
+    method: &str,
+    sig: &str,
+    args: &[JValue],
+) -> Result<JValueOwned<'local>> {
+    let sig = RuntimeMethodSignature::from_str(sig)?;
+    let class = resolve_class(env, class)?;
+    env.call_static_method(class, JNIString::new(method), sig.method_signature(), args)
+}
+
+pub fn get_static_field<'local>(
+    env: &mut Env<'local>,
+    class: &str,
+    field: &str,
+    sig: &str,
+) -> Result<JValueOwned<'local>> {
+    let sig = RuntimeFieldSignature::from_str(sig)?;
+    let class = resolve_class(env, class)?;
+    env.get_static_field(class, JNIString::new(field), sig.field_signature())
+}

android-shell/src/lib.rs

@@ -0,0 +1,131 @@
+//! The JNI bridge behind E3's two Java stub classes. See `Cargo.toml`'s
+//! package comment for what this crate is and RUST.md's E3 entry for the
+//! design decisions.
+//!
+//! Each native method is declared with `jni`'s [`native_method!`] macro
+//! rather than a hand-written `#[no_mangle] extern "system" fn Java_...`:
+//! the macro derives the mangled export name and the JNI signature from the
+//! Rust function itself, so the two cannot drift apart the way a
+//! hand-typed name string and a hand-typed `"(Landroid/...;)V"` signature
+//! routinely do. `error_policy = LogErrorAndDefault` matches
+//! `Notifications.kt`'s own posture: a failure here (a lost connection, a
+//! JNI call that threw) is reported to logcat, not thrown back into Java
+//! as an exception that would crash the app over something recoverable.
+//!
+//! Each `const _: NativeMethod = native_method! { ... };` binding is
+//! otherwise unused by name -- `_` is the idiomatic way to keep a
+//! side-effecting const (here, generating the `#[export_name]`d function
+//! the JVM resolves by the JNI naming convention) without a `dead_code`
+//! warning for a binding nothing reads.
+
+mod jcall;
+mod notify;
+mod settings;
+mod share;
+
+use jni::errors::LogErrorAndDefault;
+use jni::objects::{JClass, JObject};
+use jni::sys::jint;
+use jni::{Env, NativeMethod, native_method};
+
+/// Installs the `log` backend that routes to logcat, once per process.
+/// Without it, `LogErrorAndDefault` (every native method below) and any
+/// `log::error!` inside `jni` itself (e.g. `JString`'s `Display` fallback)
+/// call into the `log` facade's default no-op logger, and a real failure
+/// vanishes with nothing on logcat to say so -- silently *more* wrong than
+/// crashing, since nothing on screen or in the log says a notification was
+/// dropped. Called from every entry point below rather than a Java-side
+/// `Application.onCreate`, since this crate deliberately has no such class
+/// to hook (see RUST.md's E3 entry on the two-Java-classes floor).
+fn ensure_logger() {
+    static ONCE: std::sync::Once = std::sync::Once::new();
+    ONCE.call_once(|| {
+        #[cfg(target_os = "android")]
+        android_logger::init_once(
+            android_logger::Config::default()
+                .with_max_level(log::LevelFilter::Debug)
+                .with_tag("android-shell"),
+        );
+    });
+}
+
+// The parameters are spelled as their Java types, not as `JObject`: the
+// macro encodes each argument into the exported symbol's JNI signature
+// (and JNI resolves `Java_...` names *by* that signature), so a generic
+// `JObject` here would export `(Ljava/lang/Object;...)` against a Java
+// method actually declared `(Landroid/app/Activity;...)` -- two different
+// symbols that never resolve to each other, silently, with no compiler
+// error on either side. `android.app.Activity` etc. have no dedicated
+// Rust wrapper in this crate, so they fall back to plain `JObject` in the
+// implementation functions below (the "Built-in Types" note in
+// `native_method!`'s docs).
+const _: NativeMethod = native_method! {
+    java_type = "com.example.aiapp.shell.MainActivity",
+    static extern fn native_handle_intent(activity: android.app.Activity, intent: android.content.Intent) -> (),
+    error_policy = LogErrorAndDefault,
+};
+
+/// `MainActivity.nativeHandleIntent` -- called from `onCreate` and
+/// `onNewIntent`. See `share::handle_intent` for what an intent can mean.
+fn native_handle_intent<'local>(
+    env: &mut Env<'local>,
+    _class: JClass<'local>,
+    activity: JObject<'local>,
+    intent: JObject<'local>,
+) -> Result<(), jni::errors::Error> {
+    ensure_logger();
+    jcall::remember_class_loader(env, &activity)?;
+    share::handle_intent(env, &activity, &intent)
+}
+
+const _: NativeMethod = native_method! {
+    java_type = "com.example.aiapp.shell.NotificationService",
+    static extern fn native_sync(context: android.content.Context) -> (),
+    error_policy = LogErrorAndDefault,
+};
+
+/// `NotificationService.nativeSync` -- called both from `MainActivity` (an
+/// enrollment may have just landed) and from `NotificationService.sync`
+/// itself. See `notify::sync`.
+fn native_sync<'local>(
+    env: &mut Env<'local>,
+    _class: JClass<'local>,
+    context: JObject<'local>,
+) -> Result<(), jni::errors::Error> {
+    ensure_logger();
+    jcall::remember_class_loader(env, &context)?;
+    notify::sync(env, &context)
+}
+
+const _: NativeMethod = native_method! {
+    java_type = "com.example.aiapp.shell.NotificationService",
+    static extern fn native_on_start_command(service: android.app.Service) -> jint,
+    error_policy = LogErrorAndDefault,
+};
+
+/// `NotificationService.nativeOnStartCommand`. See `notify::on_start_command`.
+fn native_on_start_command<'local>(
+    env: &mut Env<'local>,
+    _class: JClass<'local>,
+    service: JObject<'local>,
+) -> Result<jint, jni::errors::Error> {
+    ensure_logger();
+    jcall::remember_class_loader(env, &service)?;
+    Ok(notify::on_start_command(env, service))
+}
+
+const _: NativeMethod = native_method! {
+    java_type = "com.example.aiapp.shell.NotificationService",
+    static extern fn native_on_destroy() -> (),
+    error_policy = LogErrorAndDefault,
+};
+
+/// `NotificationService.nativeOnDestroy`. See `notify::on_destroy`.
+fn native_on_destroy<'local>(
+    _env: &mut Env<'local>,
+    _class: JClass<'local>,
+) -> Result<(), jni::errors::Error> {
+    ensure_logger();
+    notify::on_destroy();
+    Ok(())
+}

android-shell/src/notify.rs

@@ -0,0 +1,556 @@
+//! Where a notification is said, and the foreground service that keeps
+//! the connection open while the app is closed. Ported from
+//! `Notifications.kt`'s `NotificationService`, minus the "session on
+//! screen" / "hand to the app as a banner" branches: those read
+//! process-wide state that only exists because a screen is drawn to
+//! register against, and this experiment draws no screen yet (that is
+//! E4's job, on iris). So every notification here takes the third branch
+//! Kotlin's `show` already had -- the platform's own drawer -- which is
+//! also exactly the case E3's pass condition asks for: **a notification
+//! arrives with the app closed.**
+
+use std::sync::atomic::{AtomicBool, Ordering};
+use std::time::Duration;
+
+use client_core::api::UreqTransport;
+use client_core::notifications::{SessionNotification, follow_notifications};
+use jni::Env;
+use jni::errors::Result;
+use jni::objects::{JObject, JValue};
+use jni::sys::{JNI_TRUE, jint};
+
+use crate::settings::{self, ServerSettings};
+
+const ALERT_CHANNEL: &str = "sessions";
+const ONGOING_CHANNEL: &str = "connection";
+const ONGOING_ID: i32 = 1;
+const ALERT_ID: i32 = 2;
+/// Same backoff as `Notifications.kt`'s `RECONNECT_DELAY_MS`.
+const RECONNECT_DELAY: Duration = Duration::from_millis(5_000);
+
+/// Whether the follow-loop thread is already running. **A deviation from
+/// `Notifications.kt`, found by testing rather than planned**: the Kotlin
+/// `onStartCommand` spawns a fresh `thread(isDaemon = true) { follow(...) }`
+/// on *every* call, with nothing to notice a previous one is still going --
+/// and `sync()` calling `startForegroundService` when the service is
+/// already running is an ordinary Android start, not a restart, so
+/// `onStartCommand` runs again. Enrolling from `MainActivity` (which calls
+/// `sync` once itself, then again inside `handle_enrollment` after saving
+/// the token) hits exactly this path and was observed opening **two**
+/// concurrent connections to `/notifications` from one process -- caught
+/// on this build via `adb logcat` showing two `jni::vm::java_vm: Attached
+/// thread ai-app-notifications` lines for one enrollment. Guarded here
+/// rather than left to match Kotlin's behaviour exactly, since duplicating
+/// a live connection is a resource leak with no upside; worth carrying the
+/// same guard back to `Notifications.kt` separately.
+static RUNNING: AtomicBool = AtomicBool::new(false);
+
+/// Set by `nativeOnDestroy`, checked by the follow loop between
+/// reconnects. **Known gap, recorded rather than hidden**: unlike
+/// `HttpURLConnection.disconnect()` in the Kotlin original, nothing here
+/// can interrupt a `ureq` read already blocked inside one connection --
+/// `Transport::stream` hands back a plain `Read` with no cancellation
+/// handle. So a stop lands at the next reconnect, not mid-read. `/notifications`
+/// is idle between events (a keep-alive, per `server/src/routes.rs`), so in
+/// practice this is a bounded wait rather than a hang; closing that gap
+/// for real means adding a cancellation point to `client_core::Transport`,
+/// which is a decision affecting every caller of that trait, not just this
+/// one -- left for whoever next depends on prompt shutdown.
+static STOPPING: AtomicBool = AtomicBool::new(false);
+
+fn static_int(env: &mut Env, class: &str, field: &str) -> Result<i32> {
+    crate::jcall::get_static_field(env, class, field, "I")?.i()
+}
+
+fn notification_manager<'l>(env: &mut Env<'l>, context: &JObject) -> Result<JObject<'l>> {
+    crate::jcall::call_static_method(
+        env,
+        "androidx/core/app/NotificationManagerCompat",
+        "from",
+        "(Landroid/content/Context;)Landroidx/core/app/NotificationManagerCompat;",
+        &[JValue::Object(context)],
+    )?
+    .l()
+}
+
+fn create_channel(
+    env: &mut Env,
+    manager: &JObject,
+    id: &str,
+    name: &str,
+    importance: i32,
+) -> Result<()> {
+    let id_j = crate::jcall::jstr_obj(env, id)?;
+    let builder = crate::jcall::new_object(
+        env,
+        "androidx/core/app/NotificationChannelCompat$Builder",
+        "(Ljava/lang/String;I)V",
+        &[JValue::Object(&id_j), JValue::Int(importance)],
+    )?;
+    let name_j = crate::jcall::jstr_obj(env, name)?;
+    crate::jcall::call_method(
+        env,
+        &builder,
+        "setName",
+        "(Ljava/lang/CharSequence;)Landroidx/core/app/NotificationChannelCompat$Builder;",
+        &[JValue::Object(&name_j)],
+    )?;
+    let channel = crate::jcall::call_method(
+        env,
+        &builder,
+        "build",
+        "()Landroidx/core/app/NotificationChannelCompat;",
+        &[],
+    )?
+    .l()?;
+    crate::jcall::call_method(
+        env,
+        manager,
+        "createNotificationChannel",
+        "(Landroidx/core/app/NotificationChannelCompat;)V",
+        &[JValue::Object(&channel)],
+    )?;
+    Ok(())
+}
+
+/// Two channels, because they are two different things to be told -- see
+/// `Notifications.kt`'s `createChannels` for the reasoning; the names and
+/// importances here are copied from it exactly, since a phone that has
+/// seen both apps should not learn two different vocabularies for the
+/// same fact.
+fn create_channels(env: &mut Env, context: &JObject) -> Result<()> {
+    let manager = notification_manager(env, context)?;
+    let default = static_int(
+        env,
+        "androidx/core/app/NotificationManagerCompat",
+        "IMPORTANCE_DEFAULT",
+    )?;
+    let min = static_int(
+        env,
+        "androidx/core/app/NotificationManagerCompat",
+        "IMPORTANCE_MIN",
+    )?;
+    create_channel(
+        env,
+        &manager,
+        ALERT_CHANNEL,
+        "Sessions needing attention",
+        default,
+    )?;
+    create_channel(env, &manager, ONGOING_CHANNEL, "Staying connected", min)?;
+    Ok(())
+}
+
+fn new_intent_for<'l>(
+    env: &mut Env<'l>,
+    context: &JObject,
+    class_name: &str,
+) -> Result<JObject<'l>> {
+    let target_class = crate::jcall::find_class(env, class_name)?;
+    crate::jcall::new_object(
+        env,
+        "android/content/Intent",
+        "(Landroid/content/Context;Ljava/lang/Class;)V",
+        &[JValue::Object(context), JValue::Object(&target_class)],
+    )
+}
+
+/// The intent a tap on an alert opens -- mirrors `Notifications.kt`'s
+/// `sessionIntent`, including building the URI through `Uri.Builder`
+/// rather than string concatenation, for the same reason: an id needing
+/// escaping must survive the round trip.
+fn session_intent<'l>(
+    env: &mut Env<'l>,
+    context: &JObject,
+    session_id: &str,
+) -> Result<JObject<'l>> {
+    let intent = new_intent_for(env, context, "com/example/aiapp/shell/MainActivity")?;
+    let action_view = crate::jcall::jstr_obj(env, "android.intent.action.VIEW")?;
+    crate::jcall::call_method(
+        env,
+        &intent,
+        "setAction",
+        "(Ljava/lang/String;)Landroid/content/Intent;",
+        &[JValue::Object(&action_view)],
+    )?;
+    let builder = crate::jcall::new_object(env, "android/net/Uri$Builder", "()V", &[])?;
+    let scheme = crate::jcall::jstr_obj(env, settings::SCHEME)?;
+    crate::jcall::call_method(
+        env,
+        &builder,
+        "scheme",
+        "(Ljava/lang/String;)Landroid/net/Uri$Builder;",
+        &[JValue::Object(&scheme)],
+    )?;
+    let authority = crate::jcall::jstr_obj(env, "session")?;
+    crate::jcall::call_method(
+        env,
+        &builder,
+        "authority",
+        "(Ljava/lang/String;)Landroid/net/Uri$Builder;",
+        &[JValue::Object(&authority)],
+    )?;
+    let path = crate::jcall::jstr_obj(env, session_id)?;
+    crate::jcall::call_method(
+        env,
+        &builder,
+        "appendPath",
+        "(Ljava/lang/String;)Landroid/net/Uri$Builder;",
+        &[JValue::Object(&path)],
+    )?;
+    let uri = crate::jcall::call_method(env, &builder, "build", "()Landroid/net/Uri;", &[])?.l()?;
+    crate::jcall::call_method(
+        env,
+        &intent,
+        "setData",
+        "(Landroid/net/Uri;)Landroid/content/Intent;",
+        &[JValue::Object(&uri)],
+    )?;
+    Ok(intent)
+}
+
+fn pending_activity<'l>(
+    env: &mut Env<'l>,
+    context: &JObject,
+    intent: &JObject,
+) -> Result<JObject<'l>> {
+    let update_current = static_int(env, "android/app/PendingIntent", "FLAG_UPDATE_CURRENT")?;
+    let immutable = static_int(env, "android/app/PendingIntent", "FLAG_IMMUTABLE")?;
+    crate::jcall::call_static_method(
+        env,
+        "android/app/PendingIntent",
+        "getActivity",
+        "(Landroid/content/Context;ILandroid/content/Intent;I)Landroid/app/PendingIntent;",
+        &[
+            JValue::Object(context),
+            JValue::Int(0),
+            JValue::Object(intent),
+            JValue::Int(update_current | immutable),
+        ],
+    )?
+    .l()
+}
+
+fn builder_call<'l>(
+    env: &mut Env<'l>,
+    builder: &JObject<'l>,
+    method: &str,
+    sig: &str,
+    args: &[JValue],
+) -> Result<()> {
+    crate::jcall::call_method(env, builder, method, sig, args)?;
+    Ok(())
+}
+
+/// The type Android 14+ requires a foreground service to declare, and
+/// nothing before it -- mirrors `Notifications.kt`'s `foregroundType`.
+fn foreground_type(env: &mut Env) -> Result<i32> {
+    let sdk = static_int(env, "android/os/Build$VERSION", "SDK_INT")?;
+    let upside_down_cake = static_int(env, "android/os/Build$VERSION_CODES", "UPSIDE_DOWN_CAKE")?;
+    if sdk >= upside_down_cake {
+        static_int(
+            env,
+            "android/content/pm/ServiceInfo",
+            "FOREGROUND_SERVICE_TYPE_SPECIAL_USE",
+        )
+    } else {
+        Ok(0)
+    }
+}
+
+fn ongoing_notification<'l>(env: &mut Env<'l>, context: &JObject) -> Result<JObject<'l>> {
+    let channel = crate::jcall::jstr_obj(env, ONGOING_CHANNEL)?;
+    let builder = crate::jcall::new_object(
+        env,
+        "androidx/core/app/NotificationCompat$Builder",
+        "(Landroid/content/Context;Ljava/lang/String;)V",
+        &[JValue::Object(context), JValue::Object(&channel)],
+    )?;
+    let title = crate::jcall::jstr_obj(env, "Watching for sessions that need you")?;
+    builder_call(
+        env,
+        &builder,
+        "setContentTitle",
+        "(Ljava/lang/CharSequence;)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Object(&title)],
+    )?;
+    let icon = static_int(env, "android/R$drawable", "stat_notify_sync")?;
+    builder_call(
+        env,
+        &builder,
+        "setSmallIcon",
+        "(I)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Int(icon)],
+    )?;
+    builder_call(
+        env,
+        &builder,
+        "setOngoing",
+        "(Z)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Bool(JNI_TRUE)],
+    )?;
+    let priority_min = static_int(env, "androidx/core/app/NotificationCompat", "PRIORITY_MIN")?;
+    builder_call(
+        env,
+        &builder,
+        "setPriority",
+        "(I)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Int(priority_min)],
+    )?;
+    crate::jcall::call_method(env, &builder, "build", "()Landroid/app/Notification;", &[])?.l()
+}
+
+/// Starts the service if there is a server to connect to, and stops it
+/// otherwise -- mirrors `Notifications.kt`'s `NotificationService.sync`.
+pub fn sync(env: &mut Env, context: &JObject) -> Result<()> {
+    let service_intent =
+        new_intent_for(env, context, "com/example/aiapp/shell/NotificationService")?;
+    if settings::load(env, context)?.is_none() {
+        crate::jcall::call_method(
+            env,
+            context,
+            "stopService",
+            "(Landroid/content/Intent;)Z",
+            &[JValue::Object(&service_intent)],
+        )?;
+        return Ok(());
+    }
+    create_channels(env, context)?;
+    crate::jcall::call_static_method(
+        env,
+        "androidx/core/content/ContextCompat",
+        "startForegroundService",
+        "(Landroid/content/Context;Landroid/content/Intent;)V",
+        &[JValue::Object(context), JValue::Object(&service_intent)],
+    )?;
+    Ok(())
+}
+
+/// The `Service.onStartCommand` body -- loads settings, starts the
+/// foreground notification, and spawns the follow-loop thread. Answers the
+/// platform's `START_STICKY`/`START_NOT_STICKY` constant, read from the
+/// framework rather than hardcoded so a wrong guess at their values cannot
+/// silently pick the other behaviour.
+pub fn on_start_command(env: &mut Env, service: JObject) -> jint {
+    match try_start(env, &service) {
+        Ok(true) => static_int(env, "android/app/Service", "START_STICKY").unwrap_or(1),
+        Ok(false) => {
+            let _ = crate::jcall::call_method(env, &service, "stopSelf", "()V", &[]);
+            static_int(env, "android/app/Service", "START_NOT_STICKY").unwrap_or(2)
+        }
+        Err(e) => {
+            log_error(env, "onStartCommand", &e);
+            static_int(env, "android/app/Service", "START_NOT_STICKY").unwrap_or(2)
+        }
+    }
+}
+
+fn try_start(env: &mut Env, service: &JObject) -> Result<bool> {
+    let Some(settings) = settings::load(env, service)? else {
+        return Ok(false);
+    };
+    let ca = settings::load_pinned_ca(env)?;
+    let notification = ongoing_notification(env, service)?;
+    let fg_type = foreground_type(env)?;
+    crate::jcall::call_static_method(
+        env,
+        "androidx/core/app/ServiceCompat",
+        "startForeground",
+        "(Landroid/app/Service;ILandroid/app/Notification;I)V",
+        &[
+            JValue::Object(service),
+            JValue::Int(ONGOING_ID),
+            JValue::Object(&notification),
+            JValue::Int(fg_type),
+        ],
+    )?;
+
+    // See `RUNNING`'s doc: a second `onStartCommand` while the loop from
+    // the first is still going -- the ordinary case for this service,
+    // since `sync()` is called from more than one place -- must not open
+    // a second connection.
+    if RUNNING.swap(true, Ordering::SeqCst) {
+        return Ok(true);
+    }
+
+    let vm = env.get_java_vm()?;
+    let context = env.new_global_ref(service)?;
+    STOPPING.store(false, Ordering::SeqCst);
+    std::thread::Builder::new()
+        .name("ai-app-notifications".to_string())
+        .spawn(move || {
+            // Requests a *permanent* attachment (detached only when this thread
+            // exits), matching the Kotlin original's `thread(isDaemon = true)`:
+            // this is the long-lived follow loop, not a one-shot callback.
+            let _: jni::errors::Result<()> = vm.attach_current_thread(|env| {
+                follow_loop(env, &context, settings, &ca);
+                Ok(())
+            });
+        })
+        .ok();
+    Ok(true)
+}
+
+/// Follows the backend's notification stream, reconnecting until stopped
+/// -- mirrors `Notifications.kt`'s `follow`. A dropped connection is the
+/// ordinary case, so it retries quietly and forever; nothing is shown when
+/// it cannot connect, for the same reason as the Kotlin original: a
+/// notification saying "I could not tell you whether anything happened" is
+/// noise about a condition nobody can act on.
+fn follow_loop(env: &mut Env, context: &JObject, settings: ServerSettings, ca: &[u8]) {
+    while !STOPPING.load(Ordering::SeqCst) {
+        if let Ok(transport) = UreqTransport::new(settings.base_url(), settings.token.clone(), ca) {
+            let _ = follow_notifications(&transport, |notification| {
+                if let Err(e) = show(env, context, &notification) {
+                    log_error(env, "show", &e);
+                }
+                !STOPPING.load(Ordering::SeqCst)
+            });
+        }
+        if STOPPING.load(Ordering::SeqCst) {
+            return;
+        }
+        std::thread::sleep(RECONNECT_DELAY);
+    }
+}
+
+/// One notification per session, replacing that session's previous one --
+/// mirrors `Notifications.kt`'s `show`, minus the on-screen/banner
+/// branches this module's doc comment explains.
+fn show(env: &mut Env, context: &JObject, notification: &SessionNotification) -> Result<()> {
+    let manager = notification_manager(env, context)?;
+    let sdk = static_int(env, "android/os/Build$VERSION", "SDK_INT")?;
+    let tiramisu = static_int(env, "android/os/Build$VERSION_CODES", "TIRAMISU")?;
+    let allowed = if sdk < tiramisu {
+        true
+    } else {
+        let permission = crate::jcall::jstr_obj(env, "android.permission.POST_NOTIFICATIONS")?;
+        let granted = static_int(
+            env,
+            "android/content/pm/PackageManager",
+            "PERMISSION_GRANTED",
+        )?;
+        let result = crate::jcall::call_static_method(
+            env,
+            "androidx/core/content/ContextCompat",
+            "checkSelfPermission",
+            "(Landroid/content/Context;Ljava/lang/String;)I",
+            &[JValue::Object(context), JValue::Object(&permission)],
+        )?
+        .i()?;
+        result == granted
+    };
+    let enabled =
+        crate::jcall::call_method(env, &manager, "areNotificationsEnabled", "()Z", &[])?.z()?;
+    if !allowed || !enabled {
+        return Ok(());
+    }
+    let intent = session_intent(env, context, &notification.session_id)?;
+    let pending = pending_activity(env, context, &intent)?;
+    let channel = crate::jcall::jstr_obj(env, ALERT_CHANNEL)?;
+    let builder = crate::jcall::new_object(
+        env,
+        "androidx/core/app/NotificationCompat$Builder",
+        "(Landroid/content/Context;Ljava/lang/String;)V",
+        &[JValue::Object(context), JValue::Object(&channel)],
+    )?;
+    let title = crate::jcall::jstr_obj(env, &notification.title)?;
+    builder_call(
+        env,
+        &builder,
+        "setContentTitle",
+        "(Ljava/lang/CharSequence;)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Object(&title)],
+    )?;
+    let text = crate::jcall::jstr_obj(env, notification.kind.attention_line())?;
+    builder_call(
+        env,
+        &builder,
+        "setContentText",
+        "(Ljava/lang/CharSequence;)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Object(&text)],
+    )?;
+    let icon = static_int(env, "android/R$drawable", "stat_notify_chat")?;
+    builder_call(
+        env,
+        &builder,
+        "setSmallIcon",
+        "(I)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Int(icon)],
+    )?;
+    builder_call(
+        env,
+        &builder,
+        "setContentIntent",
+        "(Landroid/app/PendingIntent;)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Object(&pending)],
+    )?;
+    builder_call(
+        env,
+        &builder,
+        "setAutoCancel",
+        "(Z)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Bool(JNI_TRUE)],
+    )?;
+    let when = (notification.at * 1000.0) as i64;
+    builder_call(
+        env,
+        &builder,
+        "setWhen",
+        "(J)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Long(when)],
+    )?;
+    builder_call(
+        env,
+        &builder,
+        "setShowWhen",
+        "(Z)Landroidx/core/app/NotificationCompat$Builder;",
+        &[JValue::Bool(JNI_TRUE)],
+    )?;
+    let built =
+        crate::jcall::call_method(env, &builder, "build", "()Landroid/app/Notification;", &[])?
+            .l()?;
+    let tag = crate::jcall::jstr_obj(env, &notification.session_id)?;
+    crate::jcall::call_method(
+        env,
+        &manager,
+        "notify",
+        "(Ljava/lang/String;ILandroid/app/Notification;)V",
+        &[
+            JValue::Object(&tag),
+            JValue::Int(ALERT_ID),
+            JValue::Object(&built),
+        ],
+    )?;
+    Ok(())
+}
+
+/// Ends the follow loop -- mirrors `Notifications.kt`'s `onDestroy`, with
+/// the gap this module's `STOPPING` doc explains.
+pub fn on_destroy() {
+    STOPPING.store(true, Ordering::SeqCst);
+    // `RUNNING`'s path out. Same race as `STOPPING` itself (this doc's own
+    // comment): the old thread may still be inside a blocked read when a
+    // new `onStartCommand` follows immediately, which would spawn a
+    // second one before the first has actually stopped. Narrower than not
+    // resetting at all -- a service destroyed and never restarted would
+    // otherwise wedge `RUNNING` true forever -- and no worse than the
+    // known gap already accepted above.
+    RUNNING.store(false, Ordering::SeqCst);
+}
+
+pub fn log_error(env: &mut Env, where_: &str, error: &jni::errors::Error) {
+    let message = format!("android-shell: {where_}: {error}");
+    let _ = (|| -> Result<()> {
+        let tag = crate::jcall::jstr_obj(env, "android-shell")?;
+        let msg = crate::jcall::jstr_obj(env, &message)?;
+        crate::jcall::call_static_method(
+            env,
+            "android/util/Log",
+            "e",
+            "(Ljava/lang/String;Ljava/lang/String;)I",
+            &[JValue::Object(&tag), JValue::Object(&msg)],
+        )?;
+        Ok(())
+    })();
+}

android-shell/src/settings.rs

@@ -0,0 +1,141 @@
+//! Enrollment: where the backend is, and the Keystore-sealed token to
+//! reach it. This crate does not reimplement the Android Keystore AES-GCM
+//! sealing in Rust -- it calls the same `wg-app-link` `ServerStore` Kotlin
+//! class the production app already uses (see `ServerConfig.kt`), through
+//! JNI, for two reasons: that code is shared with Dev Updater and already
+//! tested, and the sealed value on a real phone is keyed to the exact
+//! Keystore alias that class already uses -- reimplementing the crypto
+//! here would either duplicate it or invalidate an existing enrollment.
+
+use jni::Env;
+use jni::errors::Result;
+use jni::objects::{JObject, JString, JValue};
+
+/// Where the backend is and how to authenticate to it -- the Rust twin of
+/// `wg-app-link`'s `ServerSettings` data class, read back field by field
+/// rather than kept as a live JNI reference, so it can cross a thread
+/// boundary (a `JObject` is tied to one `Env`/thread).
+#[derive(Debug, Clone)]
+pub struct ServerSettings {
+    pub host: String,
+    pub port: i32,
+    pub token: String,
+}
+
+impl ServerSettings {
+    pub fn base_url(&self) -> String {
+        format!("https://{}:{}", self.host, self.port)
+    }
+}
+
+/// This experiment's own scheme and Keystore alias -- distinct from the
+/// production app's (`aiapp` / `aiapp-token-key`) so the two can be
+/// installed side by side on the same development device without
+/// colliding over which one a scanned QR or a deep link resolves to. See
+/// RUST.md's E3 entry for why they are not the same value.
+pub(crate) const SCHEME: &str = "aiappshell";
+const KEY_ALIAS: &str = "aiapp-shell-token-key";
+const STORE_CLASS: &str = "com/example/wgapplink/ServerStore";
+const SETTINGS_CLASS: &str = "com/example/wgapplink/ServerSettings";
+
+fn new_store<'l>(env: &mut Env<'l>) -> Result<JObject<'l>> {
+    let scheme = crate::jcall::jstr_obj(env, SCHEME)?;
+    let alias = crate::jcall::jstr_obj(env, KEY_ALIAS)?;
+    crate::jcall::new_object(
+        env,
+        STORE_CLASS,
+        "(Ljava/lang/String;Ljava/lang/String;)V",
+        &[JValue::Object(&scheme), JValue::Object(&alias)],
+    )
+}
+
+fn read_settings(env: &mut Env, settings_obj: &JObject) -> Result<ServerSettings> {
+    let host = get_string(env, settings_obj, "getHost")?;
+    let port = crate::jcall::call_method(env, settings_obj, "getPort", "()I", &[])?.i()?;
+    let token = get_string(env, settings_obj, "getToken")?;
+    Ok(ServerSettings { host, port, token })
+}
+
+fn get_string(env: &mut Env, obj: &JObject, getter: &str) -> Result<String> {
+    let value = crate::jcall::call_method(env, obj, getter, "()Ljava/lang/String;", &[])?.l()?;
+    let jstr: JString = env.cast_local::<JString>(value)?;
+    jstr.try_to_string(env)
+}
+
+/// The stored enrollment, or `None` when there is not one -- mirrors
+/// `ServerConfig.kt`'s `loadServerSettings`.
+pub fn load(env: &mut Env, context: &JObject) -> Result<Option<ServerSettings>> {
+    let store = new_store(env)?;
+    let settings_obj = crate::jcall::call_method(
+        env,
+        &store,
+        "load",
+        "(Landroid/content/Context;)Lcom/example/wgapplink/ServerSettings;",
+        &[JValue::Object(context)],
+    )?
+    .l()?;
+    if settings_obj.is_null() {
+        return Ok(None);
+    }
+    Ok(Some(read_settings(env, &settings_obj)?))
+}
+
+/// Seals and stores `settings` -- mirrors `ServerConfig.kt`'s `saveServerSettings`.
+pub fn save(env: &mut Env, context: &JObject, settings: &ServerSettings) -> Result<()> {
+    let store = new_store(env)?;
+    let host = crate::jcall::jstr_obj(env, &settings.host)?;
+    let token = crate::jcall::jstr_obj(env, &settings.token)?;
+    let settings_obj = crate::jcall::new_object(
+        env,
+        SETTINGS_CLASS,
+        "(Ljava/lang/String;ILjava/lang/String;)V",
+        &[
+            JValue::Object(&host),
+            JValue::Int(settings.port),
+            JValue::Object(&token),
+        ],
+    )?;
+    crate::jcall::call_method(
+        env,
+        &store,
+        "save",
+        "(Landroid/content/Context;Lcom/example/wgapplink/ServerSettings;)V",
+        &[JValue::Object(context), JValue::Object(&settings_obj)],
+    )?;
+    Ok(())
+}
+
+/// Parses an `aiappshell://enroll?...` URI -- mirrors `ServerConfig.kt`'s
+/// `parseEnrollmentUri`, asking the same Kotlin code that already owns the
+/// query-parameter rules rather than re-deriving them here.
+pub fn parse_enrollment_uri(env: &mut Env, uri: &JObject) -> Result<Option<ServerSettings>> {
+    let store = new_store(env)?;
+    let settings_obj = crate::jcall::call_method(
+        env,
+        &store,
+        "parseEnrollmentUri",
+        "(Landroid/net/Uri;)Lcom/example/wgapplink/ServerSettings;",
+        &[JValue::Object(uri)],
+    )?
+    .l()?;
+    if settings_obj.is_null() {
+        return Ok(None);
+    }
+    Ok(Some(read_settings(env, &settings_obj)?))
+}
+
+/// The CA this build pins, generated at build time the same way
+/// `androidApp`'s `generatePinnedCert` task does (see `build.gradle.kts`)
+/// but into a plain Java constant, since this module has no Kotlin of its
+/// own to generate into.
+pub fn load_pinned_ca(env: &mut Env) -> Result<Vec<u8>> {
+    let value = crate::jcall::get_static_field(
+        env,
+        "com/example/aiapp/shell/PinnedCa",
+        "PINNED_CA_PEM",
+        "Ljava/lang/String;",
+    )?
+    .l()?;
+    let jstr: JString = env.cast_local::<JString>(value)?;
+    Ok(jstr.try_to_string(env)?.into_bytes())
+}

android-shell/src/share.rs

@@ -0,0 +1,183 @@
+//! Deep links and the share sheet -- ported from `MainActivity.kt`'s
+//! `handleIntent`/`onNewIntent` and `Share.kt`'s `sharedContent`.
+//!
+//! **Scope cut, recorded rather than silent**: only shared *text*
+//! (`Intent.EXTRA_TEXT`) is attached to a session. `Attachments.kt`'s
+//! upload path -- `ContentResolver` reads of a shared file/photo URI,
+//! bitmap downscaling, EXIF rotation -- is real work of its own and is not
+//! ported here, because `client-core`'s `ApiClient` does not have the
+//! `/sessions/{id}/attachments` route yet either (see `CLIENT_CORE.md`'s
+//! "not covered" list). So `ACTION_SEND`/`ACTION_SEND_MULTIPLE` with a
+//! `content://` stream and no text falls through to a toast saying so,
+//! rather than silently doing nothing. Closing this gap is the same
+//! `client-core` work whichever caller needs it next.
+//!
+//! **Which session a share lands in** is also a placeholder: with no
+//! screen drawn yet (E4's job), there is no picker to ask, so this attaches
+//! to whichever session has the latest `last_activity` -- the one most
+//! likely to be what somebody meant. Worth revisiting once a real screen
+//! exists to ask instead of guessing.
+
+use client_core::api::{ApiClient, UreqTransport};
+use jni::Env;
+use jni::errors::Result;
+use jni::objects::{JObject, JString, JValue};
+
+use crate::notify;
+use crate::settings;
+
+const ACTION_SEND: &str = "android.intent.action.SEND";
+const ACTION_SEND_MULTIPLE: &str = "android.intent.action.SEND_MULTIPLE";
+const ACTION_VIEW: &str = "android.intent.action.VIEW";
+const EXTRA_TEXT: &str = "android.intent.extra.TEXT";
+
+fn get_string_method(env: &mut Env, obj: &JObject, method: &str) -> Result<Option<String>> {
+    let value = crate::jcall::call_method(env, obj, method, "()Ljava/lang/String;", &[])?.l()?;
+    if value.is_null() {
+        return Ok(None);
+    }
+    let jstr: JString = env.cast_local::<JString>(value)?;
+    Ok(Some(jstr.try_to_string(env)?))
+}
+
+fn toast(env: &mut Env, context: &JObject, message: &str) -> Result<()> {
+    let message = crate::jcall::jstr_obj(env, message)?;
+    crate::jcall::call_static_method(
+        env,
+        "com/example/aiapp/shell/MainActivity",
+        "toast",
+        "(Landroid/content/Context;Ljava/lang/String;)V",
+        &[JValue::Object(context), JValue::Object(&message)],
+    )?;
+    Ok(())
+}
+
+/// The one place an incoming intent is sorted into what it means -- mirrors
+/// `MainActivity.kt`'s `handleIntent`.
+pub fn handle_intent(env: &mut Env, activity: &JObject, intent: &JObject) -> Result<()> {
+    let action = get_string_method(env, intent, "getAction")?;
+    if matches!(
+        action.as_deref(),
+        Some(ACTION_SEND) | Some(ACTION_SEND_MULTIPLE)
+    ) {
+        return handle_share(env, activity, intent);
+    }
+    if action.as_deref() != Some(ACTION_VIEW) {
+        return Ok(());
+    }
+    let uri = crate::jcall::call_method(env, intent, "getData", "()Landroid/net/Uri;", &[])?.l()?;
+    if uri.is_null() {
+        return Ok(());
+    }
+    let scheme = get_string_method(env, &uri, "getScheme")?;
+    if scheme.as_deref() != Some(settings::SCHEME) {
+        return Ok(());
+    }
+    match get_string_method(env, &uri, "getHost")?.as_deref() {
+        Some("session") => handle_session_open(env, activity, &uri),
+        Some("enroll") => handle_enrollment(env, activity, &uri),
+        _ => Ok(()),
+    }
+}
+
+fn handle_session_open(env: &mut Env, activity: &JObject, uri: &JObject) -> Result<()> {
+    let Some(session_id) = get_string_method(env, uri, "getLastPathSegment")? else {
+        return Ok(());
+    };
+    // There is no session screen yet (E4's job); the toast is this
+    // experiment's stand-in proof that the tap was routed to the right
+    // session id.
+    toast(env, activity, &format!("Opened session {session_id}"))
+}
+
+fn handle_enrollment(env: &mut Env, activity: &JObject, uri: &JObject) -> Result<()> {
+    match settings::parse_enrollment_uri(env, uri)? {
+        Some(parsed) => {
+            settings::save(env, activity, &parsed)?;
+            notify::sync(env, activity)?;
+            toast(
+                env,
+                activity,
+                &format!("Enrolled with {}", parsed.base_url()),
+            )
+        }
+        None => toast(env, activity, "Not a valid enrollment code"),
+    }
+}
+
+/// The share sheet -- mirrors `Share.kt`'s `sharedContent` for what counts
+/// as a share, and `AttachmentButton`'s upload-then-message pattern for
+/// what happens to it, minus attachments per this module's doc comment.
+fn handle_share(env: &mut Env, activity: &JObject, intent: &JObject) -> Result<()> {
+    let extra_text = crate::jcall::jstr_obj(env, EXTRA_TEXT)?;
+    let text = crate::jcall::call_method(
+        env,
+        intent,
+        "getStringExtra",
+        "(Ljava/lang/String;)Ljava/lang/String;",
+        &[JValue::Object(&extra_text)],
+    )?
+    .l()?;
+    let text = if text.is_null() {
+        None
+    } else {
+        let jstr: JString = env.cast_local::<JString>(text)?;
+        Some(jstr.try_to_string(env)?)
+    };
+    let Some(text) = text.filter(|t| !t.trim().is_empty()) else {
+        return toast(
+            env,
+            activity,
+            "Nothing to share -- only shared text is supported so far",
+        );
+    };
+
+    // Network I/O must not run on the calling thread: `handle_intent` is
+    // called from `onCreate`/`onNewIntent`, both on the main thread, and a
+    // blocking socket read there is a `NetworkOnMainThreadException`. So
+    // the actual send happens on a JNI-attached background thread, the
+    // same shape `notify::try_start`'s follow loop uses; `toast` from that
+    // thread is safe because `MainActivity.toast` itself hops back to the
+    // main looper (see that method).
+    let vm = env.get_java_vm()?;
+    let activity_ref = env.new_global_ref(activity)?;
+    std::thread::spawn(move || {
+        let _: jni::errors::Result<()> = vm.attach_current_thread(|env| {
+            share_in_background(env, &activity_ref, text);
+            Ok(())
+        });
+    });
+    Ok(())
+}
+
+fn share_in_background(env: &mut Env, activity: &JObject, text: String) {
+    let outcome = attach_to_a_session(env, activity, &text);
+    let message = match outcome {
+        Ok(title) => format!("Shared into \"{title}\""),
+        Err(message) => message,
+    };
+    let _ = toast(env, activity, &message);
+}
+
+fn attach_to_a_session(
+    env: &mut Env,
+    activity: &JObject,
+    text: &str,
+) -> std::result::Result<String, String> {
+    let settings = settings::load(env, activity)
+        .map_err(|e| e.to_string())?
+        .ok_or_else(|| "Not enrolled yet".to_string())?;
+    let ca = settings::load_pinned_ca(env).map_err(|e| e.to_string())?;
+    let transport = UreqTransport::new(settings.base_url(), settings.token.clone(), &ca)
+        .map_err(|e| e.to_string())?;
+    let client = ApiClient::new(transport);
+    let sessions = client.fetch_sessions().map_err(|e| e.to_string())?;
+    let target = sessions
+        .into_iter()
+        .max_by(|a, b| a.last_activity.total_cmp(&b.last_activity))
+        .ok_or_else(|| "No session to share into".to_string())?;
+    client
+        .send_message(&target.id, text, &[])
+        .map_err(|e| e.to_string())?;
+    Ok(target.title)
+}

app/androidApp/build.gradle.kts

@@ -102,6 +102,16 @@ android {
         targetSdk = 37
         versionCode = 1
         versionName = "1.0"
+        // Read by MainActivity to decide, at startup, whether this is the P0 benchmark build
+        // (docs/RUST.md's P0 box) rather than the app somebody enrolled. False everywhere except
+        // the `bench` build type below, which overrides it.
+        buildConfigField("boolean", "FIXTURE_MODE", "false")
+    }
+    buildFeatures {
+        // Only for FIXTURE_MODE above; nothing else here reaches for generated BuildConfig fields.
+        buildConfig = true
+        // Only for the bench build type's resValue("string", "app_name", ...) below.
+        resValues = true
     }
     packaging {
         resources { excludes += "/META-INF/{AL2.0,LGPL2.1}" }
@@ -131,6 +141,35 @@ android {
             isMinifyEnabled = false
             if (keystore != null) signingConfig = signingConfigs.getByName("release")
         }
+        // P0's benchmark build (docs/RUST.md, docs/DECISIONS.md's 2026-09-05 entry): release
+        // optimisations so a frame time measured here means what release means everywhere else in
+        // this project, its own application id so it installs beside a real enrollment rather than
+        // replacing it, and FIXTURE_MODE so MainActivity opens straight onto the fixture session
+        // instead of asking to be enrolled. Signed with the same key as release -- it never talks
+        // to a real backend, so there is no CA of its own to mismatch, and a second keystore would
+        // be one more secret to keep off this machine's shared mount for no benefit.
+        create("bench") {
+            initWith(getByName("release"))
+            // :link (wg-app-link) has no "bench" build type of its own -- it is a library shared
+            // with dev-updater and has no reason to know this project invented one -- so this says
+            // which of its build types to link against instead.
+            matchingFallbacks += listOf("release")
+            applicationIdSuffix = ".bench"
+            // "AI Sessions bench" everywhere the OS shows the app's name (launcher, recents,
+            // Settings): this resValue overrides res/values/strings.xml's app_name for this
+            // build type alone, and AndroidManifest.xml's android:label reads @string/app_name
+            // rather than a literal so a build type can override it without touching the
+            // manifest.
+            resValue("string", "app_name", "AI Sessions bench")
+            buildConfigField("boolean", "FIXTURE_MODE", "true")
+            if (keystore != null) signingConfig = signingConfigs.getByName("release")
+        }
+    }
+    sourceSets {
+        // The fixture both bench builds (this one and iris's) open with; see
+        // app/bench-fixture/README.md. Read directly from its own directory rather than copied
+        // into androidApp/src -- one file to keep in sync with the generator, not two.
+        getByName("bench").assets.directories.add("../bench-fixture/assets")
     }
     compileOptions {
         sourceCompatibility = JavaVersion.VERSION_21

app/androidApp/src/main/AndroidManifest.xml

@@ -25,7 +25,7 @@
          the fix is a judgement about how this app should look. Drop this
          suppression when a real icon lands. -->
     <application
-        android:label="AI Sessions"
+        android:label="@string/app_name"
         android:allowBackup="true"
         android:theme="@android:style/Theme.Material.Light.NoActionBar"
         tools:ignore="MissingApplicationIcon">

app/androidApp/src/main/kotlin/com/example/aiapp/Api.kt

@@ -142,6 +142,20 @@ data class SessionSummary(
     val keepsOwnTranscript: Boolean,
     /** How much the session asks before acting; null when it was never set. */
     val permissionMode: String?,
+    /**
+     * How hard the model thinks, or null for the CLI's own default.
+     *
+     * Null is a level somebody can choose, not only one to start in -- see [EFFORT_LEVELS]. It is
+     * reported rather than assumed for the same reason [permissionMode] is.
+     */
+    val effort: String?,
+    /**
+     * Whether a thinking level does anything here -- a Claude CLI session, not a llama or echo one.
+     *
+     * Asked of the server rather than worked out from the provider's name, because this is a
+     * property of the driver's *kind* and the phone only has the name.
+     */
+    val takesEffort: Boolean,
     /**
      * Whether this continues a session the machine already had, which changes what deleting means.
      */
@@ -153,6 +167,24 @@ data class SessionSummary(
      * itself from a default is one you can turn off while believing you are reading it.
      */
     val notify: Boolean,
+    /**
+     * Whether this session sends itself a message once its account's usage limit lifts, and what
+     * that message says.
+     *
+     * The message is what the server would actually send, with its own default already filled in,
+     * so the field shows the words rather than an empty box standing for them.
+     */
+    val autoResume: Boolean,
+    val autoResumeMessage: String,
+    /**
+     * When the server next intends to check whether the limit has lifted, in epoch seconds, or null
+     * when nothing is waiting.
+     *
+     * A time to *ask*, not a time to resume: the server checks the meter at that moment and waits
+     * again if the limit is still on. Worded that way wherever it is shown, because a promise this
+     * app cannot keep is worse than no time at all.
+     */
+    val resumeAt: Double?,
     /**
      * The directory the session works in, or null where it was never given one.
      *
@@ -178,8 +210,27 @@ data class SessionSummary(
      * server because that is where a provider's kind is known.
      */
     val maxImageEdge: Int?,
+    /**
+     * Which of `GET /usage`'s snapshots is about this session, and null where nothing meters it.
+     *
+     * The rate-limit bar answers a question about an *account*, and what decides which account --
+     * if any -- is the provider this session runs, not the machine it runs on. Pairing by machine
+     * alone drew the Claude CLI's five-hour window under every echo session on a machine that also
+     * has the CLI: a quota that session cannot spend and could never run down. Decided by the
+     * server for the same reason [maxImageEdge] is -- it is a fact about the provider's kind, and
+     * this app has only its name.
+     */
+    val usageProvider: String?,
     val status: String,
     val lastActivity: Double,
+    /**
+     * How many subagents this session has, however their own status now reads.
+     *
+     * A directory listing on the server rather than a status read per subagent, so the list stays
+     * cheap; the per-subagent state is only fetched when the card is expanded. Zero on a server
+     * that predates subagents, so this app still opens against one.
+     */
+    val subagents: Int,
 )
 
 private fun parseSession(session: JSONObject) =
@@ -192,14 +243,24 @@ private fun parseSession(session: JSONObject) =
         title = session.getString("title"),
         model = session.optString("model").ifEmpty { null },
         permissionMode = session.optString("permissionMode").ifEmpty { null },
+        effort = session.optString("effort").ifEmpty { null },
+        takesEffort = session.optBoolean("takesEffort", false),
         imported = session.optBoolean("imported", false),
         notify = session.optBoolean("notify", true),
+        autoResume = session.optBoolean("autoResume", false),
+        // The server sends its own default rather than nothing, so an empty answer means an older
+        // server -- and this app's word for it is the same word.
+        autoResumeMessage =
+            session.optString("autoResumeMessage").ifEmpty { DEFAULT_RESUME_MESSAGE },
+        resumeAt = if (session.has("resumeAt")) session.getDouble("resumeAt") else null,
         cwd = session.optString("cwd").ifEmpty { null },
         contextTokens =
             if (session.has("contextTokens")) session.getLong("contextTokens") else null,
         maxImageEdge = session.optInt("maxImageEdge", 0).takeIf { it > 0 },
+        usageProvider = session.optString("usageProvider").ifEmpty { null },
         status = session.getString("status"),
         lastActivity = session.getDouble("lastActivity"),
+        subagents = session.optInt("subagents", 0),
     )
 
 fun fetchSessions(settings: ServerSettings): List<SessionSummary> =
@@ -215,6 +276,35 @@ fun fetchSessions(settings: ServerSettings): List<SessionSummary> =
 fun fetchSession(settings: ServerSettings, sessionId: String): SessionSummary =
     requestFromServer(settings, "/sessions/$sessionId") { parseSession(it.jsonObject()) }
 
+/**
+ * One row of `GET /sessions/{id}/subagents`, oldest first.
+ *
+ * A subagent is a second transcript owned by a session -- no process, no controls of its own -- so
+ * this carries only what a card needs to draw and to open it; see SUBAGENTS.md. [status] is
+ * "running", "exited" or "unknown": a subagent whose session is not itself running cannot be
+ * running, and the list says so rather than reporting a state that cannot hold.
+ */
+data class SubagentSummary(
+    val id: String,
+    val title: String,
+    val status: String,
+    val created: Double,
+    val lastActivity: Double,
+)
+
+fun fetchSubagents(settings: ServerSettings, sessionId: String): List<SubagentSummary> =
+    requestFromServer(settings, "/sessions/$sessionId/subagents") {
+        it.jsonObjects { row ->
+            SubagentSummary(
+                id = row.getString("id"),
+                title = row.getString("title"),
+                status = row.getString("status"),
+                created = row.getDouble("created"),
+                lastActivity = row.getDouble("lastActivity"),
+            )
+        }
+    }
+
 // What the server offers, so the spawn screen has no hardcoded lists: a setup added to the server's
 // config.ron appears here with no app rebuild.
 //
@@ -375,6 +465,12 @@ data class SshDetails(
      * Where files attached from here land on that machine; null for the session's own directory.
      */
     val attachmentsDir: String? = null,
+    /**
+     * Where that machine keeps its GGUF models; null for the same place the backend keeps its own
+     * (`~/.local/share/ai-app/models`, read on that machine). A llama.cpp session serves the file
+     * from the machine it runs on, so this is where its models are looked for and listed.
+     */
+    val modelsDir: String? = null,
 )
 
 private fun SshDetails.toJson() =
@@ -382,6 +478,7 @@ private fun SshDetails.toJson() =
         if (port != null) put("port", port)
         if (!identityFile.isNullOrBlank()) put("identityFile", identityFile)
         if (!attachmentsDir.isNullOrBlank()) put("attachmentsDir", attachmentsDir)
+        if (!modelsDir.isNullOrBlank()) put("modelsDir", modelsDir)
     }
 
 /** What a machine turns out to have, without saving anything. */
@@ -450,6 +547,8 @@ fun spawnSession(
     model: String? = null,
     cwd: String? = null,
     permissionMode: String? = null,
+    /** Null for whatever the server's default is; see [fetchDefaultEffort]. */
+    effort: String? = null,
     params: Map<String, String> = emptyMap(),
     /** Continue this Claude Code session instead of starting an empty one. */
     import: String? = null,
@@ -467,6 +566,7 @@ fun spawnSession(
                     if (!model.isNullOrBlank()) put("model", model)
                     if (!cwd.isNullOrBlank()) put("cwd", cwd)
                     if (!permissionMode.isNullOrBlank()) put("permissionMode", permissionMode)
+                    if (!effort.isNullOrBlank()) put("effort", effort)
                     if (!import.isNullOrBlank()) put("import", import)
                     if (params.isNotEmpty()) {
                         put("params", JSONObject(params.toMap<String, Any>()))
@@ -906,7 +1006,7 @@ fun startImport(
  */
 fun fetchTranscript(
     settings: ServerSettings,
-    sessionId: String,
+    address: TranscriptAddress,
     before: Long? = null,
     limit: Int = 80,
     // Count [limit] in rows, not events, joining a reply's streamed deltas into one -- so a page of
@@ -925,7 +1025,7 @@ fun fetchTranscript(
         if (coalesce) append("&coalesce=true")
         if (after != null) append("&after=").append(after)
     }
-    return requestFromServer(settings, "/sessions/$sessionId/transcript$query") { connection ->
+    return requestFromServer(settings, "/${address.urlPath}/transcript$query") { connection ->
         val body = JSONArray(connection.inputStream.bufferedReader().readText())
         // The text as well as the event: the transcript cache stores the one and the fold needs the
         // other, and they have to be the same line.
@@ -973,6 +1073,56 @@ fun setSessionModel(settings: ServerSettings, sessionId: String, model: String)
  */
 val PERMISSION_MODES = listOf("manual", "acceptEdits", "auto", "bypassPermissions", "plan")
 
+/**
+ * What a new session's thinking level is when nothing chose one, or null for the CLI's own.
+ *
+ * Held by the server rather than by this phone, because a second device would otherwise spawn
+ * sessions at a level the first one's owner never picked.
+ */
+fun fetchDefaultEffort(settings: ServerSettings): String? =
+    requestFromServer(settings, "/defaults") {
+        it.jsonObject().optString("effort").ifEmpty { null }
+    }
+
+/** Sets what new sessions start at. Nothing already running changes. */
+fun setDefaultEffort(settings: ServerSettings, level: String?) {
+    requestFromServer(
+        settings,
+        "/defaults",
+        method = "POST",
+        jsonBody = JSONObject().put("effort", level ?: JSONObject.NULL).toString(),
+    ) {}
+}
+
+/**
+ * How hard the model thinks, as `claude --effort` takes them, cheapest first.
+ *
+ * Not offered alongside the model and the permission mode on the session's own bar, because it does
+ * not behave like them: the CLI has a control request for those two and none for this (checked
+ * against 2.1.258), so a level is settled when the process is launched. Changing it therefore stops
+ * the process, which is what the working directory beside it in this dialog does, and why it is
+ * here rather than on a bar whose other controls take effect mid-turn.
+ */
+val EFFORT_LEVELS = listOf("low", "medium", "high", "xhigh", "max")
+
+/** What the picker shows, and sends as null, for a session that has chosen no level. */
+const val DEFAULT_EFFORT = "default"
+
+/**
+ * Records how hard a session thinks and **stops its process**, since the level is read when the
+ * process is launched. The next message, or Start, runs one that has it.
+ *
+ * [level] is null for the CLI's own default.
+ */
+fun setSessionEffort(settings: ServerSettings, sessionId: String, level: String?) {
+    requestFromServer(
+        settings,
+        "/sessions/$sessionId/effort",
+        method = "POST",
+        jsonBody = JSONObject().put("effort", level ?: JSONObject.NULL).toString(),
+    ) {}
+}
+
 /** Switches how much a running session asks before acting, also in place. */
 fun setSessionPermissionMode(settings: ServerSettings, sessionId: String, mode: String) {
     requestFromServer(
@@ -984,6 +1134,36 @@ fun setSessionPermissionMode(settings: ServerSettings, sessionId: String, mode:
 }
 
 /** Turns this session's notifications on or off. Stored on the backend -- see `SessionConfig`. */
+/**
+ * What an auto-resume says when nothing else was typed. Mirrors the server's own default, so a
+ * cleared field shows the word that would actually be sent instead of going blank.
+ */
+const val DEFAULT_RESUME_MESSAGE = "continue"
+
+/**
+ * Turns auto-resume on or off and sets what it would say, in one request because they are one
+ * decision -- see the server's `/sessions/{id}/auto-resume`.
+ */
+fun setSessionAutoResume(
+    settings: ServerSettings,
+    sessionId: String,
+    autoResume: Boolean,
+    message: String?,
+) {
+    requestFromServer(
+        settings,
+        "/sessions/$sessionId/auto-resume",
+        method = "POST",
+        jsonBody =
+            JSONObject()
+                .put("autoResume", autoResume)
+                // Empty means the server's default rather than a session poked with nothing to
+                // read, which is the same rule the server applies to the field.
+                .put("message", message?.trim()?.ifEmpty { null } ?: JSONObject.NULL)
+                .toString(),
+    ) {}
+}
+
 fun setSessionNotify(settings: ServerSettings, sessionId: String, notify: Boolean) {
     requestFromServer(
         settings,
@@ -1074,6 +1254,26 @@ private fun parseDownload(o: JSONObject) =
         error = if (o.has("error")) o.getString("error") else null,
     )
 
+/**
+ * The models on one machine, which is the list a llama.cpp session there can choose from.
+ *
+ * Not [fetchModels], which is what the *backend* has downloaded. A session serves its model from
+ * the machine it runs on, so for a machine reached over ssh those are two different lists -- and
+ * offering the backend's would name files that are not there, turning a choice that cannot work
+ * into a session that fails when it tries to load one.
+ */
+fun fetchSetupModels(settings: ServerSettings, setupId: String): List<LocalModel> =
+    requestFromServer(settings, "/setups/${setupId.urlEncoded()}/models") { connection ->
+        JSONArray(connection.inputStream.bufferedReader().readText()).mapObjects { m ->
+            LocalModel(
+                key = m.getString("key"),
+                repo = m.getString("repo"),
+                file = m.getString("file"),
+                bytes = m.getLong("bytes"),
+            )
+        }
+    }
+
 fun fetchModels(settings: ServerSettings): Models =
     requestFromServer(settings, "/models") { connection ->
         val body = JSONObject(connection.inputStream.bufferedReader().readText())

app/androidApp/src/main/kotlin/com/example/aiapp/AppRoot.kt

@@ -1,7 +1,9 @@
 package com.example.aiapp
 
 import androidx.activity.compose.BackHandler
+import androidx.compose.foundation.background
 import androidx.compose.foundation.layout.Box
+import androidx.compose.foundation.layout.fillMaxSize
 import androidx.compose.foundation.layout.imePadding
 import androidx.compose.foundation.layout.padding
 import androidx.compose.material3.AlertDialog
@@ -34,7 +36,23 @@ import kotlinx.coroutines.withContext
  * session, spawning one, and settings.
  */
 private sealed class Screen {
-    data object Main : Screen()
+    /**
+     * The session list, with a subagent's own transcript over it when [subagent] is set.
+     *
+     * A layer on this screen rather than a screen of its own, for the same reason [Session.files]
+     * is: [SessionListScreen] owns which cards are expanded and what each expansion fetched, kept
+     * in `remember`, and a subagent is opened from a card's expander. As a sibling `Screen` it was
+     * disposed and recreated on every return, which lost that state -- an expanded card collapsed
+     * itself the moment its own subagent's view was closed.
+     */
+    data class Main(val subagent: SubagentTarget? = null) : Screen()
+
+    /**
+     * One subagent's own transcript, read-only. See [SessionScreen]'s `subagent` parameter and
+     * SUBAGENTS.md's "Phone". Closing it returns to [Main] under it, not to [Session]: a subagent
+     * is opened from the session list's card rather than from inside the session it belongs to.
+     */
+    data class SubagentTarget(val summary: SessionSummary, val subagent: SubagentSummary)
 
     /**
      * One session, with the file explorer over it when [files] is set.
@@ -81,7 +99,7 @@ fun AppRoot(
     val context = LocalContext.current
     val scope = rememberCoroutineScope()
     var settings by remember(settingsVersion) { mutableStateOf(loadServerSettings(context)) }
-    var screen by remember { mutableStateOf<Screen>(Screen.Main) }
+    var screen by remember { mutableStateOf<Screen>(Screen.Main()) }
     // A notification tap this could not follow, and why. Null both before one is asked for and
     // after one succeeds, since success is a screen rather than a message.
     var failedOpen by remember { mutableStateOf<FailedOpen?>(null) }
@@ -96,7 +114,7 @@ fun AppRoot(
             share = shareRequest
             // A session already open takes it. Otherwise the list is where the choice is made,
             // whatever screen was showing: Spawn and Settings have nowhere to put a file.
-            if (screen !is Screen.Session) screen = Screen.Main
+            if (screen !is Screen.Session) screen = Screen.Main()
         }
     }
 
@@ -123,7 +141,7 @@ fun AppRoot(
                 existing = null,
                 onSaved = { saved ->
                     settings = saved
-                    screen = Screen.Main
+                    screen = Screen.Main()
                 },
                 onBack = null,
             )
@@ -136,7 +154,7 @@ fun AppRoot(
     // shows, so it always refetches.
     val goToMain = {
         reloadToken++
-        screen = Screen.Main
+        screen = Screen.Main()
     }
     if (screen !is Screen.Main) {
         BackHandler(onBack = goToMain)
@@ -185,6 +203,9 @@ fun AppRoot(
                     reloadToken = reloadToken,
                     share = share,
                     onOpen = { screen = Screen.Session(it) },
+                    onOpenSubagent = { summary, subagent ->
+                        screen = here.copy(subagent = Screen.SubagentTarget(summary, subagent))
+                    },
                     onSpawn = { screen = Screen.Spawn },
                     onImported = { imported ->
                         reloadToken++
@@ -192,6 +213,27 @@ fun AppRoot(
                     },
                     onSettings = { screen = Screen.Settings },
                 )
+                // Its own back handler is registered after MainScreen's, so it is the one the
+                // platform asks first while a subagent is open -- the same rule the files
+                // explorer's handler follows over its session, below.
+                here.subagent?.let { target ->
+                    BackHandler { screen = here.copy(subagent = null) }
+                    // Its own opaque background: this screen was always the sole content under
+                    // the theme's own Surface before, so it never had to paint one -- stacked over
+                    // the list here, the space between its own cards let the list underneath show
+                    // through without this. The same fix FilesScreen needed over its session.
+                    Box(Modifier.fillMaxSize().background(MaterialTheme.colorScheme.background)) {
+                        key(target.summary.id, target.subagent.id) {
+                            SessionScreen(
+                                settings = current,
+                                summary = target.summary,
+                                onBack = { screen = here.copy(subagent = null) },
+                                onFiles = {},
+                                subagent = target.subagent,
+                            )
+                        }
+                    }
+                }
             }
         is Screen.Session ->
             // Keyed on the id, because a different session is a different screen rather than this

app/androidApp/src/main/kotlin/com/example/aiapp/BenchFixture.kt

@@ -0,0 +1,108 @@
+package com.example.aiapp
+
+import android.content.Context
+import java.util.concurrent.CopyOnWriteArrayList
+
+/**
+ * P0's benchmark gate (see docs/RUST.md and docs/DECISIONS.md's 2026-09-05 entry): an in-process
+ * fake of the backend, so the `bench` build type can drive a real session screen -- the real
+ * [TranscriptSource], the real fold, the real paging -- with no server and no network permission.
+ *
+ * Only ever installed when [BuildConfig.FIXTURE_MODE] is true (see [MainActivity]); everything else
+ * in this build compiles it in but never calls it, since Kotlin has no per-build-type source set
+ * that both [MainActivity] (which every variant compiles) and this can share without one.
+ *
+ * The design: [requestFromServer] and [Sse] talk to `https://$FIXTURE_HOST:$FIXTURE_PORT` through
+ * ordinary `java.net.URL`, exactly as they would talk to a real server. A
+ * [java.net.URLStreamHandlerFactory] registered once for the whole process intercepts every
+ * `https://` connection to that host and answers from this object's in-memory event log instead of
+ * opening a socket -- see BenchNetwork.kt. Everything above that (TranscriptSource, SessionScreen,
+ * the fold, uniqueItems) never learns the difference.
+ */
+object BenchFixture {
+    const val FIXTURE_HOST = "bench.fixture.invalid"
+    const val FIXTURE_PORT = 1
+
+    /** How many of the fixture's events are the opening backlog; see bench-fixture/README.md. */
+    private const val BACKLOG_COUNT = 3200
+
+    val settings = ServerSettings(FIXTURE_HOST, FIXTURE_PORT, "bench")
+
+    /** The session id every bench run opens; nothing else in this build ever mints one. */
+    const val SESSION_ID = "bench-fixture-session"
+
+    /**
+     * The whole transcript, seq order, growing as [pushLive] is called during the streaming phase.
+     * Read by both the REST page handler and the SSE handler, so a page requested mid- stream and a
+     * live frame agree on what has "already happened" -- the same thing a real server's own
+     * transcript file guarantees.
+     */
+    private val log = CopyOnWriteArrayList<Pair<String, SeqEvent>>()
+
+    /** The events not yet appended to [log] -- the streaming phase's own source. */
+    private var streamTail: List<Pair<String, SeqEvent>> = emptyList()
+
+    private val images = mutableMapOf<String, ByteArray>()
+
+    @Volatile private var loaded = false
+
+    /**
+     * Parses the bundled fixture once. Safe to call more than once; only the first does anything.
+     */
+    @Synchronized
+    fun ensureLoaded(context: Context) {
+        if (loaded) return
+        val lines =
+            context.assets.open("transcript.jsonl").bufferedReader().readLines().filter {
+                it.isNotBlank()
+            }
+        val parsed = lines.map { it to parseSeqEvent(it) }
+        log.addAll(parsed.take(BACKLOG_COUNT))
+        streamTail = parsed.drop(BACKLOG_COUNT)
+        for (name in listOf("bench1.png", "bench2.png")) {
+            images[name] = context.assets.open(name).readBytes()
+        }
+        loaded = true
+    }
+
+    /** The events the streaming phase has left to send. */
+    fun remainingStreamEvents(): Int = streamTail.size
+
+    /** Sends the next fixture event onto the live log, as a real SSE frame would arrive. */
+    fun pushNextLiveEvent(): Boolean {
+        val next = streamTail.firstOrNull() ?: return false
+        streamTail = streamTail.drop(1)
+        log.add(next)
+        return true
+    }
+
+    /** Undoes [pushNextLiveEvent] and reloads the opening backlog, for running the bench twice. */
+    @Synchronized
+    fun resetToBacklog(context: Context) {
+        loaded = false
+        log.clear()
+        ensureLoaded(context)
+    }
+
+    fun fileBytes(name: String): ByteArray? = images[name]
+
+    /**
+     * Raw JSON lines with seq > [after], in order -- what an `/events?after=` connection replays.
+     */
+    fun linesAfter(after: Long): List<String> =
+        log.filter { it.second.seq > after }.map { it.first }
+
+    /**
+     * One REST page: [fetchTranscript]'s `before`/`limit`/`after`, against the growing log. Ignores
+     * `coalesce` -- the fixture's own deltas are already split the way a real reply streams, and
+     * what the benchmark exercises is the fold and the paging, not the server's row-joining, which
+     * client-core's own port tracks separately (CLIENT_CORE.md).
+     */
+    fun page(before: Long?, limit: Int, after: Long?): List<String> {
+        val upper = before ?: (log.lastOrNull()?.second?.seq?.plus(1) ?: 1L)
+        val candidates = log.filter {
+            it.second.seq < upper && (after == null || it.second.seq > after)
+        }
+        return candidates.takeLast(limit).map { it.first }
+    }
+}

app/androidApp/src/main/kotlin/com/example/aiapp/BenchNetwork.kt

@@ -0,0 +1,181 @@
+package com.example.aiapp
+
+import java.io.ByteArrayInputStream
+import java.io.IOException
+import java.io.InputStream
+import java.io.PipedInputStream
+import java.io.PipedOutputStream
+import java.net.HttpURLConnection
+import java.net.URL
+import java.net.URLStreamHandler
+import java.net.URLStreamHandlerFactory
+import java.security.Principal
+import java.security.cert.Certificate
+import javax.net.ssl.HttpsURLConnection
+import javax.net.ssl.SSLPeerUnverifiedException
+import org.json.JSONArray
+
+/**
+ * Installs the process-wide interception [BenchFixture] needs. Idempotent and safe to call more
+ * than once; the JDK only allows [URL.setURLStreamHandlerFactory] to be called successfully once
+ * per process, and a second real call throws -- so this guards it rather than relying on every
+ * caller to remember.
+ *
+ * Scoped to [BenchFixture.FIXTURE_HOST]: any other `https://` URL falls through to the platform's
+ * ordinary handler, so this only ever changes behaviour for the one host the bench build invents.
+ */
+@Synchronized
+fun installFixtureNetworkOnce() {
+    if (installed) return
+    installed = true
+    URL.setURLStreamHandlerFactory(
+        URLStreamHandlerFactory { protocol ->
+            if (protocol != "https") null
+            else
+                object : URLStreamHandler() {
+                    override fun openConnection(url: URL): HttpURLConnection =
+                        if (url.host == BenchFixture.FIXTURE_HOST) FixtureConnection(url)
+                        else
+                        // The bench build makes no other https call -- this factory is
+                        // installed only in FIXTURE_MODE (MainActivity) -- so there is
+                        // deliberately no delegate to a platform handler here: once a
+                        // URLStreamHandlerFactory is installed there is no supported way to
+                        // ask the JDK for its own default handler back, and re-entering this
+                        // same factory for the fallback would recurse forever rather than
+                        // reach one.
+                        throw java.io.IOException(
+                                "bench build's fixture network has no route to https host " +
+                                    "${url.host} -- only ${BenchFixture.FIXTURE_HOST} is served"
+                            )
+                }
+        }
+    )
+}
+
+private var installed = false
+
+/**
+ * Answers one request against [BenchFixture] instead of opening a socket. Implements just enough of
+ * [HttpsURLConnection] for [requestFromServer] and [Sse] to work unmodified: both only call
+ * `connect`/`disconnect`, set a handful of request properties they never need answered, and read
+ * `responseCode` and `inputStream`.
+ */
+private class FixtureConnection(url: URL) : HttpsURLConnection(url) {
+    private var input: InputStream? = null
+    private var writer: Thread? = null
+
+    override fun connect() {
+        if (input != null) return
+        input = route(url.path, url.query)
+    }
+
+    override fun disconnect() {
+        writer?.interrupt()
+        try {
+            input?.close()
+        } catch (_: IOException) {}
+    }
+
+    override fun usingProxy() = false
+
+    override fun getResponseCode(): Int {
+        connect()
+        return 200
+    }
+
+    override fun getInputStream(): InputStream {
+        connect()
+        return input!!
+    }
+
+    override fun getErrorStream(): InputStream? = null
+
+    // Nothing here reads any of these; implemented only because HttpsURLConnection declares them
+    // abstract. A fixture never negotiates real TLS, so each says exactly that rather than
+    // fabricating a plausible-looking certificate.
+    override fun getCipherSuite() = "none (bench fixture, no TLS)"
+
+    override fun getLocalCertificates(): Array<Certificate>? = null
+
+    override fun getServerCertificates(): Array<Certificate> =
+        throw SSLPeerUnverifiedException("bench fixture connection presents no certificate")
+
+    override fun getPeerPrincipal(): Principal =
+        throw SSLPeerUnverifiedException("bench fixture connection presents no certificate")
+
+    override fun getLocalPrincipal(): Principal? = null
+
+    /**
+     * [path] is `/sessions/{id}/...`; everything else this build's fixture is asked for is a bug.
+     */
+    private fun route(path: String, query: String?): InputStream {
+        val params =
+            (query ?: "")
+                .split("&")
+                .filter { it.contains('=') }
+                .associate {
+                    val (k, v) = it.split("=", limit = 2)
+                    k to java.net.URLDecoder.decode(v, "UTF-8")
+                }
+        return when {
+            path.endsWith("/transcript") -> {
+                val lines =
+                    BenchFixture.page(
+                        before = params["before"]?.toLongOrNull(),
+                        limit = params["limit"]?.toIntOrNull() ?: 80,
+                        after = params["after"]?.toLongOrNull(),
+                    )
+                val body = JSONArray(lines.map { org.json.JSONObject(it) })
+                ByteArrayInputStream(body.toString().toByteArray())
+            }
+            path.endsWith("/events") -> openEventsStream(params["after"]?.toLongOrNull() ?: 0L)
+            path.contains("/files/") -> {
+                val name = path.substringAfterLast("/files/")
+                val bytes =
+                    BenchFixture.fileBytes(name)
+                        ?: throw IOException("bench fixture has no file named $name")
+                ByteArrayInputStream(bytes)
+            }
+            else -> throw IOException("bench fixture has no route for $path")
+        }
+    }
+
+    /**
+     * A live SSE body: [BenchFixture.linesAfter] replayed immediately, then polled every 50ms for
+     * anything [BenchFixture.pushNextLiveEvent] has added since -- the same shape a real backend's
+     * backlog-then-follow gives [Sse], just polled instead of woken, which is a fixture's business
+     * rather than something worth a condition variable for.
+     */
+    private fun openEventsStream(after: Long): InputStream {
+        val pipeIn = PipedInputStream(1 shl 16)
+        val pipeOut = PipedOutputStream(pipeIn)
+        var sent = after
+        val thread = Thread {
+            try {
+                while (!Thread.currentThread().isInterrupted) {
+                    val fresh = BenchFixture.linesAfter(sent)
+                    for (line in fresh) {
+                        pipeOut.write("data: $line\n\n".toByteArray())
+                        pipeOut.flush()
+                        sent = org.json.JSONObject(line).getLong("seq")
+                    }
+                    Thread.sleep(50)
+                }
+            } catch (_: InterruptedException) {
+                // disconnect() -- the ordinary way this ends.
+            } catch (_: IOException) {
+                // The reader side (Sse) closed its end.
+            } finally {
+                try {
+                    pipeOut.close()
+                } catch (_: IOException) {}
+            }
+        }
+            .also {
+                it.isDaemon = true
+                it.start()
+            }
+        writer = thread
+        return pipeIn
+    }
+}

app/androidApp/src/main/kotlin/com/example/aiapp/BenchRun.kt

@@ -0,0 +1,325 @@
+package com.example.aiapp
+
+import android.content.Context
+import android.os.BatteryManager
+import android.os.Process
+import android.view.View
+import androidx.compose.foundation.gestures.FlingBehavior
+import androidx.compose.foundation.lazy.LazyListState
+import androidx.compose.ui.focus.FocusRequester
+import androidx.core.view.ViewCompat
+import androidx.core.view.WindowInsetsCompat
+import androidx.core.view.WindowInsetsControllerCompat
+import java.io.File
+import kotlinx.coroutines.CoroutineScope
+import kotlinx.coroutines.delay
+import kotlinx.coroutines.isActive
+import kotlinx.coroutines.launch
+
+/**
+ * P0's scripted benchmark, run in-process instead of by a shell script: the phone has no usable
+ * system tracing (this-machine-android's skill) and no agent can drive it, so the same scroll loop
+ * and streaming phase `transcript-bench.sh`/`stream-bench.sh` drive over `ui-trace` are reproduced
+ * here against [LazyListState] and [BenchFixture] directly. Only reachable from the `bench` build
+ * (see [SessionSettingsDialog]'s `onRunBenchmark`), but compiled into every build for the reason
+ * [BenchFixture]'s doc comment gives.
+ *
+ * **v2 (2026-09-06)**, asked for by Iris because the v1 fling was too gentle to stress-test the
+ * scroll path and said nothing about typing or the keyboard. Four phases now, each a slice of the
+ * same [FrameStats] recording ([FrameStats.markPhase]/[FrameStats.phaseLines] -- one recorder, not
+ * two): **fling** (real `FlingBehavior`, not `animateScrollBy`), **stream** (unchanged from v1),
+ * **type** (600 fixed characters into the real composer `TextFieldValue`, then deleted), and
+ * **keyboard** (five show/hide cycles). The exact constants below are also written into
+ * `docs/RUST.md`'s P0 box, "Benchmark v2 (2026-09-06)", so the iris half implements the identical
+ * spec -- changing a number here without updating that box makes the two apps measure different
+ * things while looking like the same benchmark.
+ */
+object BenchRun {
+    /** transcript-bench.sh's default: 6 cycles of 4 swipes each, kept as the pre-v2 comparison. */
+    private const val CYCLES = 6
+    private const val SWIPE_PX = 900f
+    private const val SWIPE_MS = 200
+    private const val SWIPE_PAUSE_MS = 500L
+
+    /**
+     * Fling phase (v2): a real fling through the list's own [FlingBehavior], not `animateScrollBy`
+     * -- Iris's ask was that it "travel way faster" than the old tween-based swipe, and a tween can
+     * never exceed the distance it is told to cover in the time it is given, while a real fling
+     * decays from an initial velocity the way a finger flick does. 12,000 px/s is roughly a hard,
+     * fast flick on a ~420dp/in device (about 30 dp/ms-equivalent initial speed); chosen well above
+     * the ~4,500 px/s a moderate `animateScrollBy` swipe implies, so this phase exercises the fast
+     * end of what the platform's fling decay produces rather than the gentle one v1 measured.
+     */
+    private const val FLING_VELOCITY_PX_S = 12_000f
+
+    private const val FLING_COUNT = 8
+    private const val FLING_SETTLE_CAP_MS = 3_000L
+    private const val FLING_PAUSE_MS = 300L
+
+    /** stream-bench.sh's shape: a real reply arrives as many small deltas, not one big write. */
+    private const val STREAM_EVENTS_PER_SEC = 20
+    private const val STREAM_SECONDS = 20
+
+    /**
+     * Type phase (v2): sentences built from long, multisyllabic words so the composer actually
+     * wraps across lines rather than fitting one, and long enough (600 chars) that the composer's
+     * own height grows over several frames, pushing the transcript above it upward the same way a
+     * real long message does. Exactly this string is also in `docs/RUST.md`'s P0 box so the iris
+     * half types the identical content.
+     */
+    const val TYPE_TEXT =
+        "Benchmarking this transcript screen requires unusually long, multisyllabic words so " +
+            "wrapping and reflow are properly exercised: internationalization, " +
+            "counterproductiveness, disproportionately, incomprehensibility, " +
+            "deinstitutionalization, uncharacteristically, overenthusiastically, " +
+            "misunderstanding, straightforwardness, telecommunications, and interdisciplinary " +
+            "collaboration all push a narrow composer field to wrap across several lines while " +
+            "the transcript above is pushed upward by the growing keyboard-adjacent box, which " +
+            "is exactly what a real reader typing a long message sees happening now!!!"
+
+    private const val TYPE_CHAR_DELAY_MS = 50L
+
+    /**
+     * Keyboard phase (v2): five show/hide cycles, a second apart, is enough to see whether the
+     * transition is ever actually observed rather than being a one-off fluke either way.
+     */
+    private const val KEYBOARD_CYCLES = 5
+    private const val KEYBOARD_SHOW_WAIT_MS = 1_000L
+    private const val KEYBOARD_HIDE_WAIT_MS = 1_000L
+
+    /**
+     * Scrolls, flings, streams, types and toggles the keyboard, then returns the extra report lines
+     * P0 asked for (per-phase travel/typing/keyboard counts, plus CPU time, peak RSS, battery
+     * current) -- [FrameStats] and [DebugStats] are reset first, exactly as `copyRenderReport`
+     * resets them, so the two accountings cover the same stretch of work.
+     */
+    suspend fun run(
+        context: Context,
+        scope: CoroutineScope,
+        listState: LazyListState,
+        flingBehavior: FlingBehavior,
+        composerFocus: FocusRequester,
+        setComposerText: (String) -> Unit,
+        view: View,
+    ): List<String> {
+        FrameStats.reset()
+        DebugStats.reset()
+        val cpuStartMs = Process.getElapsedCpuTime()
+
+        val battery = BatterySampler(context)
+        // Launched in the caller's scope rather than a fresh coroutineScope{} here, which would
+        // suspend this function until the sampler job ended -- and it only ends when told to.
+        val samplerJob = scope.launch {
+            while (isActive) {
+                battery.sample()
+                delay(1000)
+            }
+        }
+
+        val travel = runFlingPhase(listState, flingBehavior)
+        val sent = runStreamPhase()
+        runTypePhase(listState, composerFocus, setComposerText, view)
+        val keyboard = runKeyboardPhase(context, view)
+
+        samplerJob.cancel()
+        val cpuMs = Process.getElapsedCpuTime() - cpuStartMs
+        val rssLine = peakRssLine()
+        val batteryLine = battery.finish()
+
+        return listOf(
+            "  fling: $FLING_COUNT flings out + $FLING_COUNT back at" +
+                " ${FLING_VELOCITY_PX_S.toInt()}px/s, travel $travel",
+            "  scroll: $CYCLES cycles (${CYCLES * 4} swipes, legacy tween), " +
+                "streamed $sent/${STREAM_EVENTS_PER_SEC * STREAM_SECONDS} fixture events",
+            "  type: ${TYPE_TEXT.length} characters inserted then deleted, one per" +
+                " ${TYPE_CHAR_DELAY_MS}ms",
+            keyboard,
+            "  process CPU time over this run: ${cpuMs}ms",
+            rssLine,
+            batteryLine,
+        )
+    }
+
+    /**
+     * Phase 1: starting pinned at the newest end, [FLING_COUNT] flings away from it (toward older
+     * messages) through the list's real fling path, then [FLING_COUNT] back. Positive velocity here
+     * matches this list's existing scroll-offset convention (`TranscriptList`'s `reverseLayout`
+     * pins index 0 -- the newest item -- at the bottom; a positive scroll offset moves the viewport
+     * toward higher indices, i.e. away from the newest end and toward older content), the same sign
+     * the pre-v2 swipe loop below already used for its first two swipes.
+     */
+    private suspend fun runFlingPhase(
+        listState: LazyListState,
+        flingBehavior: FlingBehavior,
+    ): String {
+        FrameStats.markPhase("fling")
+        listState.scrollToItem(0)
+        val start = position(listState)
+        repeat(FLING_COUNT) {
+            listState.scroll { with(flingBehavior) { performFling(FLING_VELOCITY_PX_S) } }
+            waitForSettle(listState)
+            delay(FLING_PAUSE_MS)
+        }
+        val outward = position(listState)
+        repeat(FLING_COUNT) {
+            listState.scroll { with(flingBehavior) { performFling(-FLING_VELOCITY_PX_S) } }
+            waitForSettle(listState)
+            delay(FLING_PAUSE_MS)
+        }
+        val back = position(listState)
+        return "start=$start outward=$outward end=$back"
+    }
+
+    private fun position(listState: LazyListState) =
+        "idx=${listState.firstVisibleItemIndex}/off=${listState.firstVisibleItemScrollOffset}px"
+
+    /** Belt-and-suspenders on top of `performFling` already suspending until its own decay ends. */
+    private suspend fun waitForSettle(listState: LazyListState) {
+        val startedAt = System.currentTimeMillis()
+        while (
+            listState.isScrollInProgress &&
+                System.currentTimeMillis() - startedAt < FLING_SETTLE_CAP_MS
+        ) {
+            delay(16)
+        }
+    }
+
+    /**
+     * Phase 2 (unchanged from v1): pinned to the newest end before streaming starts, the way
+     * stream-bench.sh's "Jump to latest" tap is -- a reply streamed into a list parked further back
+     * arrives off-screen and the report would show nothing happened.
+     */
+    private suspend fun runStreamPhase(): Int {
+        FrameStats.markPhase("stream")
+        var sent = 0
+        val total = STREAM_EVENTS_PER_SEC * STREAM_SECONDS
+        while (sent < total && BenchFixture.remainingStreamEvents() > 0) {
+            BenchFixture.pushNextLiveEvent()
+            sent++
+            delay(1000L / STREAM_EVENTS_PER_SEC)
+        }
+        // Lets the last few deltas land and draw before the next phase starts.
+        delay(300)
+        return sent
+    }
+
+    /**
+     * Phase 3: focuses the real composer, shows the keyboard if the platform allows it, then types
+     * [TYPE_TEXT] one character at a time through the same `TextFieldValue` state a real keystroke
+     * updates, and deletes it the same way -- this is what exercises wrapping and the transcript
+     * being pushed upward, not a single big write.
+     */
+    private suspend fun runTypePhase(
+        listState: LazyListState,
+        composerFocus: FocusRequester,
+        setComposerText: (String) -> Unit,
+        view: View,
+    ) {
+        FrameStats.markPhase("type")
+        listState.scrollToItem(0)
+        composerFocus.requestFocus()
+        showIme(view.context, view)
+        // Lets focus and the keyboard's opening animation land before typing starts, so the frames
+        // this phase records are the wrap/reflow it is measuring, not the keyboard opening.
+        delay(300)
+        var typed = ""
+        for (ch in TYPE_TEXT) {
+            typed += ch
+            setComposerText(typed)
+            delay(TYPE_CHAR_DELAY_MS)
+        }
+        delay(200)
+        while (typed.isNotEmpty()) {
+            typed = typed.dropLast(1)
+            setComposerText(typed)
+            delay(TYPE_CHAR_DELAY_MS)
+        }
+    }
+
+    /**
+     * Phase 4: [KEYBOARD_CYCLES] show/hide cycles through the same [WindowInsetsControllerCompat]
+     * path a real IME toggle goes through, reporting how many of each were actually confirmed by
+     * [android.view.WindowInsets.isVisible] rather than assumed from having asked -- UI_RULES:
+     * never present an inferred value as a measured one. If the platform never shows it even once,
+     * this says so in words rather than reporting a phase with no keyboard in it.
+     */
+    private suspend fun runKeyboardPhase(context: Context, view: View): String {
+        FrameStats.markPhase("keyboard")
+        var shown = 0
+        var hidden = 0
+        repeat(KEYBOARD_CYCLES) {
+            showIme(context, view)
+            delay(KEYBOARD_SHOW_WAIT_MS)
+            if (imeVisible(view)) shown++
+            hideIme(context, view)
+            delay(KEYBOARD_HIDE_WAIT_MS)
+            if (!imeVisible(view)) hidden++
+        }
+        return if (shown == 0) {
+            "  keyboard: could not be shown ($KEYBOARD_CYCLES attempts, 0 confirmed visible)"
+        } else {
+            "  keyboard: shown $shown/$KEYBOARD_CYCLES, hidden $hidden/$KEYBOARD_CYCLES" +
+                " (confirmed via isImeVisible)"
+        }
+    }
+
+    private fun controller(context: Context, view: View): WindowInsetsControllerCompat? {
+        val window = context.activity()?.window ?: return null
+        return WindowInsetsControllerCompat(window, view)
+    }
+
+    private fun showIme(context: Context, view: View) {
+        controller(context, view)?.show(WindowInsetsCompat.Type.ime())
+    }
+
+    private fun hideIme(context: Context, view: View) {
+        controller(context, view)?.hide(WindowInsetsCompat.Type.ime())
+    }
+
+    private fun imeVisible(view: View): Boolean =
+        ViewCompat.getRootWindowInsets(view)?.isVisible(WindowInsetsCompat.Type.ime()) ?: false
+
+    /** VmHWM from /proc/self/status: the process's high-water mark, in kB, since it started. */
+    private fun peakRssLine(): String {
+        val kb =
+            try {
+                File("/proc/self/status")
+                    .readLines()
+                    .firstOrNull { it.startsWith("VmHWM:") }
+                    ?.trim()
+                    ?.removePrefix("VmHWM:")
+                    ?.trim()
+                    ?.removeSuffix("kB")
+                    ?.trim()
+                    ?.toLongOrNull()
+            } catch (_: Exception) {
+                null
+            }
+        return "  peak RSS: " +
+            (kb?.let { "${it}kB" } ?: "unavailable (/proc/self/status unreadable)")
+    }
+}
+
+/**
+ * Samples [BatteryManager.BATTERY_PROPERTY_CURRENT_NOW] (microamps) once a second for the length of
+ * a run. The property returns `Int.MIN_VALUE` on hardware that does not support it -- most
+ * emulators -- and that is reported as "unavailable" rather than folded into an average with the
+ * real samples, which would silently understate every number after it. See UI_RULES: never present
+ * an inferred value as a measured one.
+ */
+private class BatterySampler(context: Context) {
+    private val manager = context.getSystemService(BatteryManager::class.java)
+    private val samples = mutableListOf<Int>()
+
+    fun sample() {
+        val value = manager?.getIntProperty(BatteryManager.BATTERY_PROPERTY_CURRENT_NOW)
+        if (value != null && value != Int.MIN_VALUE) samples.add(value)
+    }
+
+    fun finish(): String {
+        if (samples.isEmpty()) return "  battery current: unavailable on this device"
+        val meanUa = samples.sum() / samples.size
+        return "  battery current: mean ${meanUa}µA over ${samples.size} samples" +
+            " (min ${samples.min()}, max ${samples.max()})"
+    }
+}

app/androidApp/src/main/kotlin/com/example/aiapp/DebugStats.kt

@@ -127,6 +127,20 @@ fun debugReport(
     frames: List<String>,
     accounting: List<String>,
     crash: String?,
+    /**
+     * P0's benchmark-only measurements (process CPU time, peak RSS, battery current) -- empty on
+     * every path but [BenchRun.runP0Benchmark], which is the only caller that has them. A section
+     * heading only appears when there is something to put under it, so an ordinary copy from the
+     * render-report button reads exactly as it did before this existed.
+     */
+    extra: List<String> = emptyList(),
+    /**
+     * Bench v2's per-phase frame accounting ([FrameStats.phaseLines]) --
+     * fling/stream/type/keyboard, each a slice of the same frames the whole-run sections below
+     * still cover in full. Empty on every path but the scripted bench run, same reasoning as
+     * [extra].
+     */
+    phaseFrames: List<String> = emptyList(),
 ): String = buildString {
     appendLine("ai-app render report")
     appendLine(device)
@@ -141,6 +155,11 @@ fun debugReport(
     appendLine("transcript:")
     transcript.forEach { appendLine(it) }
     appendLine()
+    if (phaseFrames.isNotEmpty()) {
+        appendLine("per phase:")
+        phaseFrames.forEach { appendLine(it) }
+        appendLine()
+    }
     appendLine("frames:")
     frames.forEach { appendLine(it) }
     appendLine()
@@ -152,6 +171,11 @@ fun debugReport(
     appendLine("work since this was last copied:")
     val work = DebugStats.lines()
     if (work.isEmpty()) appendLine("  nothing recorded") else work.forEach { appendLine(it) }
+    if (extra.isNotEmpty()) {
+        appendLine()
+        appendLine("bench:")
+        extra.forEach { appendLine(it) }
+    }
 }
 
 /** Puts [text] on the clipboard under [label], which is what the system offers as its name. */

app/androidApp/src/main/kotlin/com/example/aiapp/Dividers.kt

@@ -12,6 +12,10 @@ import androidx.compose.ui.Alignment
 import androidx.compose.ui.Modifier
 import androidx.compose.ui.graphics.Color
 import androidx.compose.ui.unit.dp
+import java.time.Instant
+import java.time.ZoneId
+import java.time.format.DateTimeFormatter
+import java.time.format.FormatStyle
 
 /**
  * A line across the transcript saying what left the session's context.
@@ -47,3 +51,40 @@ fun TranscriptDivider(text: String, color: Color, modifier: Modifier = Modifier)
 fun ClearedRow(modifier: Modifier = Modifier) {
     TranscriptDivider("Context cleared", clearedColor, modifier)
 }
+
+/**
+ * The mark running out of quota leaves.
+ *
+ * The same red the usage bar takes when a window is spent, because it is the same fact in a second
+ * place: colour by consequence, so "there is nothing left to spend" is learned once.
+ *
+ * A time rather than a countdown. The row is folded once and never re-measured, so a span would go
+ * stale on screen the moment it was drawn; and this is when the *account* said it would reset,
+ * which is not a promise about when the session picks back up. A limit the session was told no
+ * reset time for says nothing about one -- that state has its own words rather than a plausible
+ * number.
+ */
+@Composable
+fun LimitRow(item: TranscriptItem.LimitNote, modifier: Modifier = Modifier) {
+    TranscriptDivider(limitSummary(item.resetsAt, ZoneId.systemDefault()), overLimitColor, modifier)
+}
+
+/**
+ * What the row says. Split out so the wording is testable without a screen, since the two states it
+ * has to keep apart -- a reset time that arrived and one that never did -- are exactly the pair
+ * that reads the same when it goes wrong.
+ *
+ * [zone] is a parameter rather than read here so a test says the same thing wherever it runs.
+ */
+fun limitSummary(resetsAt: Double?, zone: ZoneId): String {
+    val at = resetsAt?.let {
+        try {
+            DateTimeFormatter.ofLocalizedTime(FormatStyle.SHORT)
+                .withZone(zone)
+                .format(Instant.ofEpochSecond(it.toLong()))
+        } catch (_: Exception) {
+            null
+        }
+    }
+    return if (at == null) "Usage limit reached" else "Usage limit reached • resets $at"
+}

app/androidApp/src/main/kotlin/com/example/aiapp/EventStream.kt

@@ -14,7 +14,7 @@ private const val RESET_EVENT = "reset"
  * mean. [close] from any thread ends it, and the caller owns reconnecting -- with the last seq it
  * saw as the new cursor.
  */
-class EventStream(settings: ServerSettings, private val sessionId: String) {
+class EventStream(settings: ServerSettings, private val address: TranscriptAddress) {
     private val stream = Sse(settings)
 
     fun close() = stream.close()
@@ -35,7 +35,7 @@ class EventStream(settings: ServerSettings, private val sessionId: String) {
         // one and the screen folds the other, and they have to be the same line.
         onEvent: (raw: String, event: SeqEvent) -> Unit,
     ) {
-        stream.run("/sessions/$sessionId/events?after=$after", onOpen) { name, data ->
+        stream.run("/${address.urlPath}/events?after=$after", onOpen) { name, data ->
             // A named frame carries no payload and a data frame has no name.
             if (name == RESET_EVENT) onReset()
             else if (data.isNotEmpty()) onEvent(data, parseSeqEvent(data))

app/androidApp/src/main/kotlin/com/example/aiapp/Events.kt

@@ -157,6 +157,18 @@ sealed class SessionEvent {
      */
     data object Cleared : SessionEvent()
 
+    /**
+     * The session stopped because its account's usage limit was reached.
+     *
+     * Its own event rather than an [Error] carrying the CLI's sentence, because it is a state
+     * rather than something that went wrong -- and because the raw sentence is `Claude AI usage
+     * limit reached|1788546972`, which is not readable by the person it is shown to.
+     *
+     * [resetsAt] is epoch seconds and null where the session was told nothing. Only the server acts
+     * on it; what this draws it as is a time, not a countdown, because nothing here re-measures it.
+     */
+    data class LimitReached(val resetsAt: Double?) : SessionEvent()
+
     data class Error(val message: String) : SessionEvent()
 
     /**
@@ -261,6 +273,10 @@ fun parseSeqEvent(json: String): SeqEvent {
                     trigger = body.optString("trigger").ifEmpty { null },
                 )
             "cleared" -> SessionEvent.Cleared
+            "limitReached" ->
+                SessionEvent.LimitReached(
+                    if (body.has("resetsAt")) body.getDouble("resetsAt") else null
+                )
             "error" -> SessionEvent.Error(body.getString("message"))
             else -> SessionEvent.Unknown(type)
         }

app/androidApp/src/main/kotlin/com/example/aiapp/FrameStats.kt

@@ -42,6 +42,21 @@ object FrameStats {
     private val gpu = ArrayList<Long>()
     private var since = System.currentTimeMillis()
 
+    /**
+     * Where a named phase of a scripted run (bench v2's fling/stream/type/keyboard) started, as an
+     * index into [total] and a wall-clock time -- not a second recorder, just a mark on this one,
+     * so a phase's frames are the same [FrameMetrics] the whole-run report already has, sliced.
+     */
+    private data class PhaseMark(val name: String, val startIndex: Int, val startMs: Long)
+
+    private val phaseMarks = ArrayList<PhaseMark>()
+
+    /** Call at the start of each named phase of a scripted run; see [BenchRun]. */
+    @Synchronized
+    fun markPhase(name: String) {
+        phaseMarks += PhaseMark(name, total.size, System.currentTimeMillis())
+    }
+
     @Synchronized
     fun add(metrics: FrameMetrics) {
         // The first frame after a window opens includes inflating it and is nobody's scroll.
@@ -69,6 +84,7 @@ object FrameStats {
         listOf(total, waited, input, animation, layout, draw, sync, issue, swap, gpu).forEach {
             it.clear()
         }
+        phaseMarks.clear()
         since = System.currentTimeMillis()
     }
 
@@ -95,6 +111,38 @@ object FrameStats {
         ) + if (gpu.isEmpty()) emptyList() else listOf(phase("gpu   ", gpu))
     }
 
+    /**
+     * One block per [markPhase] call: how many frames landed between that mark and the next (or the
+     * end of the run, for the last one), how many were late, the total/p50/p90/p99, the worst
+     * single frame, and how long the phase actually ran. Marks with no frames between them (a phase
+     * that finished before a frame was drawn) still get a line rather than being silently dropped
+     * -- UI_RULES' "say what you don't know" applies to a phase as much as to a single number.
+     */
+    @Synchronized
+    fun phaseLines(refreshHz: Float): List<String> {
+        if (phaseMarks.isEmpty()) return emptyList()
+        val budget = if (refreshHz > 0) 1000.0 / refreshHz else 16.7
+        val lines = ArrayList<String>()
+        phaseMarks.forEachIndexed { i, mark ->
+            val endIndex = if (i + 1 < phaseMarks.size) phaseMarks[i + 1].startIndex else total.size
+            val endMs =
+                if (i + 1 < phaseMarks.size) phaseMarks[i + 1].startMs
+                else System.currentTimeMillis()
+            val samples = total.subList(mark.startIndex, endIndex)
+            val seconds = (endMs - mark.startMs) / 1000.0
+            lines += "  ${mark.name}: ${samples.size} frames over ${"%.1f".format(seconds)}s"
+            if (samples.isEmpty()) {
+                lines += "    no frames recorded in this phase"
+            } else {
+                val late = samples.count { it / 1_000_000.0 > budget }
+                lines += "    late: $late (${percent(late, samples.size)})"
+                lines += "    " + phase("total ", samples)
+                lines += "    worst ${"%.1fms".format(samples.max() / 1_000_000.0)}"
+            }
+        }
+        return lines
+    }
+
     /** How long the frames recorded here spent in their draw phase, and how many there were. */
     @Synchronized fun drawPhase(): Pair<Long, Int> = draw.sum() to draw.size
 

app/androidApp/src/main/kotlin/com/example/aiapp/MainActivity.kt

@@ -66,6 +66,35 @@ class MainActivity : ComponentActivity() {
         // Transparent status bar on every version; the Surface below paints through underneath it
         // and content insets itself. Same reasoning as dev-updater's MainActivity.
         enableEdgeToEdge()
+
+        // The `bench` build's entire purpose (P0, docs/RUST.md): open straight onto the session
+        // screen against BenchFixture's in-process fake backend, with no enrollment, no network
+        // permission, and no notification prompt -- none of them mean anything with no server and
+        // no real device to notify. See BenchFixture.kt and BenchNetwork.kt for how a screen built
+        // to talk to a real backend is made to talk to this instead. Still needs the same
+        // status/navigation-bar padding the ordinary flow below applies: edge-to-edge is the
+        // platform's own default from Android 15 on this app's targetSdk, with or without the call
+        // above, so skipping the padding here put the header's own buttons under the status bar --
+        // there to look at, but not there for `ui-trace`'s tap-by-label to land on.
+        if (BuildConfig.FIXTURE_MODE) {
+            installFixtureNetworkOnce()
+            BenchFixture.ensureLoaded(this)
+            setContent {
+                MaterialTheme(colorScheme = AiAppColors) {
+                    Surface(modifier = Modifier.fillMaxSize()) {
+                        Box(Modifier.fillMaxSize().statusBarsPadding().navigationBarsPadding()) {
+                            SessionScreen(
+                                settings = BenchFixture.settings,
+                                summary = benchSessionSummary(),
+                                onBack = { finish() },
+                                onFiles = {},
+                            )
+                        }
+                    }
+                }
+            }
+            return
+        }
         // Dark status-bar icons only over a light background, decided from the scheme rather than
         // fixed. It was hardcoded to `true`, which was right against the default light surface and
         // became unreadable the moment the app wore Catppuccin Mocha.
@@ -147,6 +176,33 @@ class MainActivity : ComponentActivity() {
         }
     }
 
+    /** The one session the `bench` build ever shows -- BenchFixture's session id, nothing else. */
+    private fun benchSessionSummary() =
+        SessionSummary(
+            id = BenchFixture.SESSION_ID,
+            setup = "bench",
+            setupName = "bench",
+            provider = "bench",
+            title = "P0 benchmark",
+            model = null,
+            keepsOwnTranscript = false,
+            permissionMode = null,
+            effort = null,
+            takesEffort = false,
+            imported = false,
+            notify = false,
+            autoResume = false,
+            autoResumeMessage = "",
+            resumeAt = null,
+            cwd = null,
+            contextTokens = null,
+            maxImageEdge = null,
+            usageProvider = null,
+            status = "idle",
+            lastActivity = 0.0,
+            subagents = 0,
+        )
+
     // launchMode="singleTop": an enrollment scan, or a notification tapped while the app is open,
     // lands here rather than in a second activity instance.
     override fun onNewIntent(intent: Intent) {

app/androidApp/src/main/kotlin/com/example/aiapp/MainScreen.kt

@@ -48,6 +48,8 @@ fun MainScreen(
     /** What another app shared in and no session has taken yet; see [ShareRequest]. */
     share: ShareRequest? = null,
     onOpen: (SessionSummary) -> Unit,
+    /** Opens one session's subagent, from the expander under its card. */
+    onOpenSubagent: (SessionSummary, SubagentSummary) -> Unit,
     onSpawn: () -> Unit,
     onImported: (SessionSummary) -> Unit,
     onSettings: () -> Unit,
@@ -139,6 +141,7 @@ fun MainScreen(
                     settings = settings,
                     reloadToken = token,
                     onOpen = onOpen,
+                    onOpenSubagent = onOpenSubagent,
                     onSpawn = onSpawn,
                 )
             MainTab.Import ->

app/androidApp/src/main/kotlin/com/example/aiapp/SessionListScreen.kt

@@ -1,7 +1,9 @@
 package com.example.aiapp
 
 import androidx.compose.foundation.ExperimentalFoundationApi
+import androidx.compose.foundation.clickable
 import androidx.compose.foundation.combinedClickable
+import androidx.compose.foundation.layout.Arrangement
 import androidx.compose.foundation.layout.Box
 import androidx.compose.foundation.layout.Column
 import androidx.compose.foundation.layout.Row
@@ -9,6 +11,7 @@ import androidx.compose.foundation.layout.Spacer
 import androidx.compose.foundation.layout.fillMaxSize
 import androidx.compose.foundation.layout.fillMaxWidth
 import androidx.compose.foundation.layout.height
+import androidx.compose.foundation.layout.heightIn
 import androidx.compose.foundation.layout.padding
 import androidx.compose.foundation.layout.width
 import androidx.compose.foundation.lazy.LazyColumn
@@ -17,6 +20,7 @@ import androidx.compose.material3.Card
 import androidx.compose.material3.CircularProgressIndicator
 import androidx.compose.material3.FloatingActionButton
 import androidx.compose.material3.MaterialTheme
+import androidx.compose.material3.OutlinedCard
 import androidx.compose.material3.Switch
 import androidx.compose.material3.Text
 import androidx.compose.material3.TextButton
@@ -30,6 +34,8 @@ import androidx.compose.runtime.setValue
 import androidx.compose.ui.Alignment
 import androidx.compose.ui.Modifier
 import androidx.compose.ui.platform.LocalContext
+import androidx.compose.ui.semantics.contentDescription
+import androidx.compose.ui.semantics.semantics
 import androidx.compose.ui.unit.dp
 import kotlinx.coroutines.Dispatchers
 import kotlinx.coroutines.launch
@@ -47,12 +53,40 @@ fun SessionListScreen(
     settings: ServerSettings,
     reloadToken: Int,
     onOpen: (SessionSummary) -> Unit,
+    /** Opens one session's subagent, from the expander under its card. */
+    onOpenSubagent: (SessionSummary, SubagentSummary) -> Unit,
     onSpawn: () -> Unit,
 ) {
     val scope = rememberCoroutineScope()
     var listState by remember { mutableStateOf<LoadState<List<SessionSummary>>>(LoadState.Loading) }
     var confirmingDelete by remember { mutableStateOf<SessionSummary?>(null) }
 
+    // Which session cards are expanded to show their subagents, and what each expansion fetched.
+    // Ids rather than a flag on the row for the same reason `deleting` is: the rows are rebuilt
+    // from
+    // whatever the server last said, and this belongs to the reader's own choice, which survives a
+    // refresh.
+    var expandedSessions by remember { mutableStateOf(setOf<String>()) }
+    var subagentLoads by remember {
+        mutableStateOf(mapOf<String, LoadState<List<SubagentSummary>>>())
+    }
+
+    fun loadSubagents(sessionId: String) {
+        subagentLoads = subagentLoads + (sessionId to LoadState.Loading)
+        scope.launch {
+            subagentLoads =
+                subagentLoads +
+                    (sessionId to
+                        try {
+                            LoadState.Loaded(
+                                withContext(Dispatchers.IO) { fetchSubagents(settings, sessionId) }
+                            )
+                        } catch (e: ApiException) {
+                            LoadState.failed(e)
+                        })
+        }
+    }
+
     // Failures that belong to one session rather than to the list, keyed by its id and shown on its
     // own card. The two scopes are decided by whether the server answered: it answered and refused,
     // so this says nothing about the other rows.
@@ -84,6 +118,14 @@ fun SessionListScreen(
                     withContext(Dispatchers.IO) {
                         transcriptCache.retainOnly(loaded.value.map { it.id }.toSet())
                     }
+                    // A session gone from this answer cannot still be expanded, and an expanded one
+                    // that is still here asks again -- its subagents may have changed since the
+                    // last
+                    // fetch.
+                    val ids = loaded.value.map { it.id }.toSet()
+                    expandedSessions = expandedSessions intersect ids
+                    subagentLoads = subagentLoads.filterKeys { it in ids }
+                    expandedSessions.forEach(::loadSubagents)
                     loaded
                 } catch (e: ApiException) {
                     LoadState.failed(e)
@@ -127,6 +169,17 @@ fun SessionListScreen(
                                 deleting = session.id in deleting,
                                 onOpen = { onOpen(session) },
                                 onLongPress = { confirmingDelete = session },
+                                expanded = session.id in expandedSessions,
+                                subagents = subagentLoads[session.id],
+                                onToggleSubagents = {
+                                    if (session.id in expandedSessions) {
+                                        expandedSessions = expandedSessions - session.id
+                                    } else {
+                                        expandedSessions = expandedSessions + session.id
+                                        loadSubagents(session.id)
+                                    }
+                                },
+                                onOpenSubagent = { subagent -> onOpenSubagent(session, subagent) },
                             )
                             Spacer(Modifier.height(12.dp))
                         }
@@ -225,7 +278,7 @@ fun SessionListScreen(
                                     deleteSession(settings, session.id, alsoDeleteForeign)
                                     // After it succeeded, not before: a refused delete leaves the
                                     // session exactly as it was, and its transcript with it.
-                                    transcriptCache.session(session.id).purge()
+                                    transcriptCache.session(TranscriptAddress(session.id)).purge()
                                 }
                                 // Only this row, and only what changed. Refetching the list instead
                                 // put every other session back through loading and handed the
@@ -276,6 +329,12 @@ private fun SessionCard(
     deleting: Boolean,
     onOpen: () -> Unit,
     onLongPress: () -> Unit,
+    /** Whether the expander below is open. Collapsed by default; see [SessionListScreen]. */
+    expanded: Boolean,
+    /** What the expander's own fetch answered, or null before it has been asked. */
+    subagents: LoadState<List<SubagentSummary>>?,
+    onToggleSubagents: () -> Unit,
+    onOpenSubagent: (SubagentSummary) -> Unit,
 ) {
     BusyItem(label = if (deleting) "deleting" else null) {
         Card(
@@ -332,10 +391,104 @@ private fun SessionCard(
                         color = MaterialTheme.colorScheme.error,
                     )
                 }
+                // Nothing at all for a card with no subagents: a disabled expander here would be
+                // noise on every ordinary session's card. Its own row at the bottom rather than
+                // beside the title or the machine line, so opening it never displaces text that was
+                // already on screen -- see UI_RULES on a control not displacing the text beside it.
+                if (session.subagents > 0) {
+                    Spacer(Modifier.height(8.dp))
+                    // The platform's minimum touch height, not the chevron's own ten or so dp:
+                    // at the chevron's height a tap meant for it landed on the first subcard
+                    // beneath and opened a subagent instead.
+                    Row(
+                        horizontalArrangement = Arrangement.Center,
+                        verticalAlignment = Alignment.CenterVertically,
+                        modifier =
+                            Modifier.fillMaxWidth()
+                                .heightIn(min = 48.dp)
+                                .clickable(enabled = !deleting, onClick = onToggleSubagents)
+                                .semantics {
+                                    contentDescription =
+                                        if (expanded) "Collapse subagents" else "Expand subagents"
+                                },
+                    ) {
+                        Chevron(if (expanded) Pointing.Up else Pointing.Down)
+                    }
+                    if (expanded) {
+                        Spacer(Modifier.height(4.dp))
+                        Column(verticalArrangement = Arrangement.spacedBy(8.dp)) {
+                            when (subagents) {
+                                null,
+                                is LoadState.Loading ->
+                                    CircularProgressIndicator(
+                                        modifier = Modifier.width(20.dp).height(20.dp),
+                                        strokeWidth = 2.dp,
+                                    )
+                                is LoadState.Error ->
+                                    // Said here rather than left silent: a fetch that failed and an
+                                    // expander that simply found nothing must not look the same --
+                                    // see UI_RULES on designing the unknown state first.
+                                    Text(
+                                        subagents.message,
+                                        style = MaterialTheme.typography.bodySmall,
+                                        color = MaterialTheme.colorScheme.error,
+                                    )
+                                is LoadState.Loaded ->
+                                    subagents.value.forEach { subagent ->
+                                        SubagentCard(
+                                            subagent,
+                                            onClick = { onOpenSubagent(subagent) },
+                                        )
                                     }
                             }
                         }
                     }
+                }
+            }
+        }
+    }
+}
+
+/**
+ * One subagent, indented inside its session's card -- the way dev-updater draws a project's
+ * components (`ComponentCard`, `UpdaterScreen.kt`): an outlined card, not the session card's own
+ * filled one, so the nesting reads as one step rather than as another session.
+ */
+@Composable
+private fun SubagentCard(subagent: SubagentSummary, onClick: () -> Unit) {
+    OutlinedCard(Modifier.fillMaxWidth().clickable(onClick = onClick)) {
+        Column(Modifier.padding(horizontal = 12.dp, vertical = 8.dp)) {
+            Text(subagent.title, style = MaterialTheme.typography.titleSmall)
+            Spacer(Modifier.height(2.dp))
+            Row(modifier = Modifier.fillMaxWidth()) {
+                Text(
+                    subagentStatusLabel(subagent.status),
+                    style = MaterialTheme.typography.bodySmall,
+                    color = MaterialTheme.colorScheme.onSurfaceVariant,
+                    modifier = Modifier.weight(1f),
+                )
+                Text(
+                    relativeTime(subagent.lastActivity),
+                    style = MaterialTheme.typography.bodySmall,
+                    color = MaterialTheme.colorScheme.onSurfaceVariant,
+                )
+            }
+        }
+    }
+}
+
+/**
+ * The subcard's word for a subagent's status -- see SUBAGENTS.md's "Wire shape". Its own function
+ * rather than a branch inside [StatusText], because a subagent's three states are not that
+ * composable's five: "exited" reads as "finished" here, since its process was always its parent's
+ * and never something of its own to have merely stopped.
+ */
+private fun subagentStatusLabel(status: String) =
+    when (status) {
+        "running" -> "running"
+        "exited" -> "finished"
+        else -> "unknown"
+    }
 
 @Composable
 fun StatusText(status: String) {

app/androidApp/src/main/kotlin/com/example/aiapp/SessionScreen.kt

@@ -12,6 +12,7 @@ import androidx.activity.result.PickVisualMediaRequest
 import androidx.activity.result.contract.ActivityResultContracts
 import androidx.compose.foundation.background
 import androidx.compose.foundation.clickable
+import androidx.compose.foundation.gestures.ScrollableDefaults
 import androidx.compose.foundation.gestures.awaitEachGesture
 import androidx.compose.foundation.gestures.awaitFirstDown
 import androidx.compose.foundation.layout.Box
@@ -64,12 +65,15 @@ import androidx.compose.runtime.snapshots.Snapshot
 import androidx.compose.ui.Alignment
 import androidx.compose.ui.Modifier
 import androidx.compose.ui.draw.drawWithContent
+import androidx.compose.ui.focus.FocusRequester
+import androidx.compose.ui.focus.focusRequester
 import androidx.compose.ui.graphics.graphicsLayer
 import androidx.compose.ui.input.pointer.PointerEventPass
 import androidx.compose.ui.input.pointer.pointerInput
 import androidx.compose.ui.layout.onSizeChanged
 import androidx.compose.ui.platform.LocalContext
 import androidx.compose.ui.platform.LocalDensity
+import androidx.compose.ui.platform.LocalView
 import androidx.compose.ui.semantics.contentDescription
 import androidx.compose.ui.semantics.semantics
 import androidx.compose.ui.text.TextRange
@@ -216,16 +220,33 @@ fun SessionScreen(
     share: ShareRequest? = null,
     /** Said once [share] has been attached here, so it is not attached again. */
     onShareTaken: () -> Unit = {},
+    /**
+     * Draws this screen read-only, on a subagent's own transcript instead of the session's.
+     *
+     * A subagent has no process and no controls of its own -- see SUBAGENTS.md's "Phone" -- so
+     * every gate below keyed on this switches off the composer, the files button, the settings cog,
+     * the usage bar and notifications, while everything that draws a transcript (paging, cache,
+     * selection, images, the status row, stream reconnects) is reused unchanged, pointed at
+     * [address] instead of the session's own.
+     */
+    subagent: SubagentSummary? = null,
 ) {
     DebugStats.count("session screen recomposed")
+    val isSubagent = subagent != null
+    val address = TranscriptAddress(summary.id, subagent?.id)
     val scope = rememberCoroutineScope()
     val topEdgeHeld = remember { TopEdgeHold() }
     var items by remember { mutableStateOf(listOf<TranscriptItem>()) }
-    var status by remember { mutableStateOf(summary.status) }
+    var status by remember { mutableStateOf(subagent?.status ?: summary.status) }
     // Seeded from the row this screen was opened from, so a conversation already under way says how
     // much it is holding before any turn happens here. Null is "nobody has measured it", which is a
     // different answer from an empty context and is drawn differently.
-    var contextTokens by remember(summary.id) { mutableStateOf(summary.contextTokens) }
+    //
+    // A subagent has no context measurement of its own, so it always starts unmeasured rather than
+    // borrowing the parent session's figure -- see UI_RULES on not showing an inferred value as one
+    // that was measured.
+    var contextTokens by
+        remember(address) { mutableStateOf(if (isSubagent) null else summary.contextTokens) }
     // When the current compaction started. The moment comes off the `compacting` status event
     // itself -- the server timestamps every transcript line -- rather than off this device noticing
     // one, which is what makes it survive leaving the session and reopening it.
@@ -241,7 +262,13 @@ fun SessionScreen(
     val context = LocalContext.current
     // Seeded from what was left in the box last time and written back on every keystroke, so
     // leaving the screen does not throw away a half-typed message. See `Drafts.kt`.
-    var input by remember(summary.id) { mutableStateOf(atEnd(loadDraft(context, summary.id))) }
+    //
+    // A subagent has no box to type into, so it never touches a draft at all -- not this session's,
+    // which is what reading one keyed only by `summary.id` would do here.
+    var input by
+        remember(summary.id) {
+            mutableStateOf(if (isSubagent) atEnd("") else atEnd(loadDraft(context, summary.id)))
+        }
     // A model the reader has chosen and not yet confirmed. See [ModelSwitchWarning]: switching
     // makes the session re-read the whole conversation.
     var pendingModel by remember { mutableStateOf<String?>(null) }
@@ -294,27 +321,26 @@ fun SessionScreen(
     // Reload throws away what it was reading from.
     val cache = remember(settings) { TranscriptCache(cacheRoot(context, settings)) }
     val source =
-        remember(summary.id, epoch) {
-            TranscriptSource(settings, summary.id, cache.session(summary.id))
-        }
+        remember(address, epoch) { TranscriptSource(settings, address, cache.session(address)) }
     // Whether the cached tail has been shown to still be the server's own line. Nothing is resumed
     // from a cached cursor until it has, and a probe that could not be made leaves this false for
     // the stream loop to try again.
-    var probePassed by remember(summary.id, epoch) { mutableStateOf(false) }
+    var probePassed by remember(address, epoch) { mutableStateOf(false) }
     // Whether the opening effect is still settling that question. It draws the cached rows and
     // lifts [ready] before the answer arrives, which is the point of the cache -- so the stream
     // below waits for this rather than for `ready`, or it asks the same question twice.
-    var probing by remember(summary.id, epoch) { mutableStateOf(true) }
+    var probing by remember(address, epoch) { mutableStateOf(true) }
     // The oldest sequence number loaded, and whether there is more behind it. Paging backwards is
     // what keeps opening a long session cheap.
     var oldestSeq by remember { mutableLongStateOf(0L) }
-    // Where this session was last being read, from this device's own store. Read once, because the
-    // answer stops being interesting the moment the list is on screen.
-    val savedAnchor = remember(summary.id, epoch) { loadScrollAnchor(context, summary.id) }
+    // Where this transcript was last being read, from this device's own store, keyed by the address
+    // rather than the session id so a subagent's saved position cannot collide with its session's.
+    // Read once, because the answer stops being interesting the moment the list is on screen.
+    val savedAnchor = remember(address, epoch) { loadScrollAnchor(context, address.cachePath) }
     // Whether the saved position is still being put back. Nothing is drawn while it is: opening at
     // the newest end and then travelling to the anchor is exactly the journey a reader must never
     // see.
-    var restoring by remember(summary.id, epoch) { mutableStateOf(savedAnchor != null) }
+    var restoring by remember(address, epoch) { mutableStateOf(savedAnchor != null) }
     // Messages the server has taken and the session has not read yet, by the id that will resolve
     // them. From the event stream rather than from what this screen sent, so they survive leaving
     // the session -- and a message sent from another device is drawn waiting on this one too.
@@ -327,11 +353,22 @@ fun SessionScreen(
     var loadingHistory by remember { mutableStateOf(false) }
     var ready by remember { mutableStateOf(false) }
     // Replies parsed ahead of the rows that draw them; see [ParsedReplies].
-    val replies = remember(summary.id) { ParsedReplies() }
-    // Keyed like everything else describing one session's transcript. `rememberLazyListState` saves
-    // through `rememberSaveable`, and this screen restores by its own anchor instead -- two
-    // restores would fight over the first frame.
-    val listState = remember(summary.id) { LazyListState() }
+    val replies = remember(address) { ParsedReplies() }
+    // Keyed like everything else describing one transcript. `rememberLazyListState` saves through
+    // `rememberSaveable`, and this screen restores by its own anchor instead -- two restores would
+    // fight over the first frame.
+    val listState = remember(address) { LazyListState() }
+    // The list's own fling path -- what a real flick decays through -- captured here so BenchRun's
+    // fling phase can drive `LazyListState.scroll` through exactly the `FlingBehavior` this
+    // screen's
+    // `TranscriptList` already uses by not overriding it (its `LazyColumn` takes no `flingBehavior`
+    // argument, so this is the same default it gets).
+    val flingBehavior = ScrollableDefaults.flingBehavior()
+    // Where BenchRun's type phase focuses before it types, and the view it toggles the keyboard on
+    // -- both bench-only, but cheap enough (a remembered object, a CompositionLocal read) to hold
+    // unconditionally rather than behind a second code path only the bench build compiles.
+    val composerFocus = remember { FocusRequester() }
+    val view = LocalView.current
     // Whether the newest message is on screen right now. The list is reversed, so the newest end is
     // the scrolling start: nothing behind you is exactly being at the bottom. Asked of the scroll
     // state rather than of item indices, because a zero-height first item makes an index ambiguous.
@@ -637,7 +674,7 @@ fun SessionScreen(
     // ended and carries live events only. The window comes from this phone's own copy when there is
     // one, and then costs a single request to check that the server's transcript is still the one
     // it came from. See TRANSCRIPT_CACHE.md.
-    LaunchedEffect(summary.id, epoch) {
+    LaunchedEffect(address, epoch) {
         /**
          * One opening window onto the screen, whichever side it came from.
          *
@@ -667,11 +704,16 @@ fun SessionScreen(
                 // A replay is as old as the last visit; the row this screen was opened from was
                 // fetched moments ago. So the transcript comes from the cache and everything that
                 // is not the transcript comes from the summary -- otherwise a session that finished
-                // an hour ago opens saying "working" until the stream connects.
-                status = summary.status
+                // an hour ago opens saying "working" until the stream connects. A subagent's status
+                // comes from its own summary, never the parent session's: they are two different
+                // things running or not, and the parent's model and permission mode do not apply to
+                // it at all.
+                status = subagent?.status ?: summary.status
+                if (!isSubagent) {
                     model = summary.model
                     permissionMode = summary.permissionMode ?: "auto"
-                if (summary.status != "compacting") compactingSince = null
+                }
+                if (status != "compacting") compactingSince = null
                 // Nothing to put back, so these rows are the screen and the probe can return under
                 // them. A restore still has history to fetch and is gated below.
                 if (savedAnchor == null) ready = true
@@ -798,7 +840,7 @@ fun SessionScreen(
     // at the top on their return. Switching apps is a choice somebody made, not a fault to report.
     // Stopping the stream deliberately makes the drop a close rather than an error, and resuming
     // reconnects from the same cursor.
-    LaunchedEffect(summary.id, ready, epoch, lifecycleOwner) {
+    LaunchedEffect(address, ready, epoch, lifecycleOwner) {
         if (!ready) return@LaunchedEffect
         // The opening effect draws cached rows and lifts `ready` *before* it has checked that the
         // cursor under them is still the server's, so `ready` is no longer the whole gate. Without
@@ -868,10 +910,14 @@ fun SessionScreen(
     // The screen going away entirely, which the lifecycle scope above does not cover: a composable
     // can leave the composition while the activity stays started. Keyed on the epoch as well, so
     // Reload's replacement source is the one a later disposal closes.
-    DisposableEffect(summary.id, epoch) { onDispose { source.close() } }
+    DisposableEffect(address, epoch) { onDispose { source.close() } }
 
     // Nothing gets announced about the session somebody is reading; see NotificationService.
     // RESUMED rather than STARTED because "looking at it" means the foreground.
+    //
+    // Not for a subagent: it has no notifications of its own, and it is not the session this would
+    // otherwise mark as being read.
+    if (!isSubagent) {
         LaunchedEffect(summary.id, lifecycleOwner) {
             lifecycleOwner.repeatOnLifecycle(Lifecycle.State.RESUMED) {
                 NotificationService.showing(context, summary.id)
@@ -882,6 +928,7 @@ fun SessionScreen(
                 }
             }
         }
+    }
 
     // Back at the newest end, so the backlog [apply] held can land. Everything at once rather than
     // paced out: they are at the bottom, which is the one place the list is allowed to follow new
@@ -924,7 +971,7 @@ fun SessionScreen(
                 val (index, offset, awayFromNewest) = settled
                 saveScrollAnchor(
                     context,
-                    summary.id,
+                    address.cachePath,
                     // Nothing to restore at the newest end, which is where a session with no anchor
                     // opens anyway. One *before* the index, because item zero is the "below" slot.
                     if (!awayFromNewest) null
@@ -947,7 +994,7 @@ fun SessionScreen(
     //
     // There is no correction beside this one. Following the newest message is not an effect: the
     // list is reversed, so an arriving message extends the end the viewport is pinned to.
-    val unitSizes = remember(summary.id) { HashMap<Any, Int>() }
+    val unitSizes = remember(address) { HashMap<Any, Int>() }
     LaunchedEffect(listState, moreHistory) {
         snapshotFlow { listState.layoutInfo }
             .collect { info ->
@@ -983,6 +1030,8 @@ fun SessionScreen(
             }
     }
 
+    // Only for the model picker, which a subagent does not have.
+    if (!isSubagent) {
         LaunchedEffect(summary.setupName, summary.provider) {
             offeredModels =
                 try {
@@ -995,10 +1044,12 @@ fun SessionScreen(
                             .orEmpty()
                     }
                 } catch (_: Exception) {
-                // Not worth reporting: the picker simply has nothing to offer, which is visible.
+                    // Not worth reporting: the picker simply has nothing to offer, which is
+                    // visible.
                     emptyList()
                 }
         }
+    }
 
     /**
      * Asks the server to take back a message the session has not read yet.
@@ -1148,8 +1199,9 @@ fun SessionScreen(
     }
 
     // One poll for the machines' limits, read by everything on this screen that reports them.
-    val usageFeed = rememberUsageFeed(settings)
-    val usage = usageFeed.forSetup(summary.setup)
+    // Nothing meters a subagent -- it has no account of its own -- so it never starts this poll.
+    val usageFeed = if (isSubagent) null else rememberUsageFeed(settings)
+    val usage = usageFeed?.forSession(summary) ?: SessionUsage.NotMetered
     RecordFrames()
     var usageOpen by remember { mutableStateOf(false) }
     var settingsOpen by remember { mutableStateOf(false) }
@@ -1191,7 +1243,12 @@ fun SessionScreen(
     // the bench scripts keep working when this moves again. They pressed it at a hand-measured
     // coordinate until 2026-09-03, and anything that moved the header made that tap land on
     // whatever now sat there -- reporting a number that was never measured.
-    val copyRenderReport = {
+    // Shared by the ordinary "Copy" button and (bench build only) "Run benchmark": what differs
+    // between them is only whether there is a [extra] section, built by BenchRun.run beforehand --
+    // everything about assembling, copying and logging the report is exactly the same act either
+    // way, and a second copy of it beside `onRunBenchmark` below would be the two silently
+    // disagreeing about what "the report" contains the first time either one changed.
+    fun buildAndCopyReport(extra: List<String> = emptyList()) {
         val report =
             debugReport(
                 device =
@@ -1213,6 +1270,11 @@ fun SessionScreen(
                 accounting =
                     FrameStats.drawPhase().let { (nanos, count) -> drawAccounting(nanos, count) },
                 crash = lastCrash(context),
+                extra = extra,
+                // Empty outside a BenchRun.run pass -- copyRenderReport's own reset below clears
+                // the
+                // marks along with everything else, so an ordinary copy never has any to show.
+                phaseFrames = FrameStats.phaseLines(context.refreshHz()),
             )
         context.copyToClipboard("ai-app render report", report)
         // Also to the log, so a session driving the app over adb can read the same report the
@@ -1226,6 +1288,29 @@ fun SessionScreen(
         DebugStats.reset()
         Toast.makeText(context, "Copied render report", Toast.LENGTH_SHORT).show()
     }
+    val copyRenderReport = { buildAndCopyReport() }
+    // Bench build only: P0's scripted fling/stream/type/keyboard benchmark (BenchRun.kt), against
+    // the fixture session opened below instead of a real server. Null everywhere else -- see
+    // [SessionSettingsDialog]'s onRunBenchmark.
+    val runBenchmark: (() -> Unit)? =
+        if (BuildConfig.FIXTURE_MODE) {
+            {
+                settingsOpen = false
+                scope.launch {
+                    val extra =
+                        BenchRun.run(
+                            context = context,
+                            scope = scope,
+                            listState = listState,
+                            flingBehavior = flingBehavior,
+                            composerFocus = composerFocus,
+                            setComposerText = { text -> input = atEnd(text) },
+                            view = view,
+                        )
+                    buildAndCopyReport(extra)
+                }
+            }
+        } else null
     Box(Modifier.fillMaxSize()) {
         Column(Modifier.fillMaxSize()) {
             Row(
@@ -1236,20 +1321,35 @@ fun SessionScreen(
                 // A ring's worth, which is what the arrow already keeps on its other three sides.
                 Spacer(Modifier.width(GLYPH_BUTTON_MARGIN))
                 Column(Modifier.weight(1f)) {
+                    // A subagent's own title, with the session's beneath it in a smaller style --
+                    // the header says whose conversation this is as well as what it is. Otherwise
+                    // just the session's title, as before.
+                    if (subagent != null) {
+                        Text(subagent.title, style = MaterialTheme.typography.titleMedium)
+                        Text(
+                            title,
+                            style = MaterialTheme.typography.bodySmall,
+                            color = MaterialTheme.colorScheme.onSurfaceVariant,
+                        )
+                    } else {
                         Text(title, style = MaterialTheme.typography.titleMedium)
-                    // Machine first, then what runs on it -- the same order and the same wording
-                    // everywhere this pair appears, so it reads as one fact rather than two
-                    // sentences with different grammar.
+                        // Machine first, then what runs on it -- the same order and the same
+                        // wording everywhere this pair appears, so it reads as one fact rather than
+                        // two sentences with different grammar.
                         //
-                    // No model. The picker in the footer already shows what this session is set to,
-                    // and showing it twice means two things to keep in step -- they disagreed for a
-                    // moment on every model change.
+                        // No model. The picker in the footer already shows what this session is set
+                        // to, and showing it twice means two things to keep in step -- they
+                        // disagreed for a moment on every model change.
                         Text(
                             "${summary.setupName} · ${summary.provider}",
                             style = MaterialTheme.typography.bodySmall,
                             color = MaterialTheme.colorScheme.onSurfaceVariant,
                         )
                     }
+                }
+                // None of this is a subagent's: it has no files of its own to browse, no settings,
+                // and nothing meters it -- see SUBAGENTS.md's "Phone".
+                //
                 // Beside the provider it reports on, which is the line directly to its left. Its
                 // real home is this provider's settings, which do not exist yet. A session on a
                 // provider with no such service gets an honest "unavailable" rather than a hidden
@@ -1264,6 +1364,7 @@ fun SessionScreen(
                 // Usage, files, settings -- widest scope first, narrowing to the right, so the cog
                 // stays at the end where every other screen keeps it. Asked for in this order by
                 // Iris on 2026-09-03.
+                if (!isSubagent) {
                     Row {
                         GlyphButton(
                             USAGE_GLYPH,
@@ -1281,25 +1382,29 @@ fun SessionScreen(
                                     FilesTarget(
                                         setup = summary.setup,
                                         setupName = summary.setupName,
-                                    // Where this session works, and the machine's own home when it
-                                    // was never given a directory -- resolved there rather than
-                                    // guessed at here, since this app does not know that home.
+                                        // Where this session works, and the machine's own home when
+                                        // it was never given a directory -- resolved there rather
+                                        // than guessed at here, since this app does not know that
+                                        // home.
                                         start = summary.cwd?.takeIf { it.isNotBlank() } ?: "~",
                                     )
                                 )
                             },
                         )
-                    // What it opens is about this session, so it sits at the end of the session's
-                    // own row. A cog and not a word because there will be more, and a bar of words
-                    // has nowhere to put it.
+                        // What it opens is about this session, so it sits at the end of the
+                        // session's own row. A cog and not a word because there will be more, and a
+                        // bar of words has nowhere to put it.
                         GlyphButton(SETTINGS_GLYPH, "Session settings", { settingsOpen = true })
                     }
                 }
+            }
 
             // Under the header, above everything the session itself says: it is a fact about the
             // machine rather than a turn in the conversation, and it is the number that decides
-            // whether to keep going.
+            // whether to keep going. Nothing meters a subagent.
+            if (!isSubagent) {
                 SessionUsageBar(usage)
+            }
 
             (streamError ?: actionError)?.let { message ->
                 Text(
@@ -1545,6 +1650,7 @@ fun SessionScreen(
                                                 is TranscriptItem.ClearedNote -> ClearedRow()
                                                 is TranscriptItem.CompactedNote ->
                                                     CompactedRow(item)
+                                                is TranscriptItem.LimitNote -> LimitRow(item)
                                                 // Never reached: a peer message is flattened into
                                                 // its own units. Here because a `when` over the
                                                 // item kinds has to stay exhaustive.
@@ -1643,24 +1749,34 @@ fun SessionScreen(
                 )
             }
 
+            // Kept for a subagent -- see SUBAGENTS.md's "Phone" -- with the wording that turns
+            // "exited" into "finished" for one, since it has no process to leave running or stop.
             SessionStatusRow(
                 status = status,
                 compactingFor = compactingFor,
                 contextTokens = contextTokens,
+                subagent = isSubagent,
             )
 
-            // Between the transcript and the box: above what is being typed, so the list does not
-            // cover the thing the command is about, and below everything that explains it.
+            // Everything from here down is the composer: a subagent cannot be messaged, so none of
+            // it applies -- see SUBAGENTS.md's "Phone".
+            if (!isSubagent) {
+                // Between the transcript and the box: above what is being typed, so the list does
+                // not cover the thing the command is about, and below everything that explains it.
                 CommandSuggestions(
-                // Nothing to suggest about a suggestion that was just taken. `/compact` is a whole
-                // command *and* a prefix of itself, so picking it left the list standing there with
-                // the one row already chosen. Held by what was picked rather than by a flag, so
-                // typing anything else brings the list back without a second thing to reset.
-                commands = if (input.text == picked) emptyList() else suggestedCommands(input.text),
+                    // Nothing to suggest about a suggestion that was just taken. `/compact` is a
+                    // whole command *and* a prefix of itself, so picking it left the list standing
+                    // there with the one row already chosen. Held by what was picked rather than by
+                    // a flag, so typing anything else brings the list back without a second thing
+                    // to
+                    // reset.
+                    commands =
+                        if (input.text == picked) emptyList() else suggestedCommands(input.text),
                     onPick = { command ->
-                    // At the end of what was inserted, which is where the reader carries on typing:
-                    // a command with an argument is put in the box half-written, and a cursor left
-                    // at the front makes the next keystroke the first character of "/rename".
+                        // At the end of what was inserted, which is where the reader carries on
+                        // typing: a command with an argument is put in the box half-written, and a
+                        // cursor left at the front makes the next keystroke the first character of
+                        // "/rename".
                         input = atEnd(command.typed())
                         picked = command.typed()
                     },
@@ -1669,11 +1785,14 @@ fun SessionScreen(
                 // Always enabled -- a send while the session is running becomes a steering message
                 // injected at the next tool boundary, which is the point of the whole app.
                 //
-            // The field gets a row of its own, above the buttons: sharing one put the full width
-            // behind three controls, so the thing being typed into was the narrowest on the row.
+                // The field gets a row of its own, above the buttons: sharing one put the full
+                // width
+                // behind three controls, so the thing being typed into was the narrowest on the
+                // row.
                 Column(Modifier.fillMaxWidth().padding(8.dp)) {
                     // Directly above the box they will be sent from, so what is attached is visible
-                // rather than counted: the "+2" on the button below said how many and never which.
+                    // rather than counted: the "+2" on the button below said how many and never
+                    // which.
                     PendingAttachments(
                         settings = settings,
                         sessionId = summary.id,
@@ -1686,9 +1805,12 @@ fun SessionScreen(
                             input = it
                             saveDraft(context, summary.id, it.text)
                         },
-                    modifier = Modifier.fillMaxWidth(),
-                    // No longer "(+image)": the images are on screen above this, and a placeholder
-                    // saying so said it in words beside the thing itself.
+                        // BenchRun's type phase requests focus on this exact field
+                        // (`composerFocus`)
+                        // so it types through the real composer rather than a stand-in.
+                        modifier = Modifier.fillMaxWidth().focusRequester(composerFocus),
+                        // No longer "(+image)": the images are on screen above this, and a
+                        // placeholder saying so said it in words beside the thing itself.
                         placeholder = { Text("Message") },
                         maxLines = 4,
                     )
@@ -1696,17 +1818,19 @@ fun SessionScreen(
                         verticalAlignment = Alignment.CenterVertically,
                         modifier = Modifier.fillMaxWidth(),
                     ) {
-                    // Photo or file, asked here rather than by two buttons: the row is full, and
+                        // Photo or file, asked here rather than by two buttons: the row is full,
+                        // and
                         // attaching is one action whichever picker answers it.
                         var attaching by remember { mutableStateOf(false) }
                         Box {
-                        // Just "+". The count it used to carry was standing in for showing them.
+                            // Just "+". The count it used to carry was standing in for showing
+                            // them.
                             BubbleButton(onClick = { attaching = true }) { Text("+") }
                             DropdownMenu(
                                 expanded = attaching,
                                 onDismissRequest = { attaching = false },
-                            // See PickerButton: without this the menu opens a status bar's height
-                            // away from the button in an edge-to-edge activity.
+                                // See PickerButton: without this the menu opens a status bar's
+                                // height away from the button in an edge-to-edge activity.
                                 properties = PopupProperties(clippingEnabled = false),
                                 shape = BubbleMenuShape,
                             ) {
@@ -1730,11 +1854,11 @@ fun SessionScreen(
                                 )
                             }
                         }
-                    // The settings share what is left after the actions have taken what they need.
-                    // A Row hands out intrinsic widths in order and clips whatever runs past the
-                    // edge, so with these laid out first the arrival of Stop pushed Send off the
-                    // screen entirely -- the app's central control, gone at the moment it is most
-                    // in use.
+                        // The settings share what is left after the actions have taken what they
+                        // need. A Row hands out intrinsic widths in order and clips whatever runs
+                        // past the edge, so with these laid out first the arrival of Stop pushed
+                        // Send off the screen entirely -- the app's central control, gone at the
+                        // moment it is most in use.
                         Row(
                             verticalAlignment = Alignment.CenterVertically,
                             modifier = Modifier.weight(1f),
@@ -1742,16 +1866,18 @@ fun SessionScreen(
                             if (offeredModels.isNotEmpty()) {
                                 PickerButton(
                                     current = modelLabel(model),
-                                // What the machine offers, plus the state a session is in when it
-                                // has chosen none of them. The button has always been able to say
-                                // "default"; until this the list could not, so leaving it was a
-                                // one-way trip.
+                                    // What the machine offers, plus the state a session is in when
+                                    // it has chosen none of them. The button has always been able
+                                    // to
+                                    // say "default"; until this the list could not, so leaving it
+                                    // was a one-way trip.
                                     options = listOf(DEFAULT_MODEL) + offeredModels,
-                                // Not set here. The button follows what the session reports it is
-                                // set to, which arrives a moment later and is sometimes a different
-                                // answer -- a name the CLI resolved, or no change at all on a
-                                // provider whose model is fixed. Asked about first, unless there is
-                                // nothing to lose by it -- see [ModelSwitchWarning].
+                                    // Not set here. The button follows what the session reports it
+                                    // is set to, which arrives a moment later and is sometimes a
+                                    // different answer -- a name the CLI resolved, or no change at
+                                    // all on a provider whose model is fixed. Asked about first,
+                                    // unless there is nothing to lose by it -- see
+                                    // [ModelSwitchWarning].
                                     onPick = { chosen ->
                                         if (
                                             modelLabel(chosen) == modelLabel(model) ||
@@ -1772,14 +1898,16 @@ fun SessionScreen(
                                 },
                             )
                         }
-                    // The same filled shape as the button beside it, not an outlined one: these are
-                    // two things you can do about the session, and weighting one as secondary said
-                    // they were a primary action and its qualifier. What separates them is the
-                    // colour and the mark, which is what they mean.
+                        // The same filled shape as the button beside it, not an outlined one: these
+                        // are two things you can do about the session, and weighting one as
+                        // secondary said they were a primary action and its qualifier. What
+                        // separates them is the colour and the mark, which is what they mean.
                         //
-                    // Always here, rather than arriving with the turn as it used to. A control that
-                    // comes and goes makes its own presence the signal, and a button always in the
-                    // same place also cannot push Send off the end of the row by turning up.
+                        // Always here, rather than arriving with the turn as it used to. A control
+                        // that comes and goes makes its own presence the signal, and a button
+                        // always
+                        // in the same place also cannot push Send off the end of the row by turning
+                        // up.
                         val process =
                             when {
                                 running -> ProcessAction.Pause
@@ -1799,18 +1927,21 @@ fun SessionScreen(
                             Glyph(
                                 process.glyph,
                                 colour = LocalContentColor.current,
-                            modifier = Modifier.semantics { contentDescription = process.label },
+                                modifier =
+                                    Modifier.semantics { contentDescription = process.label },
                             )
                         }
                         Spacer(Modifier.width(8.dp))
-                    // The paper plane, with a clock on it while a turn is in flight: sending then
-                    // queues the message for the next tool boundary rather than starting a turn of
-                    // its own, and the two have to be told apart at a glance. The label says the
-                    // same thing to a screen reader.
+                        // The paper plane, with a clock on it while a turn is in flight: sending
+                        // then queues the message for the next tool boundary rather than starting a
+                        // turn of its own, and the two have to be told apart at a glance. The label
+                        // says the same thing to a screen reader.
                         //
-                    // Disabled while there is nothing to send, rather than pressable and silent:
-                    // `send` has always returned early on an empty composer, so the button promised
-                    // something it would not do. Disabled and not hidden, for the reason above.
+                        // Disabled while there is nothing to send, rather than pressable and
+                        // silent:
+                        // `send` has always returned early on an empty composer, so the button
+                        // promised something it would not do. Disabled and not hidden, for the
+                        // reason above.
                         Button(
                             onClick = { send() },
                             enabled = input.text.isNotBlank() || pendingAttachments.isNotEmpty(),
@@ -1827,12 +1958,13 @@ fun SessionScreen(
                 }
             }
         }
+    }
 
     // Beside the other two dialogs, and outside the list for the same reason as them: what is open
     // is the screen's business rather than any row's. See [SessionImageViewer].
     fullImage?.let { ref -> SessionImageViewer(settings, summary.id, ref) { fullImage = null } }
     if (usageOpen) {
-        UsageDialog(feed = usageFeed, onDismiss = { usageOpen = false })
+        usageFeed?.let { UsageDialog(feed = it, onDismiss = { usageOpen = false }) }
     }
     if (settingsOpen) {
         // Measured when the dialog opens rather than kept up to date: what the reader is being told
@@ -1846,6 +1978,8 @@ fun SessionScreen(
             settings = settings,
             sessionId = summary.id,
             title = title,
+            effort = summary.effort.takeIf { summary.takesEffort },
+            takesEffort = summary.takesEffort,
             cachedBytes = cachedBytes,
             // The purge finishes before the epoch moves, because the relaunched opening effect
             // reads the same directory and would otherwise draw what is about to be deleted. The
@@ -1869,6 +2003,7 @@ fun SessionScreen(
             },
             onDismiss = { settingsOpen = false },
             onCopyRenderReport = copyRenderReport,
+            onRunBenchmark = runBenchmark,
         )
     }
 }
@@ -2122,6 +2257,13 @@ private fun SessionStatusRow(
     /** Context the session is holding, or null where nothing has measured it. */
     contextTokens: Long?,
     modifier: Modifier = Modifier,
+    /**
+     * Whether this row is for a subagent rather than a session, which changes only one word:
+     * "exited" reads as "finished" there too, the same as the subagent list's own card -- a
+     * subagent's process was always its parent's, so "exited" would read as a fault rather than the
+     * ordinary way one of these ends.
+     */
+    subagent: Boolean = false,
 ) {
     DebugStats.count("status row recomposed")
     Row(
@@ -2178,7 +2320,7 @@ private fun SessionStatusRow(
                 Text(
                     when (status) {
                         "idle" -> "idle"
-                        "exited" -> "exited"
+                        "exited" -> if (subagent) "finished" else "exited"
                         "awaitingInput" -> "your turn"
                         "unknown" -> "can't tell"
                         else -> status
@@ -2249,7 +2391,7 @@ private const val ONE_TAP_MS = 250L
  * session is set to without spending a second line on saying it.
  */
 @Composable
-private fun PickerButton(current: String, options: List<String>, onPick: (String) -> Unit) {
+fun PickerButton(current: String, options: List<String>, onPick: (String) -> Unit) {
     var open by remember { mutableStateOf(false) }
     // When an outside touch last closed the menu.
     //

app/androidApp/src/main/kotlin/com/example/aiapp/SessionSettingsDialog.kt

@@ -6,8 +6,10 @@ import androidx.compose.foundation.layout.Spacer
 import androidx.compose.foundation.layout.fillMaxWidth
 import androidx.compose.foundation.layout.height
 import androidx.compose.foundation.layout.width
+import androidx.compose.foundation.rememberScrollState
 import androidx.compose.foundation.text.KeyboardActions
 import androidx.compose.foundation.text.KeyboardOptions
+import androidx.compose.foundation.verticalScroll
 import androidx.compose.material3.AlertDialog
 import androidx.compose.material3.CircularProgressIndicator
 import androidx.compose.material3.MaterialTheme
@@ -26,6 +28,10 @@ import androidx.compose.ui.Alignment
 import androidx.compose.ui.Modifier
 import androidx.compose.ui.text.input.ImeAction
 import androidx.compose.ui.unit.dp
+import java.time.Instant
+import java.time.ZoneId
+import java.time.format.DateTimeFormatter
+import java.time.format.FormatStyle
 import kotlinx.coroutines.Dispatchers
 import kotlinx.coroutines.launch
 import kotlinx.coroutines.withContext
@@ -54,6 +60,16 @@ fun SessionSettingsDialog(
      */
     title: String,
     onRenamed: (String) -> Unit,
+    /**
+     * How hard the model thinks, as the session reports it, or null for the CLI's own default.
+     *
+     * Taken from the row this dialog was opened over rather than fetched, because unlike the
+     * notification switch there is nothing else that changes it: the level is this app's to set and
+     * the server does not resolve it into something else.
+     */
+    effort: String?,
+    /** Whether a level does anything here; the row is left out entirely where it does not. */
+    takesEffort: Boolean,
     /**
      * What this phone is holding of the conversation, or null while that is being measured -- see
      * the Reload row below, which is what would discard it.
@@ -66,9 +82,18 @@ fun SessionSettingsDialog(
      * measures is that screen's own state.
      */
     onCopyRenderReport: () -> Unit,
+    /**
+     * Runs P0's scripted scroll-and-stream benchmark and copies the extended report, or null on
+     * every build but `bench` -- see [BuildConfig.FIXTURE_MODE] and BenchRun.kt. Null rather than
+     * always-present-but-disabled: this has no meaning at all outside the bench build, and a
+     * control with nothing behind it on every other build is not a state worth drawing.
+     */
+    onRunBenchmark: (() -> Unit)? = null,
 ) {
     val scope = rememberCoroutineScope()
     var name by remember(sessionId) { mutableStateOf(title) }
+    var level by remember(sessionId) { mutableStateOf(effort) }
+    var effortError by remember { mutableStateOf<String?>(null) }
     var saving by remember { mutableStateOf(false) }
     var error by remember { mutableStateOf<String?>(null) }
     // Null until the server has been asked. The row this dialog was opened over is a snapshot of
@@ -77,6 +102,16 @@ fun SessionSettingsDialog(
     // and a spinner sits beside it, which is what not knowing looks like.
     var notify by remember(sessionId) { mutableStateOf<Boolean?>(null) }
     var notifyError by remember { mutableStateOf<String?>(null) }
+    // The same three-state shape the notification switch has, for the same reason: until the
+    // server has answered, the switch is disabled rather than showing a position nothing confirmed.
+    var autoResume by remember(sessionId) { mutableStateOf<Boolean?>(null) }
+    var resumeMessage by remember(sessionId) { mutableStateOf(DEFAULT_RESUME_MESSAGE) }
+    // When the server next intends to ask whether the limit has lifted, or null when nothing is
+    // waiting. Read once with everything else: it moves on the server's schedule, not this
+    // screen's, and a figure that redrew itself here would be this app re-measuring what it was
+    // told.
+    var resumeAt by remember(sessionId) { mutableStateOf<Double?>(null) }
+    var resumeError by remember { mutableStateOf<String?>(null) }
     // Where the session works. Null until the server has been asked, for the same reason the switch
     // above is. An empty answer is a session that was never given a directory, which is not the
     // same as one whose directory is unknown -- the field is only enabled once one of those is
@@ -90,6 +125,9 @@ fun SessionSettingsDialog(
         try {
             val fresh = withContext(Dispatchers.IO) { fetchSession(settings, sessionId) }
             notify = fresh.notify
+            autoResume = fresh.autoResume
+            resumeMessage = fresh.autoResumeMessage
+            resumeAt = fresh.resumeAt
             cwd = fresh.cwd.orEmpty()
             typedCwd = fresh.cwd.orEmpty()
         } catch (e: ApiException) {
@@ -97,6 +135,8 @@ fun SessionSettingsDialog(
             // instead of offering a position nothing confirmed.
             notifyError = e.message
             notify = null
+            resumeError = e.message
+            autoResume = null
         }
     }
 
@@ -125,6 +165,26 @@ fun SessionSettingsDialog(
         }
     }
 
+    /**
+     * Chooses a thinking level, which ends the process the old level was launched with.
+     *
+     * Put back if the request is refused, for the reason the notification switch below gives: a
+     * control that stays where it was put after a refusal is stating something untrue.
+     */
+    fun setEffort(chosen: String?) {
+        val was = level
+        level = chosen
+        effortError = null
+        scope.launch {
+            try {
+                withContext(Dispatchers.IO) { setSessionEffort(settings, sessionId, chosen) }
+            } catch (e: ApiException) {
+                level = was
+                effortError = e.message
+            }
+        }
+    }
+
     // Moved optimistically so the switch answers the finger that moved it, and put back if the
     // request is refused -- a switch that waits for a round trip reads as broken on a slow tunnel,
     // and one that stays moved after a refusal lies.
@@ -142,6 +202,39 @@ fun SessionSettingsDialog(
         }
     }
 
+    /**
+     * Turns auto-resume on or off, or changes what it would say.
+     *
+     * One request for both, because the server takes one: switching it on and typing the message
+     * are two halves of the same decision, and sending them separately would leave a moment where
+     * the session is armed with the old words.
+     *
+     * Put back if refused, like the notification switch. Turning it off also clears what was
+     * scheduled -- said here rather than only on the server, or the row would go on naming a time
+     * that no longer exists.
+     */
+    fun setAutoResume(on: Boolean, message: String) {
+        val wasOn = autoResume
+        val wasMessage = resumeMessage
+        val wasAt = resumeAt
+        autoResume = on
+        resumeMessage = message
+        if (!on) resumeAt = null
+        resumeError = null
+        scope.launch {
+            try {
+                withContext(Dispatchers.IO) {
+                    setSessionAutoResume(settings, sessionId, on, message)
+                }
+            } catch (e: ApiException) {
+                autoResume = wasOn
+                resumeMessage = wasMessage
+                resumeAt = wasAt
+                resumeError = e.message
+            }
+        }
+    }
+
     // Nothing to do when the name has not changed, so the button says so rather than sending a
     // request whose success would look exactly like the failure of having typed nothing.
     val changed = name.trim().isNotEmpty() && name.trim() != title
@@ -168,7 +261,10 @@ fun SessionSettingsDialog(
         onDismissRequest = onDismiss,
         title = { Text("Session settings") },
         text = {
-            Column {
+            // Scrollable, because this dialog grew past a screenful: a Material dialog constrains
+            // its own height and clips what does not fit, so the last control on the list is one
+            // large system font away from being unreachable with nothing on screen to say so.
+            Column(Modifier.verticalScroll(rememberScrollState())) {
                 OutlinedTextField(
                     value = name,
                     onValueChange = { name = it },
@@ -212,6 +308,70 @@ fun SessionSettingsDialog(
                     )
                 }
                 Spacer(Modifier.height(8.dp))
+                Row(
+                    verticalAlignment = Alignment.CenterVertically,
+                    modifier = Modifier.fillMaxWidth(),
+                ) {
+                    Text("Resume after a usage limit", modifier = Modifier.weight(1f))
+                    if (autoResume == null && resumeError == null) {
+                        CircularProgressIndicator(
+                            modifier = Modifier.width(16.dp).height(16.dp),
+                            strokeWidth = 2.dp,
+                        )
+                        Spacer(Modifier.width(8.dp))
+                    }
+                    Switch(
+                        checked = autoResume == true,
+                        onCheckedChange = { setAutoResume(it, resumeMessage) },
+                        enabled = autoResume != null,
+                    )
+                }
+                // Disabled rather than hidden while the switch is off: a field that comes and goes
+                // makes its own presence the signal, and a visible one teaches what the switch will
+                // do. Committed on the keyboard's Done rather than on every keystroke, so typing a
+                // sentence is one request instead of one per letter.
+                OutlinedTextField(
+                    value = resumeMessage,
+                    onValueChange = { resumeMessage = it },
+                    label = { Text("Message to send") },
+                    // What an empty field means, in the field: the server's own word rather than a
+                    // session poked with nothing to read.
+                    placeholder = { Text(DEFAULT_RESUME_MESSAGE) },
+                    singleLine = true,
+                    enabled = autoResume == true,
+                    modifier = Modifier.fillMaxWidth(),
+                    keyboardOptions = KeyboardOptions(imeAction = ImeAction.Done),
+                    keyboardActions =
+                        KeyboardActions(onDone = { setAutoResume(true, resumeMessage) }),
+                )
+                // What it does and what it costs, in the order it happens. The last sentence is the
+                // one that matters: the time below is when the server will *ask*, not a promise
+                // about when the session speaks.
+                Text(
+                    "When this session stops because the account is out of quota, the server " +
+                        "checks the limit and sends this message once it has lifted. It checks " +
+                        "again if the limit is still on.",
+                    style = MaterialTheme.typography.bodySmall,
+                    color = MaterialTheme.colorScheme.onSurfaceVariant,
+                )
+                // Only where something is actually waiting. Absent is not a state worth a row: a
+                // session that has not hit a limit has nothing scheduled, which the reader can see
+                // from the switch.
+                resumeAt?.let { at ->
+                    Text(
+                        "Waiting now -- next check ${formatCheckTime(at)}.",
+                        style = MaterialTheme.typography.bodySmall,
+                        color = MaterialTheme.colorScheme.onSurfaceVariant,
+                    )
+                }
+                resumeError?.let {
+                    Text(
+                        it,
+                        color = MaterialTheme.colorScheme.error,
+                        style = MaterialTheme.typography.bodySmall,
+                    )
+                }
+                Spacer(Modifier.height(8.dp))
                 Row(
                     verticalAlignment = Alignment.CenterVertically,
                     modifier = Modifier.fillMaxWidth(),
@@ -257,6 +417,44 @@ fun SessionSettingsDialog(
                         style = MaterialTheme.typography.bodySmall,
                     )
                 }
+                // Left out rather than disabled, the one place this dialog does that: a disabled
+                // control teaches what the thing can do, and a llama session cannot do this at all
+                // -- the row would be teaching something false about it.
+                if (takesEffort) {
+                    Spacer(Modifier.height(8.dp))
+                    Row(
+                        verticalAlignment = Alignment.CenterVertically,
+                        modifier = Modifier.fillMaxWidth(),
+                    ) {
+                        Text("Thinking", modifier = Modifier.weight(1f))
+                        PickerButton(
+                            current = level ?: DEFAULT_EFFORT,
+                            // The level the CLI picks for itself is in the list as well as in the
+                            // button, so leaving a level is not a one-way trip -- the same
+                            // correction the model picker carries.
+                            options = listOf(DEFAULT_EFFORT) + EFFORT_LEVELS,
+                            onPick = { chosen ->
+                                setEffort(chosen.takeIf { it != DEFAULT_EFFORT })
+                            },
+                        )
+                    }
+                    // What it costs, said where it is about to be pressed, like Move above: the
+                    // CLI reads the level when it launches and has no control request for
+                    // changing one.
+                    Text(
+                        "Changing this stops the session's process. It starts again with the " +
+                            "next message, or with Start.",
+                        style = MaterialTheme.typography.bodySmall,
+                        color = MaterialTheme.colorScheme.onSurfaceVariant,
+                    )
+                    effortError?.let {
+                        Text(
+                            it,
+                            color = MaterialTheme.colorScheme.error,
+                            style = MaterialTheme.typography.bodySmall,
+                        )
+                    }
+                }
                 Spacer(Modifier.height(8.dp))
                 Row(
                     verticalAlignment = Alignment.CenterVertically,
@@ -317,6 +515,21 @@ fun SessionSettingsDialog(
                     Text("Render timings", modifier = Modifier.weight(1f))
                     TextButton(onClick = onCopyRenderReport) { Text("Copy") }
                 }
+                // Bench-build only: see [onRunBenchmark]. Named exactly "Run benchmark" because
+                // ui-trace and the emulator smoke run find it by that label, the same way every
+                // other control here is found -- see AGENTS.md's "Driving the UI".
+                onRunBenchmark?.let { run ->
+                    Spacer(Modifier.height(8.dp))
+                    Row(
+                        verticalAlignment = Alignment.CenterVertically,
+                        modifier = Modifier.fillMaxWidth(),
+                    ) {
+                        Glyph(SPEED_GLYPH, colour = MaterialTheme.colorScheme.onSurface)
+                        Spacer(Modifier.width(8.dp))
+                        Text("P0 benchmark", modifier = Modifier.weight(1f))
+                        TextButton(onClick = run) { Text("Run benchmark") }
+                    }
+                }
             }
         },
         // Disabled rather than absent while there is nothing to save: a button that comes and goes
@@ -329,3 +542,21 @@ fun SessionSettingsDialog(
         dismissButton = { TextButton(onClick = onDismiss) { Text("Close") } },
     )
 }
+
+/**
+ * When the server will next look, as a local time.
+ *
+ * A time rather than a countdown, for the reason the transcript's own limit row gives: this screen
+ * reads the figure once, and a span drawn from a value nothing refreshes goes stale while somebody
+ * is looking at it.
+ */
+private fun formatCheckTime(epochSeconds: Double): String =
+    try {
+        DateTimeFormatter.ofLocalizedTime(FormatStyle.SHORT)
+            .withZone(ZoneId.systemDefault())
+            .format(Instant.ofEpochSecond(epochSeconds.toLong()))
+    } catch (_: Exception) {
+        // A time that cannot be read is not a time to show: the sentence above still says a check
+        // is coming, which is the part the reader can act on.
+        "soon"
+    }

app/androidApp/src/main/kotlin/com/example/aiapp/SessionUsageBar.kt

@@ -70,12 +70,21 @@ class UsageFeed(
     /** Ask the backend again now. The dialog's refresh button; the poll does it on its own. */
     val refresh: () -> Unit,
 ) {
-    /** What [setup]'s own limits came back as. See [usageFor] for why the states are these. */
-    fun forSetup(setup: String): SessionUsage =
-        when (val state = snapshots) {
+    /**
+     * What meters [session], and what that meter came back as. See [usageFor] for the states.
+     *
+     * A session rather than a machine, because a machine is not what is metered: one machine runs
+     * the Claude CLI and an echo session side by side, and only the first of them spends anything.
+     */
+    fun forSession(session: SessionSummary): SessionUsage {
+        // Settled without asking anybody: a session nothing meters has nothing to check, and
+        // "checking" is what the fetch's own states would say about it for as long as one is out.
+        val provider = session.usageProvider ?: return SessionUsage.NotMetered
+        return when (val state = snapshots) {
             is LoadState.Loading -> SessionUsage.Waiting
             is LoadState.Error -> SessionUsage.Unavailable(state.message)
-            is LoadState.Loaded -> usageFor(state.value, setup)
+            is LoadState.Loaded -> usageFor(state.value, session.setup, provider)
+        }
     }
 }
 
@@ -158,9 +167,15 @@ fun SessionUsageBar(usage: SessionUsage, modifier: Modifier = Modifier) {
         }
     }
 
-    // Nothing at all for a machine that meters nothing: a row saying "unknown" there would report a
-    // problem about a setup somebody chose, on every screen, forever.
-    if (usage is SessionUsage.NotMetered) {
+    // Nothing at all for a session that meters nothing: a row saying "unknown" there would report
+    // a problem about a setup somebody chose, on every screen, forever.
+    //
+    // And nothing while the first fetch is out, which is a different silence. A request in flight
+    // is not a state to report -- and the session that meters nothing is exactly the one this
+    // cannot yet tell apart, so "5-hour usage: checking" appeared under an echo session for half a
+    // second and was then taken away. A row that has to be withdrawn is worse than one that
+    // arrives late.
+    if (usage is SessionUsage.NotMetered || usage is SessionUsage.Waiting) {
         return
     }
 
@@ -171,9 +186,10 @@ fun SessionUsageBar(usage: SessionUsage, modifier: Modifier = Modifier) {
         // Words, not a colour and not an empty bar: every one of these is a different kind of
         // answer from "this much is used", and only words carry a difference in kind.
         when (val state = usage) {
-            SessionUsage.NotMetered -> Unit
+            // Both handled above, before the row exists at all.
+            SessionUsage.NotMetered,
+            SessionUsage.Waiting -> Unit
             is SessionUsage.Unavailable -> UsageNote("5-hour usage unknown -- ${state.why}")
-            SessionUsage.Waiting -> UsageNote("5-hour usage: checking")
             is SessionUsage.Known -> {
                 val window = state.windows.firstOrNull { it.kind == "session" }
                 if (window == null) {
@@ -234,16 +250,22 @@ private fun fiveHourLabel(window: UsageWindow, now: OffsetDateTime): String {
 }
 
 /**
- * One machine's snapshot, out of every machine's.
+ * One meter's snapshot, out of every machine's: [setup]'s row for [provider].
+ *
+ * Both halves are needed to pick it. A machine can hold more than one meter -- the Claude CLI's
+ * account and, while a test has one set, an echo session's invented one -- and a snapshot is one
+ * service on one machine.
  *
  * Every way of having *failed* to get numbers is [SessionUsage.Unavailable] with the reason in it.
  * None of them may look like zero, and none may look like [SessionUsage.NotMetered], which is the
  * machine having no quota rather than the question going unanswered.
  */
-fun usageFor(snapshots: List<UsageSnapshot>, setup: String): SessionUsage {
-    // No snapshot at all means the backend never asked, which it only does for a machine with
-    // nothing metered on it. That is a different answer from having asked and failed.
-    val mine = snapshots.firstOrNull { it.setup == setup } ?: return SessionUsage.NotMetered
+fun usageFor(snapshots: List<UsageSnapshot>, setup: String, provider: String): SessionUsage {
+    // No snapshot at all means the backend never asked, which it only does where there is nothing
+    // to ask about. That is a different answer from having asked and failed.
+    val mine =
+        snapshots.firstOrNull { it.setup == setup && it.provider == provider }
+            ?: return SessionUsage.NotMetered
     if (mine.state != "ok") {
         return SessionUsage.Unavailable(mine.detail ?: mine.state)
     }

app/androidApp/src/main/kotlin/com/example/aiapp/SetupsScreen.kt

@@ -236,6 +236,7 @@ private fun AddSetupDialog(
     var address by remember { mutableStateOf("") }
     var identity by remember { mutableStateOf("") }
     var attachmentsDir by remember { mutableStateOf("") }
+    var modelsDir by remember { mutableStateOf("") }
     var tested by remember { mutableStateOf<String?>(null) }
     var testing by remember { mutableStateOf(false) }
 
@@ -250,6 +251,7 @@ private fun AddSetupDialog(
                     port = typedPort,
                     identityFile = identity.trim().ifEmpty { null },
                     attachmentsDir = attachmentsDir.trim().ifEmpty { null },
+                    modelsDir = modelsDir.trim().ifEmpty { null },
                 )
             }
 
@@ -293,6 +295,14 @@ private fun AddSetupDialog(
                     label = { Text("Folder for attached files (optional)") },
                     singleLine = true,
                 )
+                // Where that machine's GGUFs are, for a llama.cpp session on it. Blank means
+                // the same place this backend keeps its own downloads, read on that machine.
+                OutlinedTextField(
+                    value = modelsDir,
+                    onValueChange = { modelsDir = it },
+                    label = { Text("Folder for models (optional)") },
+                    singleLine = true,
+                )
                 tested?.let {
                     Spacer(Modifier.height(8.dp))
                     Text(it, style = MaterialTheme.typography.bodySmall)

app/androidApp/src/main/kotlin/com/example/aiapp/SpawnScreen.kt

@@ -61,18 +61,31 @@ fun SpawnScreen(
     // "auto" rather than "manual": on a phone every ask is a round trip to a question card, and
     // answering "allow Bash?" dozens of times per task is what this app exists to avoid.
     var permissionMode by remember { mutableStateOf("auto") }
+    // Null until the server has been asked, and null again if it answers "no level chosen" -- the
+    // two are told apart by [defaultsAsked], because a picker that shows a level before the answer
+    // arrives is one you can spawn at without having chosen it.
+    var effort by remember { mutableStateOf<String?>(null) }
+    var defaultsAsked by remember { mutableStateOf(false) }
     var busy by remember { mutableStateOf(false) }
     // Only the spawn's own failure. The fetch's lives in `options`: this one leaves a filled-in
     // form worth keeping, and that one leaves nothing to fill in.
     var spawnError by remember { mutableStateOf<String?>(null) }
-    // Downloaded models, for a llama provider to choose between. Kept separate from the setups: a
-    // Claude session needs none, so failing to list them must not stop the screen rendering.
+    // The models on the *chosen machine*, for a llama provider to choose between. Kept separate
+    // from the setups: a Claude session needs none, so failing to list them must not stop the
+    // screen rendering. Refetched when the machine changes, because a model is a file on one
+    // machine -- see [fetchSetupModels].
     var models by remember { mutableStateOf<List<LocalModel>>(emptyList()) }
     var modelKey by remember { mutableStateOf<String?>(null) }
     var contextSize by remember { mutableStateOf("") }
     var temperature by remember { mutableStateOf("") }
 
     LaunchedEffect(Unit) {
+        // Separate from the setups fetch below and deliberately not fatal: failing to learn the
+        // default must leave a screen you can still spawn from, so the picker stays on "default"
+        // and says so rather than the whole form refusing to draw.
+        runCatching { withContext(Dispatchers.IO) { fetchDefaultEffort(settings) } }
+            .onSuccess { effort = it }
+        defaultsAsked = true
         options =
             try {
                 val fetched = withContext(Dispatchers.IO) { fetchSetups(settings) }
@@ -83,9 +96,6 @@ fun SpawnScreen(
             } catch (e: ApiException) {
                 LoadState.failed(e)
             }
-        models =
-            runCatching { withContext(Dispatchers.IO) { fetchModels(settings).local } }
-                .getOrDefault(emptyList())
     }
 
     Column(Modifier.fillMaxSize().verticalScroll(rememberScrollState()).padding(16.dp)) {
@@ -115,6 +125,17 @@ fun SpawnScreen(
                 is LoadState.Loaded -> state.value
             }
         val setup = setups.firstOrNull { it.name == setupName }
+        // Whichever machine is chosen now, asked again when that changes. The old machine's list
+        // is dropped first rather than left on screen: a file name from another machine looks
+        // exactly like one from this one.
+        LaunchedEffect(setup?.id) {
+            models = emptyList()
+            modelKey = null
+            val id = setup?.id ?: return@LaunchedEffect
+            models =
+                runCatching { withContext(Dispatchers.IO) { fetchSetupModels(settings, id) } }
+                    .getOrDefault(emptyList())
+        }
         val current = setup?.providers?.firstOrNull { it.name == providerName }
         // Only the Claude CLI has models, a working directory and permission modes; keying the
         // extra fields on the kind rather than the provider name keeps a second Claude provider
@@ -175,12 +196,13 @@ fun SpawnScreen(
         )
 
         if (isLlama) {
-            // A llama session names one of the models this backend has downloaded, so the choice is
-            // that list rather than free text -- a name that is not on disk is a session that
-            // cannot start.
+            // A llama session names one of the models on the machine it will run on, so the
+            // choice is that list rather than free text -- a name that is not on that machine's
+            // disk is a session that cannot start.
             if (models.isEmpty()) {
                 Text(
-                    "No models downloaded yet. Get one from the Models screen first.",
+                    "No models on ${setup?.name ?: "this machine"}. The Models screen downloads " +
+                        "to the backend; another machine needs the file put there itself.",
                     style = MaterialTheme.typography.bodyMedium,
                     color = MaterialTheme.colorScheme.onSurfaceVariant,
                 )
@@ -251,6 +273,19 @@ fun SpawnScreen(
                 selected = permissionMode,
                 onSelect = { permissionMode = it },
             )
+            Spacer(Modifier.height(16.dp))
+
+            // Says what it does to *later* spawns as well, because it does: the level chosen here
+            // is stored as the default, which is the whole way that default is set. A picker that
+            // quietly changed a global would be the same control with the fact left out.
+            ChipGroup(
+                label = "Thinking (kept as the default for new sessions)",
+                options = listOf(DEFAULT_EFFORT) + EFFORT_LEVELS,
+                // The CLI's own default is a level in the list, so this cannot be a one-way trip.
+                // Disabled-looking until the server has answered, for the reason above.
+                selected = if (defaultsAsked) effort ?: DEFAULT_EFFORT else null,
+                onSelect = { chosen -> effort = chosen.takeIf { it != DEFAULT_EFFORT } },
+            )
         }
         Spacer(Modifier.height(24.dp))
 
@@ -268,6 +303,13 @@ fun SpawnScreen(
                     try {
                         val spawned =
                             withContext(Dispatchers.IO) {
+                                // Stored before the spawn and not after it: choosing a level is
+                                // an intent about new sessions in general, so a spawn that then
+                                // fails must not also lose the choice. Non-fatal for the same
+                                // reason the fetch above is -- the session is what was asked for.
+                                if (isClaude) {
+                                    runCatching { setDefaultEffort(settings, effort) }
+                                }
                                 spawnSession(
                                     settings,
                                     // The id, not the label: labels are editable and the server
@@ -280,6 +322,7 @@ fun SpawnScreen(
                                         if (isLlama) modelKey else model.trim().takeIf { isClaude },
                                     cwd = cwd.trim().takeIf { isClaude },
                                     permissionMode = permissionMode.takeIf { isClaude },
+                                    effort = effort.takeIf { isClaude },
                                     // Sent only when set, so blank means "whatever llama.cpp does
                                     // by default" rather than a zero.
                                     params =

app/androidApp/src/main/kotlin/com/example/aiapp/TranscriptAddress.kt

@@ -0,0 +1,27 @@
+package com.example.aiapp
+
+/**
+ * Where one transcript lives: a session's own, or one of its subagents'.
+ *
+ * The single mechanism [fetchTranscript], [EventStream], [TranscriptSource] and
+ * [TranscriptCache.session] all take, rather than each growing its own branch between a session and
+ * a subagent -- see SUBAGENTS.md's "Phone" and "Wire shape". A caller that has only a session id
+ * builds one with the one-argument constructor; a subagent's screen supplies both ids.
+ */
+data class TranscriptAddress(val sessionId: String, val subagentId: String? = null) {
+    /** The URL segment naming this transcript, before `/transcript` or `/events`. */
+    val urlPath: String
+        get() =
+            if (subagentId == null) "sessions/$sessionId"
+            else "sessions/$sessionId/subagents/$subagentId"
+
+    /**
+     * Where this transcript's cache lives on the phone, relative to the cache root.
+     *
+     * A subagent's nests under its session's directory rather than sitting beside it, so deleting a
+     * session's cache directory takes its subagents' with it -- the same one-way door the server's
+     * own storage describes.
+     */
+    val cachePath: String
+        get() = if (subagentId == null) sessionId else "$sessionId/subagents/$subagentId"
+}

app/androidApp/src/main/kotlin/com/example/aiapp/TranscriptCache.kt

@@ -35,8 +35,15 @@ class TranscriptCache(
     private val root: File,
     private val warn: (String) -> Unit = { Log.w("ai-app", it) },
 ) {
-    /** The cache for one session, whether or not anything has been stored for it yet. */
-    fun session(id: String): SessionCache = SessionCache(File(root, id), warn)
+    /**
+     * The cache for one transcript, whether or not anything has been stored for it yet.
+     *
+     * A subagent's [TranscriptAddress.cachePath] nests it under its session's directory, so
+     * deleting the session (below) takes its subagents' caches with it -- there is no separate
+     * purge for one.
+     */
+    fun session(address: TranscriptAddress): SessionCache =
+        SessionCache(File(root, address.cachePath), warn)
 
     /**
      * Deletes every session directory not in [ids], called after a successful list fetch. The path

app/androidApp/src/main/kotlin/com/example/aiapp/TranscriptItems.kt

@@ -175,6 +175,18 @@ sealed class TranscriptItem {
         val preTokens: Long?,
         val postTokens: Long?,
     ) : TranscriptItem()
+
+    /**
+     * The account ran out of quota, so the turn stopped here.
+     *
+     * A divider rather than an error: nothing failed, and what a reader scrolling back needs from
+     * it is the same thing a clear or a compaction gives them -- why the conversation stops at this
+     * line.
+     *
+     * [resetsAt] is epoch seconds and null where the session was told nothing, which is a state the
+     * row has words for rather than a time it invents.
+     */
+    data class LimitNote(override val seq: Long, val resetsAt: Double?) : TranscriptItem()
 }
 
 /**
@@ -458,6 +470,7 @@ fun foldEvent(items: List<TranscriptItem>, entry: SeqEvent): List<TranscriptItem
             } else {
                 items + TranscriptItem.ImageItem(entry.seq, event.ref)
             }
+        is SessionEvent.LimitReached -> items + TranscriptItem.LimitNote(entry.seq, event.resetsAt)
         is SessionEvent.Cleared -> items + TranscriptItem.ClearedNote(entry.seq)
         is SessionEvent.Compacted ->
             items + TranscriptItem.CompactedNote(entry.seq, event.preTokens, event.postTokens)

app/androidApp/src/main/kotlin/com/example/aiapp/TranscriptSource.kt

@@ -18,7 +18,7 @@ import java.util.concurrent.atomic.AtomicReference
  */
 class TranscriptSource(
     private val settings: ServerSettings,
-    private val sessionId: String,
+    private val address: TranscriptAddress,
     val cache: SessionCache,
 ) {
     private val stream = AtomicReference<EventStream?>(null)
@@ -65,7 +65,7 @@ class TranscriptSource(
         val tail = cache.tail() ?: return false
         // `before = seq + 1` is the newest event with seq <= the cursor, which is the event *at*
         // the cursor when the server still has one there.
-        val answer = fetchTranscript(settings, sessionId, before = tail.seq + 1, limit = 1)
+        val answer = fetchTranscript(settings, address, before = tail.seq + 1, limit = 1)
         val matches =
             answer.size == 1 &&
                 try {
@@ -83,7 +83,7 @@ class TranscriptSource(
      */
     suspend fun fetchOpening(): List<SeqEvent> {
         DebugStats.count("transcript page from server")
-        val page = fetchTranscript(settings, sessionId, limit = OPENING_WINDOW)
+        val page = fetchTranscript(settings, address, limit = OPENING_WINDOW)
         page.forEach { (line, entry) -> cache.append(line, entry.seq) }
         cache.flush()
         return page.map { it.second }
@@ -108,7 +108,7 @@ class TranscriptSource(
         val page =
             fetchTranscript(
                 settings,
-                sessionId,
+                address,
                 before = before,
                 limit = limit,
                 coalesce = coalesce,
@@ -131,7 +131,7 @@ class TranscriptSource(
      * well lose.
      */
     fun follow(after: Long, onOpen: () -> Unit, onReset: () -> Unit, onEvent: (SeqEvent) -> Unit) {
-        val opened = EventStream(settings, sessionId)
+        val opened = EventStream(settings, address)
         stream.getAndSet(opened)?.close()
         try {
             opened.run(after, onOpen, onReset) { raw, entry ->

app/androidApp/src/main/res/values/strings.xml

@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="utf-8"?>
+<resources>
+    <!-- Overridden by the `bench` build type's resValue (build.gradle.kts) to "AI Sessions bench",
+         so the two are never mistaken for each other in the launcher or in Settings. -->
+    <string name="app_name">AI Sessions</string>
+</resources>

app/androidApp/src/test/kotlin/com/example/aiapp/LimitRowTest.kt

@@ -0,0 +1,32 @@
+package com.example.aiapp
+
+import java.time.ZoneId
+import kotlin.test.Test
+import kotlin.test.assertEquals
+import kotlin.test.assertTrue
+
+/**
+ * What the transcript says where a session ran out of quota.
+ *
+ * The pair worth a test is the one that reads the same when it goes wrong: a reset time that
+ * arrived and one that never did. The second must not turn into a plausible-looking time, because a
+ * reader has no way of telling an invented one from a reported one.
+ */
+class LimitRowTest {
+    private val utc = ZoneId.of("UTC")
+
+    @Test
+    fun `a reported reset time is shown as a time`() {
+        // 2026-09-05T12:00:00Z. Asserted as a prefix and the clock reading rather than as the
+        // whole string: the platform's own short-time format is what this asks for, and it
+        // differs by JDK and locale down to which space character separates the meridiem.
+        val summary = limitSummary(1_788_609_600.0, utc)
+        assertTrue(summary.startsWith("Usage limit reached • resets "), summary)
+        assertTrue(summary.contains("12:00"), summary)
+    }
+
+    @Test
+    fun `a limit with no reset time says only what is known`() {
+        assertEquals("Usage limit reached", limitSummary(null, utc))
+    }
+}

app/androidApp/src/test/kotlin/com/example/aiapp/TranscriptCacheTest.kt

@@ -23,7 +23,7 @@ class TranscriptCacheTest {
 
     private fun cache() = TranscriptCache(File(temp, "v1/host_8443")) { said += it }
 
-    private fun session(id: String = "s") = cache().session(id)
+    private fun session(id: String = "s") = cache().session(TranscriptAddress(id))
 
     private fun line(seq: Long, type: String = "toolStart") =
         """{"seq":$seq,"ts":1.5,"type":"$type","id":"x"}"""

app/bench-fixture/README.md

@@ -0,0 +1,28 @@
+# The P0 benchmark fixture
+
+`transcript.jsonl` is a synthetic transcript in the app's own event model (the JSON lines
+`GET /sessions/{id}/transcript` returns; see `Events.kt`'s `parseSeqEvent` and
+`server/src/session/driver.rs`) -- never a real one. It is what both the Compose `bench` build
+and iris's bench build open with no server, so the two apps draw exactly the same content and a
+frame-time comparison is measuring the renderer rather than the data.
+
+Generated by `./generate.py` (Python stdlib only, seeded -- `SEED = 20260905` -- so re-running it
+reproduces the same file byte for byte). It writes into `assets/` -- a separate directory from this
+script and README, because the Compose `bench` build type points its own asset source set straight
+at `assets/` (`app/androidApp/build.gradle.kts`'s `sourceSets { getByName("bench") }`), and a Python
+script and a markdown file have no business inside an APK:
+
+- `transcript.jsonl` -- 3,601 events. The first 3,200 (`BACKLOG_COUNT`) are the scrolled-back
+  history the benchmark opens with: user turns, tool calls with kilobyte-scale input/output,
+  assistant replies built from headings, bold/italic/inline code, a link, fenced code blocks that
+  rotate through rust/kotlin/python/sh/json/toml, a markdown table, two embedded images, and
+  periodic `usageDelta`/`compacted` events. The remaining 400 (`STREAM_COUNT`) are not part of the
+  opening window -- both bench harnesses replay them at a fixed rate (20/s) through the same live
+  fold path a real SSE reply arrives on, which is P0's "streaming phase."
+- `bench1.png`, `bench2.png` -- tiny (8x8) flat-colour PNGs, base64-free on disk but served the
+  same way a real attachment is (`GET /sessions/{id}/files/{name}`), referenced by the two
+  `"type":"image"` events in the transcript.
+
+Regenerate after changing the shape (a new event type, a different backlog/stream split) with
+`./generate.py`, and commit the result -- it is checked in rather than generated at build time so
+both apps' bench builds embed the identical bytes without needing this script at build time.

app/bench-fixture/generate.py

@@ -0,0 +1,186 @@
+#!/usr/bin/env python3
+"""Generates transcript.jsonl -- the synthetic fixture P0's benchmark opens in both apps.
+
+Deterministic (fixed seed), so a Compose bench APK and an iris bench APK draw byte-identical
+content: the point of the fixture is a like-for-like comparison, not a realistic one.
+
+Never a real transcript -- see AGENTS.md's ui-sandbox.sh, which this borrows its vocabulary
+style from (headings, code fences, a table, a link) rather than reusing its Claude-Code JSONL
+shape. This file's shape is the *app's own event model* instead: one JSON object per line,
+matching what GET /sessions/{id}/transcript returns and what Events.kt's parseSeqEvent reads
+(server/src/session/driver.rs is the source of truth for the field names).
+
+    ./generate.py                      writes transcript.jsonl and bench1.png/bench2.png here
+
+BACKLOG_COUNT events (seq 1..BACKLOG_COUNT) are the scrolled-back history the benchmark opens
+with. A further STREAM_COUNT events (seq BACKLOG_COUNT+1..) are not part of the opening window;
+both bench harnesses replay them at a fixed rate as the "streaming reply" phase, appended through
+the same live path a real SSE reply arrives on. Keeping both halves in one file means one
+generator and one seed to keep in sync, rather than two fixtures that can drift apart.
+"""
+import base64
+import json
+import random
+import struct
+import zlib
+from pathlib import Path
+
+SEED = 20260905
+BACKLOG_COUNT = 3200
+STREAM_COUNT = 400
+HERE = Path(__file__).resolve().parent / "assets"
+
+random.seed(SEED)
+
+LANGUAGES = ["rust", "kotlin", "python", "sh", "json", "toml"]
+
+CODE_SNIPPETS = {
+    "rust": '''fn fold_event(items: Vec<Item>, seq: u64) -> Vec<Item> {
+    // a comment worth keeping: this is the fold the app's own screen runs
+    let mut out = items;
+    out.push(Item::new(seq));
+    out
+}''',
+    "kotlin": '''fun foldEvent(items: List<TranscriptItem>, entry: SeqEvent): List<TranscriptItem> {
+    // mirrors the server's own event model, one item per line
+    return items + TranscriptItem.from(entry)
+}''',
+    "python": '''def render_report(frames, cpu_ms, rss_kb):
+    # printed for a human to paste back, so every number carries its unit
+    return f"{frames} frames, {cpu_ms}ms cpu, {rss_kb}kb peak rss"''',
+    "sh": '''#!/bin/sh
+# scripted scroll loop, the shape transcript-bench.sh drives on a phone
+for i in $(seq 1 24); do
+    ui-trace record --do "swipe 540 700 540 1600 200"
+done''',
+    "json": '{"seq": 1, "type": "status", "state": "running"}',
+    "toml": '''[package]
+name = "bench-fixture"
+version = "0.1.0"''',
+}
+
+HEADINGS = [
+    "## Plan",
+    "## What changed",
+    "## Why this approach",
+    "### Open questions",
+    "## Results",
+]
+
+WORDS = (
+    "session render report frame budget scroll transcript fold event cache "
+    "cursor probe stream backlog swipe fixture bench compose iris widget layout "
+    "measure place draw tool call token context window anchor"
+).split()
+
+
+def paragraph(n=24):
+    words = [random.choice(WORDS) for _ in range(n)]
+    words[0] = words[0].capitalize()
+    text = " ".join(words) + "."
+    # Sprinkle markdown inline spans so the syntax highlighter/markdown parser sees a real mix.
+    text = text.replace(" fold ", " **fold** ", 1)
+    text = text.replace(" cursor ", " *cursor* ", 1)
+    text = text.replace(" cache ", " `cache` ", 1)
+    if "bench" in text:
+        text = text.replace(
+            " bench ", " [bench](https://example.com/bench) ", 1
+        )
+    return text
+
+
+def make_png(rgb, size=8):
+    """A tiny, valid PNG -- flat colour, no external dependency."""
+
+    def chunk(tag, data):
+        c = tag + data
+        return struct.pack(">I", len(data)) + c + struct.pack(">I", zlib.crc32(c))
+
+    sig = b"\x89PNG\r\n\x1a\n"
+    ihdr = struct.pack(">IIBBBBB", size, size, 8, 2, 0, 0, 0)
+    raw = b""
+    for _ in range(size):
+        raw += b"\x00" + bytes(rgb) * size
+    idat = zlib.compress(raw)
+    return sig + chunk(b"IHDR", ihdr) + chunk(b"IDAT", idat) + chunk(b"IEND", b"")
+
+
+def main():
+    HERE.mkdir(exist_ok=True)
+    lines = []
+    seq = 1
+    ts = 1_788_000_000.0
+
+    def emit(type_, **fields):
+        nonlocal seq, ts
+        obj = {"seq": seq, "ts": round(ts, 3), "type": type_}
+        obj.update(fields)
+        lines.append(json.dumps(obj, separators=(",", ":")))
+        seq += 1
+        ts += random.uniform(0.05, 2.0)
+
+    emit("status", state="running")
+    emit("settings", model="bench-model", permissionMode="auto")
+
+    image_refs = []
+    turn = 0
+    while seq <= BACKLOG_COUNT:
+        turn += 1
+        emit("userMessage", text=f"Turn {turn}: {paragraph(12)}", id=None, attachments=[])
+
+        # A tool call with kilobyte-scale input/output every few turns.
+        if turn % 3 == 0:
+            tool_id = f"tool-{turn}"
+            big_input = json.dumps({"path": f"/repo/file_{turn}.rs", "content": paragraph(400)})
+            emit("toolStart", id=tool_id, tool="Edit", input=big_input)
+            big_output = "\n".join(paragraph(60) for _ in range(20))
+            emit("toolUpdate", id=tool_id, output=big_output[: len(big_output) // 2])
+            emit("toolEnd", id=tool_id, output=big_output)
+
+        # A reply: a heading, prose, a fenced block in a rotating language, a table, then deltas.
+        emit("assistantText", delta=f"{random.choice(HEADINGS)}\n\n")
+        emit("assistantText", delta=paragraph(30) + "\n\n")
+        lang = LANGUAGES[turn % len(LANGUAGES)]
+        emit("assistantText", delta=f"```{lang}\n{CODE_SNIPPETS[lang]}\n```\n\n")
+        if turn % 5 == 0:
+            emit(
+                "assistantText",
+                delta="| column | value |\n|---|---|\n| a | " + paragraph(3) + " |\n\n",
+            )
+        # A run of small deltas -- the shape a live reply actually streams in.
+        for _ in range(random.randint(3, 8)):
+            emit("assistantText", delta=paragraph(6) + " ")
+
+        # A couple of images, base64 PNGs, the way a real transcript embeds a screenshot.
+        if turn in (10, 40):
+            ref = f"bench{len(image_refs) + 1}.png"
+            image_refs.append(ref)
+            emit("image", ref=ref, about=None)
+
+        emit("usageDelta", tokens=random.randint(200, 4000), context=random.randint(2000, 180000))
+
+        if turn % 15 == 0:
+            emit(
+                "compacted",
+                preTokens=180000,
+                postTokens=20000,
+                trigger="auto",
+            )
+
+    # The streaming-phase tail: one long reply, built entirely from text deltas, the shape a
+    # bench harness replays at a fixed events/sec through the live fold path.
+    emit("userMessage", text="One more, streamed live for the benchmark's timing phase.", id=None, attachments=[])
+    while seq <= BACKLOG_COUNT + STREAM_COUNT:
+        emit("assistantText", delta=paragraph(5) + " ")
+    emit("status", state="idle")
+
+    (HERE / "transcript.jsonl").write_text("\n".join(lines) + "\n")
+
+    (HERE / "bench1.png").write_bytes(make_png((220, 90, 90)))
+    (HERE / "bench2.png").write_bytes(make_png((90, 150, 220)))
+
+    print(f"wrote {len(lines)} events ({BACKLOG_COUNT} backlog + {STREAM_COUNT} stream) to transcript.jsonl")
+
+
+if __name__ == "__main__":
+    main()

app/build-apk.sh

@@ -4,6 +4,11 @@
 #   ./build-apk.sh            the release build, signed (what the phone runs)
 #   ./build-apk.sh debug      the debug build, for reproducing something the
 #                             emulator scripts would build anyway
+#   ./build-apk.sh bench      P0's benchmark build (own app id, "AI Sessions
+#                             bench" label, opens straight onto the fixture
+#                             session -- see docs/RUST.md's P0 box and
+#                             app/bench-fixture/README.md). Signed the same
+#                             as release; never touches the CA it pins.
 #
 # Dev Updater's `.dev-updater.ron` at the checkout root spells these out as
 # build modes, one command line each; it passes nothing else, so the word
@@ -25,8 +30,9 @@ VARIANT=${1:-release}
 case "$VARIANT" in
 release) TASK=assembleRelease ;;
 debug) TASK=assembleDebug ;;
+bench) TASK=assembleBench ;;
 *)
-    echo "build-apk.sh: unknown variant '$VARIANT' (release, debug)" >&2
+    echo "build-apk.sh: unknown variant '$VARIANT' (release, debug, bench)" >&2
     exit 2
     ;;
 esac
@@ -81,7 +87,7 @@ fi
 # uninstalling it first: the signatures differ, and Android refuses to
 # update across them.
 KEYSTORE="${AI_APP_KEYSTORE:-${XDG_CONFIG_HOME:-$HOME/.config}/ai-app/release.jks}"
-if [ "$VARIANT" = release ] && [ ! -f "$KEYSTORE" ]; then
+if { [ "$VARIANT" = release ] || [ "$VARIANT" = bench ]; } && [ ! -f "$KEYSTORE" ]; then
     KEYTOOL="${JAVA_HOME:+$JAVA_HOME/bin/keytool}"
     KEYTOOL="${KEYTOOL:-keytool}"
     if ! command -v "$KEYTOOL" >/dev/null 2>&1; then
@@ -97,7 +103,7 @@ if [ "$VARIANT" = release ] && [ ! -f "$KEYSTORE" ]; then
         -keyalg RSA -keysize 2048 -validity 10000 \
         -storepass "$PASSWORD" -keypass "$PASSWORD" -dname "CN=ai-app" >/dev/null 2>&1)
 fi
-if [ "$VARIANT" = release ]; then
+if [ "$VARIANT" = release ] || [ "$VARIANT" = bench ]; then
     AI_APP_KEYSTORE="$KEYSTORE"
     AI_APP_KEYSTORE_PASSWORD=$(cat "$KEYSTORE.password")
     export AI_APP_KEYSTORE AI_APP_KEYSTORE_PASSWORD

app/iris-scroll.sh

@@ -0,0 +1,34 @@
+#!/bin/sh
+# RUST.md's I5 "Where iris's frame time goes" pass (2026-09-05). The same
+# 24-swipe/6-cycle loop as transcript-bench.sh's, extracted for iris's own
+# demo app -- transcript-bench.sh itself is Compose-specific (opens by
+# session title through the Compose app's own UI) and cannot be called
+# directly against dev.iris.android.demo.
+#
+# MUST be run from inside this checkout (not /tmp): ui-trace/adb pick which
+# emulator to target from the current directory's basename (the
+# per-checkout-AVD rule), and a previous pass lost two attempts to a `cd`
+# into /tmp that made this resolve to a nonexistent "tmp" checkout.
+set -eu
+cd "$(dirname "$0")"
+. ./android-env.sh >/dev/null 2>&1
+
+cycles=${1:-6}
+
+ui-trace record -d 3000 --do "tap 'Reset frame report'" -o /tmp/iris-bench-reset.txt >/dev/null
+adb logcat -c
+
+DO=""
+i=0
+while [ "$i" -lt "$cycles" ]; do
+    DO="$DO --do 'swipe 540 700 540 1600 200' --do 'wait 500'"
+    DO="$DO --do 'swipe 540 700 540 1600 200' --do 'wait 500'"
+    DO="$DO --do 'swipe 540 1600 540 700 200' --do 'wait 500'"
+    DO="$DO --do 'swipe 540 1600 540 700 200' --do 'wait 500'"
+    i=$((i + 1))
+done
+eval ui-trace record -d $((cycles * 16000 + 20000)) $DO -o /tmp/iris-bench-scroll.txt >/dev/null
+
+ui-trace record -d 3000 --do "tap 'Frame report'" -o /tmp/iris-bench-report.txt >/dev/null
+sleep 1
+adb logcat -d -s iris-android-app:I | grep "iris frame report:"

app/settings.gradle.kts

@@ -17,6 +17,11 @@ dependencyResolutionManagement {
 
 include(":androidApp")
 
+// E3 (RUST.md): the Kotlin/Java shell over android-shell's JNI bridge, a
+// separate module from :androidApp so the ~13,000 lines of working Compose
+// UI there are untouched. See shellApp/build.gradle.kts's module comment.
+include(":shellApp")
+
 // The app half of wg-app-link, resolved by path through the submodule so
 // this checkout and the crate it consumes move together -- the same
 // arrangement `server/` uses for the Rust half. See that repo's README.

app/shellApp/build.gradle.kts

@@ -0,0 +1,163 @@
+plugins { alias(libs.plugins.androidApplication) }
+
+// E3 (RUST.md): the Kotlin/Java shell being replaced by a thin JNI bridge
+// into Rust (`../../android-shell`). Deliberately its own module rather
+// than a rewrite of `:androidApp` in place -- that module is ~13,000 lines
+// of working Compose UI this experiment does not touch, and the two can be
+// installed side by side on the same development device (see
+// `settings.SCHEME`'s doc in `android-shell` for why the deep-link scheme
+// and Keystore alias are not the production app's). No Compose plugin, no
+// Kotlin source of its own: `MainActivity`/`NotificationService` are plain
+// Java, and the CA constant below is generated as Java too.
+//
+// The CA this build pins is baked in the same way `androidApp`'s does --
+// see that module's `build.gradle.kts` comment for the reasoning (the
+// trust boundary follows the machine that builds, never a pasted copy).
+// `PinnedCa.java`'s package must match `android-shell`'s
+// `settings::load_pinned_ca` lookup (`com/example/aiapp/shell/PinnedCa`).
+val pinnedCaPath: String =
+    System.getenv("AI_APP_CA")
+        ?: "${System.getenv("XDG_CONFIG_HOME") ?: "${System.getProperty("user.home")}/.config"}" +
+            "/ai-app/certs/ca.pem"
+
+abstract class GeneratePinnedCa : DefaultTask() {
+    @get:Input abstract val caPath: Property<String>
+
+    @get:InputFile
+    @get:Optional
+    @get:PathSensitive(PathSensitivity.NONE)
+    abstract val caCertificate: RegularFileProperty
+
+    @get:OutputDirectory abstract val outputDir: DirectoryProperty
+
+    @TaskAction
+    fun generate() {
+        val path = caPath.get()
+        val ca = File(path)
+        if (!ca.isFile) {
+            throw GradleException(
+                "No CA certificate at $path.\n" +
+                    "Start ai-server (or app/ui-sandbox.sh) once on this machine first -- it " +
+                    "generates the CA this build pins.\n" +
+                    "Set AI_APP_CA=/path/to/ca.pem to build against a different one."
+            )
+        }
+        val pem = ca.readText().trim()
+        if (!pem.startsWith("-----BEGIN CERTIFICATE-----")) {
+            throw GradleException("$path is not a PEM certificate.")
+        }
+        val dir = outputDir.get().dir("com/example/aiapp/shell").asFile
+        dir.mkdirs()
+        // Same reasoning as androidApp's generatePinnedCert: the text block
+        // must start immediately after the opening `"""`, or
+        // CertificateFactory stops recognising the "-----BEGIN" preamble.
+        File(dir, "PinnedCa.java")
+            .writeText(
+                """
+                |// Generated from $path by the generatePinnedCa task. Do not edit.
+                |package com.example.aiapp.shell;
+                |
+                |public final class PinnedCa {
+                |    private PinnedCa() {}
+                |    public static final String PINNED_CA_PEM = ""${'"'}
+                |$pem""${'"'};
+                |}
+                |"""
+                    .trimMargin()
+            )
+    }
+}
+
+val generatePinnedCa =
+    tasks.register<GeneratePinnedCa>("generatePinnedCa") {
+        val ca = file(pinnedCaPath)
+        caPath.set(pinnedCaPath)
+        if (ca.isFile) {
+            caCertificate.set(ca)
+        }
+    }
+
+android {
+    namespace = "com.example.aiapp.shell"
+    compileSdk = 37
+
+    defaultConfig {
+        applicationId = "com.example.aiapp.shell"
+        minSdk = 24
+        targetSdk = 37
+        versionCode = 1
+        versionName = "1.0"
+    }
+    // Same reasoning and same key as androidApp's (see that module's comment): E5 (RUST.md)
+    // signs its own, Gradle-free build with this same keystore, and the two can only
+    // `adb install -r` over each other if they carry the same certificate.
+    val keystore = System.getenv("AI_APP_KEYSTORE")
+    signingConfigs {
+        if (keystore != null) {
+            create("release") {
+                storeFile = file(keystore)
+                storePassword = System.getenv("AI_APP_KEYSTORE_PASSWORD")
+                keyAlias = "ai-app"
+                keyPassword = storePassword
+            }
+        }
+    }
+    buildTypes {
+        getByName("release") {
+            isMinifyEnabled = false
+            if (keystore != null) signingConfig = signingConfigs.getByName("release")
+        }
+    }
+    compileOptions {
+        sourceCompatibility = JavaVersion.VERSION_21
+        targetCompatibility = JavaVersion.VERSION_21
+    }
+}
+
+// E5 (RUST.md): the xtask dexes and packages this module's Java sources itself, but it does
+// not resolve Maven dependencies -- reimplementing a dependency resolver was out of scope for a
+// packaging step, so this one task is the single place Gradle still runs in that pipeline. It
+// asks the dependency graph for the *post-transform* jars (AARs already unpacked to a classes
+// jar, the same artifact type AGP's own dexing task consumes) rather than the raw configuration,
+// which would hand back .aar files d8 cannot read directly.
+val artifactType = Attribute.of("artifactType", String::class.java)
+
+tasks.register("printRuntimeClasspathJars") {
+    description = "Writes the resolved release runtime classpath jars, one per line, for xtask."
+    val outputFile = layout.buildDirectory.file("xtask/runtime-classpath.txt")
+    outputs.file(outputFile)
+    val jars =
+        configurations
+            .getByName("releaseRuntimeClasspath")
+            .incoming
+            .artifactView { attributes.attribute(artifactType, "android-classes-jar") }
+            .files
+    // Captured as a plain FileCollection (not the ArtifactView itself, which the
+    // configuration cache cannot serialize) so this task is still cacheable.
+    inputs.files(jars)
+    doLast {
+        val file = outputFile.get().asFile
+        file.parentFile.mkdirs()
+        file.writeText(jars.joinToString("\n") { it.absolutePath })
+    }
+}
+
+androidComponents {
+    onVariants { variant ->
+        variant.sources.java?.addGeneratedSourceDirectory(generatePinnedCa, GeneratePinnedCa::outputDir)
+    }
+}
+
+dependencies {
+    // The Keystore-sealed enrollment (ServerStore/ServerSettings) --
+    // android-shell's settings.rs calls into this Kotlin class directly
+    // over JNI rather than re-sealing the token in Rust; see that file's
+    // module doc.
+    implementation(project(":link"))
+    // NotificationCompat/NotificationManagerCompat/NotificationChannelCompat/
+    // ServiceCompat -- android-shell's notify.rs calls these classes over
+    // JNI so the pre-26 fallback behaviour (no channels) lives once, in
+    // the library that already has it, rather than being re-derived as a
+    // set of Build.VERSION.SDK_INT branches in Rust.
+    implementation(libs.androidx.core.ktx)
+}

app/shellApp/src/main/AndroidManifest.xml

@@ -0,0 +1,62 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    xmlns:tools="http://schemas.android.com/tools">
+    <!-- Mirrors androidApp's manifest (AGENTS.md: reuse it rather than
+         re-deriving it) for the permissions and declarations E3 actually
+         exercises. Not carried over: the QR scanner activity (this
+         experiment enrolls via the aiappshell://enroll deep link directly,
+         per AGENTS.md's ui-sandbox.sh banner) and the app icon warning
+         suppression below, for the same reason androidApp's is there. -->
+    <uses-permission android:name="android.permission.INTERNET" />
+    <uses-permission android:name="android.permission.ACCESS_LOCAL_NETWORK" />
+    <uses-permission android:name="android.permission.POST_NOTIFICATIONS" />
+    <uses-permission android:name="android.permission.FOREGROUND_SERVICE" />
+    <uses-permission android:name="android.permission.FOREGROUND_SERVICE_SPECIAL_USE" />
+
+    <application
+        android:label="AI Sessions (shell)"
+        android:allowBackup="true"
+        android:theme="@android:style/Theme.Material.Light.NoActionBar"
+        tools:ignore="MissingApplicationIcon">
+        <activity
+            android:name=".MainActivity"
+            android:exported="true"
+            android:launchMode="singleTop">
+            <intent-filter>
+                <action android:name="android.intent.action.MAIN" />
+                <category android:name="android.intent.category.LAUNCHER" />
+            </intent-filter>
+            <!-- Enrollment: aiappshell://enroll?host=...&port=...&token=...,
+                 per AGENTS.md's ui-sandbox.sh banner (fed to this app with
+                 `adb shell am start -a android.intent.action.VIEW -d
+                 'aiappshell://enroll?...'`, or -n'd at this component
+                 directly if a second app also claims the aiapp scheme). -->
+            <intent-filter>
+                <action android:name="android.intent.action.VIEW" />
+                <category android:name="android.intent.category.DEFAULT" />
+                <category android:name="android.intent.category.BROWSABLE" />
+                <data android:scheme="aiappshell" android:host="enroll" />
+            </intent-filter>
+            <!-- The share sheet - see android-shell's share.rs. -->
+            <intent-filter>
+                <action android:name="android.intent.action.SEND" />
+                <action android:name="android.intent.action.SEND_MULTIPLE" />
+                <category android:name="android.intent.category.DEFAULT" />
+                <data android:mimeType="*/*" />
+            </intent-filter>
+        </activity>
+
+        <!-- specialUse, not dataSync, for the reason androidApp's manifest
+             gives: a connection that has to keep listening overnight
+             cannot accept dataSync's six-hour cap. -->
+        <service
+            android:name=".NotificationService"
+            android:exported="false"
+            android:foregroundServiceType="specialUse">
+            <property
+                android:name="android.app.PROPERTY_SPECIAL_USE_FGS_SUBTYPE"
+                android:value="E3 experiment: holds one connection to the sandbox server so a
+                    session that needs an answer can be reported while the app is closed." />
+        </service>
+    </application>
+</manifest>

app/shellApp/src/main/java/com/example/aiapp/shell/MainActivity.java

@@ -0,0 +1,50 @@
+package com.example.aiapp.shell;
+
+import android.app.Activity;
+import android.content.Context;
+import android.content.Intent;
+import android.os.Bundle;
+import android.os.Handler;
+import android.os.Looper;
+import android.widget.Toast;
+
+/**
+ * E3's floor, per RUST.md's "How much Java is unavoidable": a class the framework
+ * constructs by name from the manifest, with its lifecycle methods handing straight to Rust
+ * (android-shell's {@code share::handle_intent}). No Compose, no layout -- there is no screen to
+ * draw yet (that is E4's job, on iris); {@link #toast} is this experiment's stand-in for showing
+ * something happened.
+ */
+public class MainActivity extends Activity {
+    static {
+        System.loadLibrary("android_shell");
+    }
+
+    @Override
+    protected void onCreate(Bundle savedInstanceState) {
+        super.onCreate(savedInstanceState);
+        NotificationService.sync(this);
+        nativeHandleIntent(this, getIntent());
+    }
+
+    // launchMode="singleTop": a notification tap or a share while this activity is already on
+    // top lands here rather than in a second instance -- same reasoning as MainActivity.kt's.
+    @Override
+    protected void onNewIntent(Intent intent) {
+        super.onNewIntent(intent);
+        setIntent(intent);
+        nativeHandleIntent(this, intent);
+    }
+
+    /**
+     * Called from android-shell, sometimes from a background thread (a share's network call is
+     * never made on the calling thread -- see share.rs). {@code Toast} itself is main-thread-only,
+     * so this hops there with a {@link Handler} rather than assuming the caller already has.
+     */
+    static void toast(Context context, String message) {
+        new Handler(Looper.getMainLooper())
+                .post(() -> Toast.makeText(context, message, Toast.LENGTH_LONG).show());
+    }
+
+    private static native void nativeHandleIntent(Activity activity, Intent intent);
+}

app/shellApp/src/main/java/com/example/aiapp/shell/NotificationService.java

@@ -0,0 +1,45 @@
+package com.example.aiapp.shell;
+
+import android.app.Service;
+import android.content.Context;
+import android.content.Intent;
+import android.os.IBinder;
+
+/**
+ * E3's second unavoidable Java class (RUST.md): a foreground service constructed by the framework
+ * from the manifest, existing only to hand its lifecycle to android-shell's {@code notify} module
+ * -- the SSE follow loop, deciding what a notification says, and posting it are all Rust reached
+ * through these three native calls. See {@code Notifications.kt}'s {@code NotificationService} for
+ * the Kotlin original this mirrors.
+ */
+public class NotificationService extends Service {
+    static {
+        System.loadLibrary("android_shell");
+    }
+
+    @Override
+    public IBinder onBind(Intent intent) {
+        return null;
+    }
+
+    @Override
+    public int onStartCommand(Intent intent, int flags, int startId) {
+        return nativeOnStartCommand(this);
+    }
+
+    @Override
+    public void onDestroy() {
+        nativeOnDestroy();
+    }
+
+    /** Starts this service if there is a server to connect to, and stops it otherwise. */
+    static void sync(Context context) {
+        nativeSync(context);
+    }
+
+    private static native void nativeSync(Context context);
+
+    private static native int nativeOnStartCommand(Service service);
+
+    private static native void nativeOnDestroy();
+}

client-core/Cargo.lock

@@ -47,6 +47,7 @@ name = "client-core"
 version = "0.1.0"
 dependencies = [
  "event-model",
+ "pulldown-cmark",
  "serde",
  "serde_json",
  "tempfile",
@@ -173,6 +174,15 @@ dependencies = [
  "percent-encoding",
 ]
 
+[[package]]
+name = "getopts"
+version = "0.2.24"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cfe4fbac503b8d1f88e6676011885f34b7174f46e59956bba534ba83abded4df"
+dependencies = [
+ "unicode-width",
+]
+
 [[package]]
 name = "getrandom"
 version = "0.2.17"
@@ -425,6 +435,25 @@ dependencies = [
  "unicode-ident",
 ]
 
+[[package]]
+name = "pulldown-cmark"
+version = "0.13.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e9f068eba8e7071c5f9511831b44f32c740d5adf574e990f946ddb53db2f314e"
+dependencies = [
+ "bitflags",
+ "getopts",
+ "memchr",
+ "pulldown-cmark-escape",
+ "unicase",
+]
+
+[[package]]
+name = "pulldown-cmark-escape"
+version = "0.11.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "007d8adb5ddab6f8e3f491ac63566a7d5002cc7ed73901f72057943fa71ae1ae"
+
 [[package]]
 name = "quote"
 version = "1.0.47"
@@ -661,12 +690,24 @@ dependencies = [
  "zerovec",
 ]
 
+[[package]]
+name = "unicase"
+version = "2.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dbc4bc3a9f746d862c45cb89d705aa10f187bb96c76001afab07a0d35ce60142"
+
 [[package]]
 name = "unicode-ident"
 version = "1.0.24"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e6e4313cd5fcd3dad5cafa179702e2b244f760991f45397d14d4ebf38247da75"
 
+[[package]]
+name = "unicode-width"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b4ac048d71ede7ee76d585517add45da530660ef4390e49b098733c6e897f254"
+
 [[package]]
 name = "untrusted"
 version = "0.9.0"

client-core/Cargo.toml

@@ -7,7 +7,7 @@ edition = "2024"
 # with `server/` via `event-model`), the REST + SSE clients for its HTTP
 # surface (see `server/src/routes.rs`'s module doc for the table), the
 # transcript fold and cache, the markdown block model, the syntax
-# highlighter and the ANSI parser. See CLIENT_CORE.md at the repo root for
+# highlighter and the ANSI parser. See `docs/CLIENT_CORE.md` for
 # what this holds today, what it does not yet, and how it corresponds to
 # the Kotlin it replaces.
 #
@@ -17,7 +17,12 @@ edition = "2024"
 [dependencies]
 event-model = { path = "../event-model" }
 serde = { version = "1", features = ["derive"] }
-serde_json = { version = "1", features = ["float_roundtrip"] }
+# "raw_value" is `fetch_transcript_lines`'s reason -- it needs the exact
+# bytes the server sent, not this crate's own re-serialization of a parsed
+# `Value`, so a cached line and a live SSE frame for the same event agree
+# byte-for-byte (see that method's doc). "float_roundtrip" is why they
+# agree on a `ts` at all -- see server/Cargo.toml's identical comment.
+serde_json = { version = "1", features = ["float_roundtrip", "raw_value"] }
 # The blocking HTTP client for the REST calls and the long-lived SSE GETs.
 # `server/` already depends on ureq for its own outbound HTTPS (the usage
 # poll in usage.rs) and it is rustls-backed like the rest of this project's
@@ -27,6 +32,12 @@ serde_json = { version = "1", features = ["float_roundtrip"] }
 # no need of an async runtime, and RUST.md's brief for this port is
 # "lightweight" throughout.
 ureq = { version = "3", features = ["json"] }
+# The markdown block split (`markdown_blocks`), which has to agree with the
+# renderer in `iris/transcript-ui` about where a block begins -- so it is
+# the same parser at the same version, rather than a hand-written splitter
+# that would drift from it.
+pulldown-cmark = "0.13.4"
+
 
 [dev-dependencies]
 tempfile = "3"

client-core/src/api.rs

@@ -10,6 +10,7 @@
 
 use std::io::Read;
 
+use event_model::SeqEvent;
 use serde::Deserialize;
 use serde_json::Value;
 
@@ -116,6 +117,14 @@ impl<T: Transport> ApiClient<T> {
         Self { transport }
     }
 
+    /// The transport underneath, for a caller that needs the raw SSE
+    /// stream (`event_stream::follow_session_events`) rather than one of
+    /// this client's typed REST calls -- `transcript_source::TranscriptSource`
+    /// is the one that does.
+    pub fn transport(&self) -> &T {
+        &self.transport
+    }
+
     fn json_request<R: for<'de> Deserialize<'de>>(
         &self,
         method: &str,
@@ -266,6 +275,61 @@ impl<T: Transport> ApiClient<T> {
         limit: u32,
         coalesce: bool,
     ) -> Result<Vec<Value>, ApiError> {
+        self.json_request(
+            "GET",
+            &transcript_path(session_id, before, limit, coalesce, None),
+            None,
+        )
+    }
+
+    /// A page of transcript history, each line handed back paired with the
+    /// exact text it came from, and bounded below by `after` -- the shape
+    /// `crate::transcript_source::TranscriptSource` needs to store what it
+    /// fetched in the transcript cache without a second round trip to fetch
+    /// the raw text separately. Ported from `Api.kt`'s `fetchTranscript`.
+    ///
+    /// Uses [`serde_json::value::RawValue`] rather than re-serializing a
+    /// parsed [`Value`], so the stored line is the exact bytes the server
+    /// sent (key order and float literal included) rather than this
+    /// crate's own idea of how to write them back out -- the cache and a
+    /// live SSE frame must agree byte-for-byte on the same event, which is
+    /// exactly what caught the `serde_json` float-rounding bug this
+    /// project's `AGENTS.md` records.
+    pub fn fetch_transcript_lines(
+        &self,
+        session_id: &str,
+        before: Option<u64>,
+        limit: u32,
+        coalesce: bool,
+        after: Option<u64>,
+    ) -> Result<Vec<(String, SeqEvent)>, ApiError> {
+        let path = transcript_path(session_id, before, limit, coalesce, after);
+        let raw: Vec<Box<serde_json::value::RawValue>> = self.json_request("GET", &path, None)?;
+        raw.into_iter()
+            .map(|value| {
+                let line = value.get().to_string();
+                let event: SeqEvent = serde_json::from_str(&line).map_err(|e| ApiError {
+                    message: format!(
+                        "the server sent a transcript line this build couldn't parse: {e}"
+                    ),
+                    status: None,
+                })?;
+                Ok((line, event))
+            })
+            .collect()
+    }
+}
+
+/// The query string shared by [`ApiClient::fetch_transcript_page`] and
+/// [`ApiClient::fetch_transcript_lines`], so the two agree on how each
+/// parameter is written rather than keeping two copies to drift.
+fn transcript_path(
+    session_id: &str,
+    before: Option<u64>,
+    limit: u32,
+    coalesce: bool,
+    after: Option<u64>,
+) -> String {
     let mut path = format!("/sessions/{session_id}/transcript?limit={limit}");
     if let Some(before) = before {
         path.push_str(&format!("&before={before}"));
@@ -273,8 +337,10 @@ impl<T: Transport> ApiClient<T> {
     if coalesce {
         path.push_str("&coalesce=true");
     }
-        self.json_request("GET", &path, None)
+    if let Some(after) = after {
+        path.push_str(&format!("&after={after}"));
     }
+    path
 }
 
 /// The blocking [`Transport`] backed by `ureq`, the same crate `server/`

client-core/src/config.rs

@@ -0,0 +1,152 @@
+//! What a Rust client needs to reach one enrolled server: host, port and
+//! bearer token. Mirrors the shape `ServerConfig.kt`/`Api.kt`'s
+//! `handleEnrollment` parses out of an `aiapp://enroll?host=H&port=P&token=T`
+//! deep link -- the exact link `wg-app-link`'s `enroll` module mints and
+//! `app/ui-sandbox.sh`'s banner prints, so any Rust client can enrol from
+//! the same text a phone would scan as a QR, with no second format
+//! invented for it (RUST.md's E4).
+//!
+//! What this type deliberately does not decide: where it is persisted, and
+//! under what file permissions. A phone seals its token in the Android
+//! Keystore; a desktop client has its own `$XDG_CONFIG_HOME/<app>/`
+//! directory and its own file-mode conventions (MACHINE.md: owner-only,
+//! never in the repo). Both are caller-specific, so they stay out of this
+//! crate per the code rules' "ask for the least you need" -- see
+//! `iris/desktop-app/src/config.rs` for the desktop instance.
+
+use serde::{Deserialize, Serialize};
+
+/// One enrolled server: reachable at `https://{host}:{port}`, authenticated
+/// with `token` as a bearer header. Does not carry the pinned CA -- that is
+/// a public certificate rather than a secret, and where to find it differs
+/// by caller (a phone pins the one its APK was built against; a desktop
+/// client is told a path).
+#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
+pub struct EnrolledServer {
+    pub host: String,
+    pub port: u16,
+    pub token: String,
+}
+
+impl EnrolledServer {
+    /// Parses `aiapp://enroll?host=H&port=P&token=T` (query order does not
+    /// matter; unrecognised keys are ignored). `token` is percent-decoded,
+    /// since `ui-sandbox.sh` encodes it precisely because a raw token can
+    /// contain `+`, which turns into a space if left to a naive splitter.
+    pub fn parse_link(link: &str) -> Result<Self, String> {
+        let query = link.split_once('?').map(|(_, q)| q).ok_or_else(|| {
+            format!(
+                "'{link}' has no query string (expected \
+                 aiapp://enroll?host=...&port=...&token=...)"
+            )
+        })?;
+
+        let mut host = None;
+        let mut port = None;
+        let mut token = None;
+        for pair in query.split('&') {
+            let Some((key, value)) = pair.split_once('=') else {
+                continue;
+            };
+            let value = percent_decode(value);
+            match key {
+                "host" => host = Some(value),
+                "port" => port = Some(value),
+                "token" => token = Some(value),
+                _ => {}
+            }
+        }
+
+        let host = host.ok_or_else(|| format!("'{link}' is missing 'host'"))?;
+        let port_str = port.ok_or_else(|| format!("'{link}' is missing 'port'"))?;
+        let port: u16 = port_str
+            .parse()
+            .map_err(|e| format!("'{link}''s port ('{port_str}') is not a number: {e}"))?;
+        let token = token.ok_or_else(|| format!("'{link}' is missing 'token'"))?;
+
+        Ok(Self { host, port, token })
+    }
+
+    /// Where a `client_core::api::UreqTransport` reaches this server.
+    pub fn base_url(&self) -> String {
+        format!("https://{}:{}", self.host, self.port)
+    }
+}
+
+fn percent_decode(s: &str) -> String {
+    let bytes = s.as_bytes();
+    let mut out = Vec::with_capacity(bytes.len());
+    let mut i = 0;
+    while i < bytes.len() {
+        if bytes[i] == b'%'
+            && i + 2 < bytes.len()
+            && let Ok(byte) =
+                u8::from_str_radix(std::str::from_utf8(&bytes[i + 1..i + 3]).unwrap_or(""), 16)
+        {
+            out.push(byte);
+            i += 3;
+            continue;
+        }
+        out.push(bytes[i]);
+        i += 1;
+    }
+    String::from_utf8_lossy(&out).into_owned()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn parses_host_port_and_token() {
+        let server =
+            EnrolledServer::parse_link("aiapp://enroll?host=127.0.0.1&port=8547&token=abcDEF123")
+                .unwrap();
+        assert_eq!(
+            server,
+            EnrolledServer {
+                host: "127.0.0.1".to_string(),
+                port: 8547,
+                token: "abcDEF123".to_string(),
+            }
+        );
+        assert_eq!(server.base_url(), "https://127.0.0.1:8547");
+    }
+
+    #[test]
+    fn field_order_does_not_matter() {
+        let server =
+            EnrolledServer::parse_link("aiapp://enroll?token=tok&port=443&host=example.com")
+                .unwrap();
+        assert_eq!(server.host, "example.com");
+        assert_eq!(server.port, 443);
+        assert_eq!(server.token, "tok");
+    }
+
+    #[test]
+    fn a_percent_encoded_token_is_decoded() {
+        // ui-sandbox.sh's own reason for encoding: a raw '+' would
+        // otherwise arrive as a space.
+        let server =
+            EnrolledServer::parse_link("aiapp://enroll?host=h&port=1&token=a%2Bb%2Fc").unwrap();
+        assert_eq!(server.token, "a+b/c");
+    }
+
+    #[test]
+    fn a_missing_field_is_named_in_the_error() {
+        let err = EnrolledServer::parse_link("aiapp://enroll?host=h&port=1").unwrap_err();
+        assert!(
+            err.contains("token"),
+            "error should name the missing field: {err}"
+        );
+    }
+
+    #[test]
+    fn a_non_numeric_port_is_named_in_the_error() {
+        let err = EnrolledServer::parse_link("aiapp://enroll?host=h&port=x&token=t").unwrap_err();
+        assert!(
+            err.contains("port"),
+            "error should name the offending field: {err}"
+        );
+    }
+}

client-core/src/durations.rs

@@ -0,0 +1,100 @@
+//! A span of milliseconds, written the way somebody reads it -- the port
+//! of `Durations.kt`'s `formatMillis`/`formatMillisText`, with its tests.
+//!
+//! Only the tool-timeout half is here. `formatSpan` (the usage
+//! countdown's rounding-up rule) belongs with whatever draws the usage
+//! bar, and nothing in this crate needs it yet.
+
+/// A span of milliseconds, written the way somebody reads it.
+///
+/// A tool's timeout arrives as `480000`, which nobody reads as eight
+/// minutes. The rule has two halves, because a short span and a long one
+/// are read for different things. Under a minute the question is "roughly
+/// how long", so only the largest unit is shown and a fraction carries the
+/// rest -- `2.5s`. At a minute or more the question is "how long exactly",
+/// so every unit with something in it is written out -- `5d 12h 4m`. Empty
+/// units are left out rather than written as zero.
+///
+/// Sub-second precision is dropped past a minute: nothing that takes days
+/// is measured in milliseconds.
+pub fn format_millis(ms: i64) -> String {
+    if ms < 0 {
+        return format!("-{}", format_millis(-ms));
+    }
+    if ms < 1000 {
+        return format!("{ms}ms");
+    }
+    if ms < 60_000 {
+        let tenths = (ms + 50) / 100;
+        let (whole, rest) = (tenths / 10, tenths % 10);
+        return if rest == 0 {
+            format!("{whole}s")
+        } else {
+            format!("{whole}.{rest}s")
+        };
+    }
+    let seconds = ms / 1000;
+    [
+        ("d", seconds / 86_400),
+        ("h", seconds / 3600 % 24),
+        ("m", seconds / 60 % 60),
+        ("s", seconds % 60),
+    ]
+    .iter()
+    .filter(|(_, n)| *n > 0)
+    .map(|(unit, n)| format!("{n}{unit}"))
+    .collect::<Vec<_>>()
+    .join(" ")
+}
+
+/// `text` as a span when it is a whole number of milliseconds, and
+/// unchanged when it is not.
+pub fn format_millis_text(text: &str) -> String {
+    match text.trim().parse::<i64>() {
+        Ok(ms) => format_millis(ms),
+        Err(_) => text.to_string(),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// The two ways a span of time is written here, and the rule each of
+    /// them follows -- ported from `DurationsTest.kt`, whose doc says why:
+    /// both are read off a screen to make a decision, so what matters is
+    /// that the shortest form that answers the question is what appears.
+    #[test]
+    fn under_a_minute_is_the_largest_unit_alone() {
+        assert_eq!(format_millis(30), "30ms");
+        assert_eq!(format_millis(999), "999ms");
+        assert_eq!(format_millis(1000), "1s");
+        assert_eq!(format_millis(2500), "2.5s");
+        // One decimal, rounded rather than cut: 2.46s is nearer two and a
+        // half than two and four.
+        assert_eq!(format_millis(2460), "2.5s");
+        assert_eq!(format_millis(59_900), "59.9s");
+    }
+
+    #[test]
+    fn a_minute_or_more_is_every_unit_that_has_something_in_it() {
+        // The figure this rule was written for: a tool timeout, which
+        // arrives as milliseconds and is unreadable as 480000.
+        assert_eq!(format_millis(480_000), "8m");
+        assert_eq!(format_millis(60_000), "1m");
+        assert_eq!(format_millis(90_000), "1m 30s");
+        assert_eq!(format_millis(475_440_000), "5d 12h 4m");
+        // Empty units are left out rather than written as zero: the labels
+        // say which is which, and "5d 0h 4m" is only longer.
+        assert_eq!(format_millis(432_240_000), "5d 4m");
+    }
+
+    #[test]
+    fn only_a_whole_number_of_milliseconds_is_rewritten() {
+        assert_eq!(format_millis_text(" 480000 "), "8m");
+        // A timeout a tool expressed some other way is its own words,
+        // passed through rather than guessed at.
+        assert_eq!(format_millis_text("2 minutes"), "2 minutes");
+        assert_eq!(format_millis_text(""), "");
+    }
+}

client-core/src/lib.rs

@@ -1,13 +1,19 @@
 //! The app's pure logic, shared between the server and any Rust client --
-//! see `CLIENT_CORE.md` at the repo root for what lives here and what does
+//! see `docs/CLIENT_CORE.md` for what lives here and what does
 //! not yet.
 
 pub mod ansi;
 pub mod api;
+pub mod config;
+pub mod durations;
 pub mod event_stream;
 pub mod highlight;
+pub mod markdown_blocks;
+pub mod notifications;
 pub mod sse;
+pub mod tool_summary;
 pub mod transcript_cache;
 pub mod transcript_fold;
+pub mod transcript_source;
 
 pub use event_model::*;

client-core/src/markdown_blocks.rs

@@ -0,0 +1,325 @@
+//! Split a markdown message into its top-level **blocks** -- one
+//! paragraph, heading, fenced code block, list, table or quote each, as a
+//! byte slice of the original source.
+//!
+//! This exists for streaming. A transcript row used to be one text widget
+//! holding the whole message, so a single streamed delta re-shaped every
+//! paragraph of it through the text engine again; the phone's bench v2 put
+//! the stream phase at p50 18.2ms against Compose's 13.4ms for exactly
+//! that reason (docs/IRIS_TODO.md). A row is a column of one widget per
+//! block now, and a delta that lands in the last block leaves every
+//! earlier block's layout alone. `docs/DECISIONS.md`'s 2026-09-06 entry has
+//! what that rejected and why the split lives here rather than in the UI
+//! crate: `docs/CLIENT_CORE.md` already wanted a block model for P1, and
+//! keeping it here means iris stays a text renderer that knows nothing
+//! about markdown.
+//!
+//! **Blocks only.** Inline styling (bold, links, inline code) is still the
+//! renderer's own job, per block -- this deliberately does not build a
+//! full AST, because nothing needs one yet.
+//!
+//! ## Appending is not guaranteed to leave earlier blocks alone
+//!
+//! It nearly always does, which is what makes the fast path worth having,
+//! but markdown has no such rule: appending a "```" line can turn text
+//! that was three paragraphs into one fenced block, and appending "---"
+//! under a paragraph turns that paragraph into a heading. So a caller
+//! taking the O(last block) path **must compare the prefix it is about to
+//! keep** rather than assume it. [`common_prefix`] is that comparison, and
+//! it is cheap next to laying the text out again.
+
+use pulldown_cmark::{Event, Options, Parser, Tag};
+
+/// What a block is, for a renderer that wants to style or space blocks
+/// differently. `Other` is deliberately present rather than a panic or a
+/// silent fallback to `Paragraph`: markdown has more block kinds than this
+/// list and more get added, and a renderer treating an unknown one as
+/// prose is right, but it should be able to *tell* that is what it is
+/// doing.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum BlockKind {
+    Paragraph,
+    Heading,
+    /// A fenced or indented code block.
+    Code,
+    List,
+    Table,
+    Quote,
+    /// A thematic break, raw HTML, a footnote -- anything with no
+    /// distinguished treatment here.
+    Other,
+}
+
+/// One top-level block: its kind and the exact source that produced it.
+/// `source` is a slice of the input with trailing whitespace removed, so
+/// two splits of the same prefix compare equal even when one of them had a
+/// delta arriving after it.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct Block {
+    pub kind: BlockKind,
+    pub source: String,
+}
+
+fn kind_of(tag: &Tag) -> BlockKind {
+    match tag {
+        Tag::Paragraph => BlockKind::Paragraph,
+        Tag::Heading { .. } => BlockKind::Heading,
+        Tag::CodeBlock(_) => BlockKind::Code,
+        Tag::List(_) => BlockKind::List,
+        Tag::Table(_) => BlockKind::Table,
+        Tag::BlockQuote(_) => BlockKind::Quote,
+        _ => BlockKind::Other,
+    }
+}
+
+fn options() -> Options {
+    // The same set `transcript-ui`'s renderer parses with, so a block
+    // boundary here and the styling there cannot disagree about what the
+    // source means.
+    Options::ENABLE_STRIKETHROUGH | Options::ENABLE_TABLES | Options::ENABLE_TASKLISTS
+}
+
+/// Split `src` into its top-level blocks, in source order. An empty or
+/// whitespace-only input gives no blocks; text the parser does not put
+/// inside any block (a stray fence marker mid-stream) still comes back,
+/// as `Other`, rather than being dropped.
+pub fn split_blocks(src: &str) -> Vec<Block> {
+    let mut out: Vec<Block> = Vec::new();
+    let mut depth = 0usize;
+    let mut kind = BlockKind::Other;
+    for (event, range) in Parser::new_ext(src, options()).into_offset_iter() {
+        match event {
+            Event::Start(tag) => {
+                if depth == 0 {
+                    kind = kind_of(&tag);
+                }
+                depth += 1;
+            }
+            Event::End(_) => {
+                depth -= 1;
+                if depth == 0 {
+                    push(&mut out, kind, &src[range]);
+                }
+            }
+            // A top-level event that is not part of any block -- a
+            // thematic break, a block of raw HTML. Inside one, it is the
+            // enclosing block's business and this does nothing.
+            _ => {
+                if depth == 0 {
+                    push(&mut out, BlockKind::Other, &src[range]);
+                }
+            }
+        }
+    }
+    out
+}
+
+fn push(out: &mut Vec<Block>, kind: BlockKind, source: &str) {
+    let source = source.trim_end();
+    if source.is_empty() {
+        return;
+    }
+    out.push(Block {
+        kind,
+        source: source.to_string(),
+    });
+}
+
+/// How many leading blocks of `old` and `new` are identical -- what a
+/// caller may keep the laid-out widgets for. See the module doc for why
+/// this is a comparison rather than an assumption.
+pub fn common_prefix(old: &[Block], new: &[Block]) -> usize {
+    old.iter().zip(new).take_while(|(a, b)| a == b).count()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn kinds(src: &str) -> Vec<BlockKind> {
+        split_blocks(src).into_iter().map(|b| b.kind).collect()
+    }
+
+    #[test]
+    fn a_message_splits_into_its_top_level_blocks() {
+        let src = "# Title\n\nFirst para.\n\n```rust\nfn main() {}\n```\n\n- a\n- b\n";
+        assert_eq!(
+            kinds(src),
+            vec![
+                BlockKind::Heading,
+                BlockKind::Paragraph,
+                BlockKind::Code,
+                BlockKind::List
+            ]
+        );
+        let blocks = split_blocks(src);
+        assert_eq!(blocks[1].source, "First para.");
+        assert_eq!(blocks[2].source, "```rust\nfn main() {}\n```");
+    }
+
+    #[test]
+    fn blank_input_has_no_blocks() {
+        assert!(split_blocks("").is_empty());
+        assert!(split_blocks("   \n\n  ").is_empty());
+    }
+
+    /// The property the streaming fast path rests on, in its ordinary
+    /// shape: a delta landing in the last paragraph must leave every
+    /// earlier block byte-identical.
+    #[test]
+    fn a_delta_into_the_last_paragraph_leaves_earlier_blocks_untouched() {
+        let before = split_blocks("# Title\n\nFirst para.\n\nSecond par");
+        let after = split_blocks("# Title\n\nFirst para.\n\nSecond paragraph now.");
+        assert_eq!(common_prefix(&before, &after), 2);
+        assert_eq!(before.len(), 3);
+        assert_eq!(after.len(), 3);
+        assert_ne!(before[2], after[2]);
+    }
+
+    /// A delta that starts a *new* block keeps every old block, including
+    /// the one that was last -- so the fast path appends rather than
+    /// replacing.
+    #[test]
+    fn a_delta_that_starts_a_new_block_keeps_every_old_one() {
+        let before = split_blocks("First para.\n\nSecond para.");
+        let after = split_blocks("First para.\n\nSecond para.\n\nThird");
+        assert_eq!(common_prefix(&before, &after), 2);
+        assert_eq!(after.len(), 3);
+    }
+
+    /// A code fence arrives one delta at a time and is unterminated for
+    /// most of its life. It must still be *one* block the whole way, or
+    /// every delta would re-split the message into a different number of
+    /// pieces.
+    #[test]
+    fn an_unterminated_fence_is_one_block_while_it_streams() {
+        for src in [
+            "Here:\n\n```rust\n",
+            "Here:\n\n```rust\nfn main() {\n",
+            "Here:\n\n```rust\nfn main() {\n    println!(\"hi\");\n",
+        ] {
+            assert_eq!(
+                kinds(src),
+                vec![BlockKind::Paragraph, BlockKind::Code],
+                "{src:?}"
+            );
+        }
+    }
+
+    /// The half the fast path had no reason to touch, and the reason
+    /// `common_prefix` is a comparison rather than an assumption:
+    /// appending can rewrite what came before. `---` under a paragraph
+    /// turns that paragraph into a setext heading, so the block that was
+    /// already laid out is not the block it is now.
+    #[test]
+    fn appending_can_rewrite_an_earlier_block_and_the_prefix_says_so() {
+        let before = split_blocks("Not a heading\n\nsecond");
+        let after = split_blocks("Not a heading\n\nsecond\n---");
+        assert_eq!(before[1].kind, BlockKind::Paragraph);
+        assert_eq!(after[1].kind, BlockKind::Heading);
+        assert_eq!(
+            common_prefix(&before, &after),
+            1,
+            "the rewritten block must not be reported as keepable"
+        );
+    }
+
+    #[test]
+    fn a_thematic_break_is_its_own_block() {
+        assert_eq!(
+            kinds("one\n\n---\n\ntwo"),
+            vec![BlockKind::Paragraph, BlockKind::Other, BlockKind::Paragraph]
+        );
+    }
+
+    /// The shapes a real transcript actually contains, each checked for
+    /// the one property the streaming fast path needs: the *number* of
+    /// blocks and every earlier block's source stay put while the message
+    /// grows. A fence's own blank lines, a `---` inside one, a nested
+    /// list and a table are all places where a naive line-based split
+    /// would break the message into more pieces than there are blocks.
+    #[test]
+    fn the_transcripts_own_block_shapes_survive_a_split() {
+        let fence_with_blanks = "Intro.\n\n```rust\nfn a() {}\n\nfn b() {}\n```\n\nAfter.";
+        assert_eq!(
+            kinds(fence_with_blanks),
+            vec![BlockKind::Paragraph, BlockKind::Code, BlockKind::Paragraph],
+            "a blank line inside a fence is not a block boundary"
+        );
+        assert_eq!(
+            kinds("```\n---\n```"),
+            vec![BlockKind::Code],
+            "a thematic break inside a fence is code, not a break"
+        );
+        assert_eq!(
+            kinds("- a\n  - a1\n  - a2\n- b"),
+            vec![BlockKind::List],
+            "a nested list is one top-level block"
+        );
+        assert_eq!(
+            kinds("## Heading\n```sh\nls\n```"),
+            vec![BlockKind::Heading, BlockKind::Code],
+            "a fence directly under a heading, with no blank line"
+        );
+        assert_eq!(
+            kinds("| a | b |\n|---|---|\n| 1 | 2 |"),
+            vec![BlockKind::Table]
+        );
+        assert_eq!(
+            kinds("> quoted\n> more\n\nplain"),
+            vec![BlockKind::Quote, BlockKind::Paragraph]
+        );
+    }
+
+    /// `apply_delta`'s precondition, stated as the property rather than
+    /// the arithmetic: for every prefix of a realistic streamed message,
+    /// the blocks before the last one must be exactly the blocks the
+    /// previous prefix had. Where markdown breaks that (the `---` case
+    /// above), `common_prefix` has to *say* so -- which is what the
+    /// `>= len - 1` assertion below checks: the split may rewrite the
+    /// last block, never an earlier one, or `RowBlocks::apply_delta`
+    /// would keep a widget whose text is no longer what it holds.
+    #[test]
+    fn every_prefix_of_a_streamed_message_keeps_all_but_its_last_block() {
+        let full = "# Report\n\nFirst finding, at some length.\n\n```rust\nfn main() {\n\n                        println!(\"hi\");\n}\n```\n\n- one\n  - nested\n- two\n\n| a | b |\n                    |---|---|\n| 1 | 2 |\n\n> and a closing quote.";
+        // Every character boundary, so a delta landing mid-word and one
+        // landing exactly on a fence's closing backtick are both covered.
+        let mut prev = Vec::new();
+        for end in full.char_indices().map(|(i, _)| i).chain([full.len()]) {
+            let now = split_blocks(&full[..end]);
+            let common = common_prefix(&prev, &now);
+            assert!(
+                prev.is_empty() || common + 1 >= prev.len(),
+                "at {end} bytes the split rewrote block {common} of {}, not just the last one:\n                 before={prev:#?}\nafter={now:#?}",
+                prev.len()
+            );
+            prev = now;
+        }
+    }
+
+    /// The half a growing message cannot show: a fence that never closes.
+    /// The stream ends there and the block must still be the code block
+    /// it has been all along, not re-split into paragraphs.
+    #[test]
+    fn a_stream_that_ends_inside_a_fence_still_ends_with_one_code_block() {
+        let src = "Here is the patch:\n\n```diff\n- old line\n+ new line";
+        let blocks = split_blocks(src);
+        assert_eq!(
+            blocks.iter().map(|b| b.kind).collect::<Vec<_>>(),
+            vec![BlockKind::Paragraph, BlockKind::Code]
+        );
+        assert_eq!(blocks[1].source, "```diff\n- old line\n+ new line");
+    }
+
+    /// A delta that closes a fence changes the *last* block only, so the
+    /// fast path takes it -- the case the module doc says is the reason
+    /// `common_prefix` is a comparison.
+    #[test]
+    fn the_delta_that_closes_a_fence_changes_only_the_last_block() {
+        let before = split_blocks("Text.\n\n```\ncode\n");
+        let after = split_blocks("Text.\n\n```\ncode\n```");
+        assert_eq!(before.len(), after.len());
+        assert_eq!(common_prefix(&before, &after), 1);
+        assert_ne!(before[1], after[1]);
+    }
+}

client-core/src/notifications.rs

@@ -0,0 +1,162 @@
+//! `GET /notifications`, the attention stream PLAN.md's "Notifications: two
+//! places, never both" describes. Ported from the parsing half of
+//! `app/.../Notifications.kt`'s `NotificationService` -- the framing
+//! ([`crate::sse`]) and the wire shape ([`SessionNotification`],
+//! [`NotificationKind`], mirroring `server/src/session/mod.rs`'s
+//! `Notification`/`NotificationKind`).
+//!
+//! What is deliberately **not** here, because it is a decision rather than
+//! logic: whether a given notification is shown at all (the session on
+//! screen gets nothing), handed to the app as a banner, or posted to the
+//! platform's own notification drawer. That three-way choice reads
+//! process-wide state (what screen is open, whether the app is in front)
+//! that has no meaning to a pure crate with no UI and no Android in it --
+//! see `android-shell` for where it lives for this port.
+
+use std::io::{BufRead, BufReader};
+
+use serde::Deserialize;
+
+use crate::api::{ApiError, Transport};
+use crate::sse::SseReader;
+
+/// One frame of `GET /notifications`, matching `server/src/session/mod.rs`'s
+/// `Notification` field for field.
+#[derive(Debug, Clone, PartialEq, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct SessionNotification {
+    pub session_id: String,
+    pub title: String,
+    pub kind: NotificationKind,
+    /// Epoch seconds, so a phone that was asleep can say how long ago.
+    pub at: f64,
+}
+
+/// Mirrors `server/src/session/mod.rs`'s `NotificationKind` -- serialized
+/// the same way, so this deserializes the wire's `"awaitingInput"` /
+/// `"finished"` directly rather than through a string match.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub enum NotificationKind {
+    AwaitingInput,
+    Finished,
+}
+
+impl NotificationKind {
+    /// What a notification asks of the reader, in the words they see --
+    /// ported verbatim from `Notifications.kt`'s `attentionLine`. One
+    /// function because the same fact is shown in two places (the
+    /// platform's drawer and the app's own banner) and two mappings of one
+    /// word drift.
+    pub fn attention_line(self) -> &'static str {
+        match self {
+            NotificationKind::AwaitingInput => "Waiting for you",
+            NotificationKind::Finished => "Finished",
+        }
+    }
+}
+
+/// Follows `/notifications`, calling `on_notification` for each frame until
+/// the connection drops or the callback asks to stop (by returning
+/// `false`). Reconnecting is the caller's job -- mirroring
+/// `NotificationService.follow`'s retry loop, which is a platform policy
+/// (how long to wait, whether to give up) rather than parsing logic.
+pub fn follow_notifications(
+    transport: &dyn Transport,
+    mut on_notification: impl FnMut(SessionNotification) -> bool,
+) -> Result<(), ApiError> {
+    let body = transport.stream("/notifications")?;
+    let mut lines = BufReader::new(body).lines();
+    let mut reader = SseReader::new();
+    while let Some(line) = lines.next().transpose().map_err(|e| ApiError {
+        message: format!("Can't reach the server -- retrying. ({e})"),
+        status: None,
+    })? {
+        let Some(frame) = reader.feed_line(&line) else {
+            continue;
+        };
+        if frame.data.is_empty() {
+            continue;
+        }
+        let notification: SessionNotification =
+            serde_json::from_str(&frame.data).map_err(|e| ApiError {
+                message: format!("The server sent a notification this build couldn't parse: {e}"),
+                status: None,
+            })?;
+        if !on_notification(notification) {
+            return Ok(());
+        }
+    }
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::api::{Body, RawResponse};
+    use std::io::Cursor;
+
+    struct FixtureTransport {
+        body: &'static str,
+    }
+
+    impl Transport for FixtureTransport {
+        fn request(
+            &self,
+            _method: &str,
+            _path: &str,
+            _body: Option<Body>,
+        ) -> Result<RawResponse, ApiError> {
+            unimplemented!("this fixture only serves a stream")
+        }
+
+        fn stream(&self, _path: &str) -> Result<Box<dyn std::io::Read + Send>, ApiError> {
+            Ok(Box::new(Cursor::new(self.body.as_bytes().to_vec())))
+        }
+    }
+
+    #[test]
+    fn a_notification_frame_parses_both_kinds() {
+        let transport = FixtureTransport {
+            body: "data:{\"sessionId\":\"s1\",\"title\":\"fix the bug\",\"kind\":\"awaitingInput\",\"at\":1.0}\n\n\
+                   data:{\"sessionId\":\"s2\",\"title\":\"add tests\",\"kind\":\"finished\",\"at\":2.0}\n\n",
+        };
+        let mut seen = Vec::new();
+        follow_notifications(&transport, |n| {
+            seen.push((n.session_id, n.kind));
+            true
+        })
+        .unwrap();
+        assert_eq!(
+            seen,
+            vec![
+                ("s1".to_string(), NotificationKind::AwaitingInput),
+                ("s2".to_string(), NotificationKind::Finished),
+            ]
+        );
+    }
+
+    #[test]
+    fn the_caller_can_stop_early() {
+        let transport = FixtureTransport {
+            body: "data:{\"sessionId\":\"s1\",\"title\":\"a\",\"kind\":\"finished\",\"at\":1.0}\n\n\
+                   data:{\"sessionId\":\"s2\",\"title\":\"b\",\"kind\":\"finished\",\"at\":2.0}\n\n",
+        };
+        let mut count = 0;
+        follow_notifications(&transport, |_| {
+            count += 1;
+            count < 1
+        })
+        .unwrap();
+        assert_eq!(count, 1);
+    }
+
+    #[test]
+    fn attention_line_matches_the_kotlin_original() {
+        assert_eq!(
+            NotificationKind::AwaitingInput.attention_line(),
+            "Waiting for you"
+        );
+        assert_eq!(NotificationKind::Finished.attention_line(), "Finished");
+    }
+}

client-core/src/tool_summary.rs

@@ -0,0 +1,244 @@
+//! A tool call's input, read rather than dumped -- the port of
+//! `ToolInput.kt`'s `parseToolInput`, which is what both the collapsed
+//! card's one-line summary and the expanded card's key/value list are
+//! derived from.
+//!
+//! Every tool's input arrives as JSON, and showing it raw makes the reader
+//! parse `{"command":"…","timeout":120000}` themselves to find the one
+//! line they care about. So the fields that carry the meaning are pulled
+//! out, and anything left over is still shown, because dropping a field
+//! would be claiming the tool has no other input when it might.
+//!
+//! Pure, and here rather than in the widget crate, for the reason the rest
+//! of this crate exists: the derivation is the same on a phone and on a
+//! desktop, and it is testable without a renderer.
+
+use crate::durations::format_millis_text;
+use crate::highlight::Language;
+use serde_json::{Map, Value};
+
+/// A tool call's input, split into the parts a card draws separately.
+#[derive(Debug, Clone, PartialEq, Eq, Default)]
+pub struct ToolInput {
+    /// The thing that will actually be run or read, if this tool has one.
+    pub subject: Option<String>,
+    /// The language [`ToolInput::subject`] is written in, for
+    /// highlighting.
+    pub language: Option<Language>,
+    /// The tool's own one-line summary, when it wrote one.
+    pub description: Option<String>,
+    /// How long the call may take, in the largest units it fits. Shown
+    /// apart because it is a limit on the call rather than part of what
+    /// the call does.
+    pub timeout: Option<String>,
+    /// Everything else, as `name: value` lines. Never dropped.
+    pub rest: Vec<String>,
+}
+
+impl ToolInput {
+    /// The one line to show when there is only room for one: what this
+    /// call is for.
+    pub fn title(&self) -> Option<&str> {
+        self.description
+            .as_deref()
+            .or(self.subject.as_deref())
+            // A subject that is only whitespace would draw as an empty
+            // summary line, which reads as a tool with nothing to say
+            // rather than as one whose subject was blank.
+            .filter(|t| !t.trim().is_empty())
+    }
+}
+
+/// Which field of which tool is the subject.
+///
+/// A table rather than a chain of `if`s: adding a tool is a row, and the
+/// shape stops any of them from being the special case that gets its own
+/// code path. Unknown tools fall through to "no subject, everything is
+/// rest".
+const SUBJECTS: &[(&str, &str, Option<Language>)] = &[
+    ("Bash", "command", Some(Language::Shell)),
+    ("Read", "file_path", None),
+    ("Write", "file_path", None),
+    ("Edit", "file_path", None),
+    ("Glob", "pattern", None),
+    ("Grep", "pattern", None),
+    ("WebFetch", "url", None),
+];
+
+/// Fields that are the tool's own prose about itself rather than input to
+/// it.
+const DESCRIPTIONS: &[&str] = &["description", "prompt"];
+
+/// One JSON value as the Kotlin's `JSONObject.optString`/`get` wrote it: a
+/// string is its own characters, anything else is its JSON form.
+///
+/// One function rather than two, because the same coercion decides both
+/// what a subject reads as and what a leftover field's value reads as, and
+/// two copies would eventually disagree about a number.
+fn as_text(value: &Value) -> String {
+    match value {
+        Value::String(s) => s.clone(),
+        other => other.to_string(),
+    }
+}
+
+fn non_blank(value: Option<&Value>) -> Option<String> {
+    let text = as_text(value?);
+    (!text.trim().is_empty()).then_some(text)
+}
+
+/// Split `input` (a tool call's JSON) into the parts a card draws.
+///
+/// Input that is not a JSON object -- older transcripts and some tools
+/// send a bare string -- is still the input, so it is still shown, as the
+/// whole of `rest`.
+pub fn parse_tool_input(tool: &str, input: &str) -> ToolInput {
+    let Ok(Value::Object(json)) = serde_json::from_str::<Value>(input) else {
+        return ToolInput {
+            rest: match input.trim().is_empty() {
+                true => Vec::new(),
+                false => vec![input.to_string()],
+            },
+            ..ToolInput::default()
+        };
+    };
+    parse_object(tool, &json)
+}
+
+fn parse_object(tool: &str, json: &Map<String, Value>) -> ToolInput {
+    let (subject_key, language) = SUBJECTS
+        .iter()
+        .find(|(name, ..)| *name == tool)
+        .map(|(_, key, language)| (Some(*key), *language))
+        .unwrap_or((None, None));
+    let subject = subject_key.and_then(|key| non_blank(json.get(key)));
+    let description = DESCRIPTIONS
+        .iter()
+        .find_map(|key| non_blank(json.get(*key)));
+    let timeout = non_blank(json.get("timeout")).map(|t| format_millis_text(&t));
+
+    // Sorted, so the leftovers are in the same order every time this call
+    // is drawn rather than in whatever order the JSON happened to arrive
+    // in. A field is left out only when it is already drawn somewhere
+    // else on the card.
+    let mut keys: Vec<&String> = json
+        .keys()
+        .filter(|k| Some(k.as_str()) != subject_key || subject.is_none())
+        .filter(|k| !DESCRIPTIONS.contains(&k.as_str()) || description.is_none())
+        .filter(|k| k.as_str() != "timeout" || timeout.is_none())
+        .collect();
+    keys.sort();
+    let rest = keys
+        .into_iter()
+        .map(|key| format!("{key}: {}", as_text(&json[key])))
+        .collect();
+
+    ToolInput {
+        subject,
+        language,
+        description,
+        timeout,
+        rest,
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn each_tool_in_the_table_has_its_own_subject() {
+        // One assertion per row of `SUBJECTS`, because the table is the
+        // whole of the rule and a row lost in an edit would otherwise
+        // only show up as a card with no summary line.
+        let cases = [
+            ("Bash", r#"{"command":"ls -la"}"#, "ls -la"),
+            ("Read", r#"{"file_path":"/tmp/x.rs"}"#, "/tmp/x.rs"),
+            ("Write", r#"{"file_path":"/tmp/y.rs"}"#, "/tmp/y.rs"),
+            ("Edit", r#"{"file_path":"/tmp/z.rs"}"#, "/tmp/z.rs"),
+            ("Glob", r#"{"pattern":"**/*.rs"}"#, "**/*.rs"),
+            ("Grep", r#"{"pattern":"fn main"}"#, "fn main"),
+            ("WebFetch", r#"{"url":"https://x/y"}"#, "https://x/y"),
+        ];
+        for (tool, input, expected) in cases {
+            let parsed = parse_tool_input(tool, input);
+            assert_eq!(parsed.subject.as_deref(), Some(expected), "{tool}");
+            assert_eq!(parsed.title(), Some(expected), "{tool}");
+            assert!(parsed.rest.is_empty(), "{tool}: {:?}", parsed.rest);
+        }
+        assert_eq!(
+            parse_tool_input("Bash", r#"{"command":"ls"}"#).language,
+            Some(Language::Shell),
+            "a Bash command is shell, and is the one row that names a language"
+        );
+    }
+
+    #[test]
+    fn a_tools_own_description_is_what_the_one_line_says() {
+        // The description wins over the subject: it is the tool's own
+        // prose about what this call is for, which is what a reader
+        // scanning a collapsed run is looking for.
+        let parsed = parse_tool_input(
+            "Bash",
+            r#"{"command":"cargo test -p iris","description":"Run the iris tests"}"#,
+        );
+        assert_eq!(parsed.title(), Some("Run the iris tests"));
+        assert_eq!(parsed.subject.as_deref(), Some("cargo test -p iris"));
+        assert!(parsed.rest.is_empty(), "{:?}", parsed.rest);
+    }
+
+    #[test]
+    fn a_timeout_is_read_as_a_span_and_kept_apart_from_the_rest() {
+        let parsed = parse_tool_input("Bash", r#"{"command":"sleep 500","timeout":480000}"#);
+        assert_eq!(parsed.timeout.as_deref(), Some("8m"));
+        assert!(parsed.rest.is_empty(), "{:?}", parsed.rest);
+    }
+
+    #[test]
+    fn every_field_not_drawn_elsewhere_is_still_shown() {
+        // The half the "never dropped" promise is about: a tool this
+        // build has never heard of has no subject, so *everything* is
+        // rest -- and a known tool's extra fields are too.
+        let parsed = parse_tool_input(
+            "Edit",
+            r#"{"file_path":"/a.rs","old_string":"x","new_string":"y","replace_all":true}"#,
+        );
+        assert_eq!(
+            parsed.rest,
+            vec![
+                "new_string: y".to_string(),
+                "old_string: x".to_string(),
+                "replace_all: true".to_string(),
+            ],
+            "sorted, and a non-string value written as JSON"
+        );
+        let unknown = parse_tool_input("SomeNewTool", r#"{"b":2,"a":"one"}"#);
+        assert_eq!(unknown.subject, None);
+        assert_eq!(unknown.rest, vec!["a: one".to_string(), "b: 2".to_string()]);
+    }
+
+    #[test]
+    fn input_that_is_not_an_object_is_still_the_input() {
+        // Older transcripts and some tools send a bare string; a card
+        // that dropped it would claim the call had no input at all.
+        assert_eq!(
+            parse_tool_input("Bash", "just a string").rest,
+            vec!["just a string".to_string()]
+        );
+        assert_eq!(parse_tool_input("Bash", "   ").rest, Vec::<String>::new());
+        assert_eq!(parse_tool_input("Bash", "").title(), None);
+    }
+
+    #[test]
+    fn a_blank_subject_is_no_subject_rather_than_an_empty_summary_line() {
+        let parsed = parse_tool_input("Bash", r#"{"command":"   ","other":1}"#);
+        assert_eq!(parsed.subject, None);
+        assert_eq!(parsed.title(), None);
+        // Not dropped just because it was blank -- it is still a field
+        // the call carried.
+        assert_eq!(
+            parsed.rest,
+            vec!["command:    ".to_string(), "other: 1".to_string()]
+        );
+    }
+}

client-core/src/transcript_cache.rs

@@ -1,7 +1,7 @@
 //! This phone's copy of the transcripts it has already been sent, so
 //! reopening a session does not download it again. Ported from
-//! `app/.../TranscriptCache.kt`; see `TRANSCRIPT_CACHE.md` at the repo root
-//! for the design and `CLIENT_CORE.md` for how this file corresponds to it.
+//! `app/.../TranscriptCache.kt`; see `docs/TRANSCRIPT_CACHE.md`
+//! for the design and `docs/CLIENT_CORE.md` for how this file corresponds to it.
 //!
 //! What is stored is the server's own JSON for one event per line, in
 //! transcript order. Reading the cache means running the same [`seq_of`]
@@ -361,6 +361,10 @@ impl SessionCache {
             {
                 return Ok(false);
             }
+            debug_assert!(
+                lines.iter().all(|l| !l.contains('\n')),
+                "a stored page's lines must each be one line"
+            );
             fs::create_dir_all(&this.dir)?;
             let kind = if rows { "rows" } else { "raw" };
             let mut content = lines.join("\n");
@@ -389,8 +393,16 @@ impl SessionCache {
                 return Ok(());
             };
             // Written as it arrived. A newline inside it would split one
-            // event into two unreadable halves, but neither source can
-            // produce one.
+            // event into two unreadable halves. No source here can produce
+            // one -- an SSE `data:` field cannot hold a raw newline, and a
+            // fetched line is one element of a compact JSON array -- but
+            // that is a fact about the *server's* serializer rather than
+            // anything this file controls, so it is checked rather than
+            // trusted.
+            debug_assert!(
+                !line.contains('\n'),
+                "a cached transcript line must be one line: {line}"
+            );
             use std::io::Write;
             writer.write_all(line.as_bytes())?;
             writer.write_all(b"\n")?;

client-core/src/transcript_fold.rs

@@ -78,6 +78,11 @@ pub enum TranscriptItem {
         input: String,
         output: String,
         done: bool,
+        /// Whether the result that arrived said the call failed
+        /// ([`Event::ToolEnd`]'s `is_error`). Meaningless while `done` is
+        /// false, and [`ToolState::of`] is the only thing that reads the
+        /// pair, so the two cannot be combined wrongly at a call site.
+        failed: bool,
         asks: Vec<QuestionCard>,
         images: Vec<String>,
     },
@@ -112,6 +117,13 @@ pub enum TranscriptItem {
     ClearedNote {
         seq: u64,
     },
+    /// The account's usage limit stopped the turn; `resets_at` is epoch
+    /// seconds when the dialect said when it lifts (`LimitNote` in
+    /// `TranscriptItems.kt`).
+    LimitNote {
+        seq: u64,
+        resets_at: Option<f64>,
+    },
     CompactedNote {
         seq: u64,
         pre_tokens: Option<u64>,
@@ -131,6 +143,7 @@ impl TranscriptItem {
             | Self::CommandRow { seq, .. }
             | Self::Note { seq, .. }
             | Self::ClearedNote { seq }
+            | Self::LimitNote { seq, .. }
             | Self::CompactedNote { seq, .. } => *seq,
             Self::QuestionCard(card) => card.seq,
         }
@@ -286,6 +299,201 @@ fn split_run(tail: &[TranscriptItem], behind: Option<&str>) -> Vec<TranscriptIte
     out
 }
 
+/// Puts a page of older items in front of the ones already loaded, healing
+/// whatever the page boundary cut in two. Ported from `TranscriptItems.kt`'s
+/// `joinPages`.
+///
+/// Two things straddle a boundary: a tool call separated from its result,
+/// and a message separated from the rest of itself. Both were one thing
+/// before the transcript was cut into pages.
+///
+/// A boundary lands wherever it lands, and roughly half the time that is
+/// between a call and its result. The newer page then holds a `ToolEnd`
+/// whose start it never saw, which `fold_event` draws as a row of its own
+/// -- correctly, because a call that renders as nothing is indistinguishable
+/// from one that never happened. When the older page arrives it brings the
+/// real `ToolStart`, and concatenating the two lists left *both*: the same
+/// call twice.
+///
+/// Merged by the call's own id rather than by position, because position is
+/// exactly what a page boundary destroys. The older row wins on what a
+/// start knows and the newer on what an end knows, which is the only way
+/// round that loses nothing.
+///
+/// The third thing is the *run*, and it is the one the Kotlin original used
+/// to miss (AGENTS.md's "things that have bitten"): every page ends up
+/// here, but `adopt_run` must run on *every* join, not only the one where a
+/// split call was found -- a boundary landing cleanly between two finished
+/// calls, which is most of them, would otherwise leave the older page's
+/// calls under the run name they were folded with. On screen: one run of
+/// tool calls drawn as two groups, with the seam wherever the reader
+/// happened to have paged.
+pub fn join_pages(earlier: &[TranscriptItem], later: &[TranscriptItem]) -> Vec<TranscriptItem> {
+    let (older, newer) = heal_split_message(earlier, later);
+    let started_earlier: std::collections::HashSet<&str> = older
+        .iter()
+        .filter_map(TranscriptItem::as_tool_run)
+        .collect();
+    // Owned rather than borrowed from `newer`: `kept` below needs to consume `newer` by
+    // value, and a map borrowing it would keep that alive.
+    let ended_later: std::collections::HashMap<String, TranscriptItem> = newer
+        .iter()
+        .filter_map(|item| item.as_tool_run().map(|id| (id.to_string(), item.clone())))
+        .filter(|(id, _)| started_earlier.contains(id.as_str()))
+        .collect();
+    let healed: Vec<TranscriptItem> = older
+        .into_iter()
+        .map(|row| match row {
+            TranscriptItem::ToolRun {
+                seq,
+                id,
+                run_id,
+                tool,
+                input,
+                asks: row_asks,
+                images: row_images,
+                ..
+            } if ended_later.contains_key(id.as_str()) => {
+                let &TranscriptItem::ToolRun {
+                    ref output,
+                    done,
+                    failed,
+                    asks: ref half_asks,
+                    images: ref half_images,
+                    ..
+                } = &ended_later[id.as_str()]
+                else {
+                    unreachable!("filtered to ToolRun above");
+                };
+                TranscriptItem::ToolRun {
+                    seq,
+                    id,
+                    run_id,
+                    tool,
+                    input,
+                    output: output.clone(),
+                    done,
+                    failed,
+                    // Kept from both halves: a question or an image can be
+                    // attached to either, depending on which side of the
+                    // boundary its event fell.
+                    asks: row_asks.into_iter().chain(half_asks.clone()).collect(),
+                    images: row_images.into_iter().chain(half_images.clone()).collect(),
+                }
+            }
+            other => other,
+        })
+        .collect();
+    let kept: Vec<TranscriptItem> = newer
+        .into_iter()
+        .filter(|item| match item.as_tool_run() {
+            Some(id) => !ended_later.contains_key(id),
+            None => true,
+        })
+        .collect();
+    let mut out = adopt_run(&healed, &kept);
+    out.extend(kept);
+    // What this function exists to prevent, checked rather than assumed: the same
+    // call drawn twice, once from the page that saw its start and once from the page
+    // that saw its end. Not a seq-ordering check -- a peer note is stamped with the
+    // seq its turn began at, which can be older than the page it arrived in, so the
+    // two pages' seqs legitimately interleave at the boundary.
+    debug_assert!(
+        {
+            let mut ids: Vec<&str> = out.iter().filter_map(TranscriptItem::as_tool_run).collect();
+            let before = ids.len();
+            ids.sort_unstable();
+            ids.dedup();
+            ids.len() == before
+        },
+        "join_pages left the same tool call in both halves"
+    );
+    out
+}
+
+/// Rejoins a message the page boundary cut, and hands back the two pages to
+/// concatenate. Ported from `TranscriptItems.kt`'s `healSplitMessage`.
+///
+/// `fold_event` never leaves two assistant messages next to each other
+/// inside one page, so two meeting at a join are always the two halves of
+/// one reply, and leaving them apart drew a single answer as two with a
+/// paragraph break through the middle of a sentence.
+///
+/// The newer half keeps its identity, for the reason `adopt_run`'s doc
+/// gives. It grows by what the older half brings, which is safe here and
+/// nowhere else -- the join is at the oldest end of what is loaded, so the
+/// growth extends off the top of the screen.
+fn heal_split_message(
+    earlier: &[TranscriptItem],
+    later: &[TranscriptItem],
+) -> (Vec<TranscriptItem>, Vec<TranscriptItem>) {
+    let (
+        Some(TranscriptItem::AssistantMsg {
+            text: head_text, ..
+        }),
+        Some(TranscriptItem::AssistantMsg {
+            seq: tail_seq,
+            text: tail_text,
+            settled: tail_settled,
+        }),
+    ) = (earlier.last(), later.first())
+    else {
+        return (earlier.to_vec(), later.to_vec());
+    };
+    let merged = TranscriptItem::AssistantMsg {
+        seq: *tail_seq,
+        text: format!("{head_text}{tail_text}"),
+        settled: *tail_settled,
+    };
+    let mut newer = vec![merged];
+    newer.extend(later[1..].iter().cloned());
+    (earlier[..earlier.len() - 1].to_vec(), newer)
+}
+
+/// Hands the older calls at the join the name of the run they are joining.
+/// Ported from `TranscriptItems.kt`'s `adoptRun`.
+///
+/// The two pages were folded separately, so a run split by the boundary
+/// came back as two runs with two names. Naming the joined run after the
+/// *older* half would be the obvious way round and is wrong: the newer half
+/// is the part already on screen, and renaming it is renaming the row the
+/// reader is looking at, which is how a list loses its anchor.
+fn adopt_run(earlier: &[TranscriptItem], later: &[TranscriptItem]) -> Vec<TranscriptItem> {
+    let Some(TranscriptItem::ToolRun { run_id, tool, .. }) = later.first() else {
+        return earlier.to_vec();
+    };
+    // A question is in a run of its own on both sides of the join, the same as it would be
+    // had the two pages been folded as one. Without this the heal would merge a group
+    // straight through the row the reader was asked something on.
+    if tool == ASK_USER_QUESTION {
+        return earlier.to_vec();
+    }
+    let joining = run_id.clone();
+    let tail_len = earlier
+        .iter()
+        .rev()
+        .take_while(|item| matches!(item, TranscriptItem::ToolRun { tool, .. } if tool != ASK_USER_QUESTION))
+        .count();
+    if tail_len == 0 {
+        return earlier.to_vec();
+    }
+    let split = earlier.len() - tail_len;
+    let mut out = earlier[..split].to_vec();
+    out.extend(earlier[split..].iter().cloned().map(|mut item| {
+        // `take_while` above already restricted this slice to non-question tool calls;
+        // this just guards the invariant rather than trusting it silently.
+        debug_assert!(
+            matches!(&item, TranscriptItem::ToolRun { tool, .. } if tool != ASK_USER_QUESTION),
+            "adopt_run must never rename a question's own run"
+        );
+        if let TranscriptItem::ToolRun { run_id, .. } = &mut item {
+            *run_id = joining.clone();
+        }
+        item
+    }));
+    out
+}
+
 /// Folds one transcript event onto `items`, the way `foldEvent` does in
 /// `TranscriptItems.kt`. Every wire event has a case; see the module doc
 /// for the one difference from the Kotlin original (no `Unknown` fallback
@@ -361,6 +569,7 @@ pub fn fold_event(items: &[TranscriptItem], entry: &SeqEvent) -> Vec<TranscriptI
                 input: input.to_string(),
                 output: String::new(),
                 done: false,
+                failed: false,
                 asks: Vec::new(),
                 images: Vec::new(),
             });
@@ -371,15 +580,23 @@ pub fn fold_event(items: &[TranscriptItem], entry: &SeqEvent) -> Vec<TranscriptI
                 *out = output.clone();
             }
         }),
-        Event::ToolEnd { id, output } => {
+        Event::ToolEnd {
+            id,
+            output,
+            is_error,
+        } => {
             if items.iter().any(|i| i.as_tool_run() == Some(id.as_str())) {
                 update_tool(items, id, |item| {
                     if let TranscriptItem::ToolRun {
-                        output: out, done, ..
+                        output: out,
+                        done,
+                        failed,
+                        ..
                     } = item
                     {
                         *out = output.clone();
                         *done = true;
+                        *failed = *is_error;
                     }
                 })
             } else {
@@ -393,6 +610,7 @@ pub fn fold_event(items: &[TranscriptItem], entry: &SeqEvent) -> Vec<TranscriptI
                     input: String::new(),
                     output: output.clone(),
                     done: true,
+                    failed: *is_error,
                     asks: Vec::new(),
                     images: Vec::new(),
                 });
@@ -449,6 +667,7 @@ pub fn fold_event(items: &[TranscriptItem], entry: &SeqEvent) -> Vec<TranscriptI
                     input,
                     output,
                     done,
+                    failed,
                     images,
                 } if asks.iter().any(|a| &a.id == id) => {
                     for ask in asks.iter_mut() {
@@ -464,6 +683,7 @@ pub fn fold_event(items: &[TranscriptItem], entry: &SeqEvent) -> Vec<TranscriptI
                         input,
                         output,
                         done,
+                        failed,
                         asks,
                         images,
                     }
@@ -525,6 +745,14 @@ pub fn fold_event(items: &[TranscriptItem], entry: &SeqEvent) -> Vec<TranscriptI
             items.push(TranscriptItem::ClearedNote { seq });
             items
         }
+        Event::LimitReached { resets_at } => {
+            let mut items = items.to_vec();
+            items.push(TranscriptItem::LimitNote {
+                seq,
+                resets_at: *resets_at,
+            });
+            items
+        }
         Event::Compacted {
             pre_tokens,
             post_tokens,
@@ -541,6 +769,74 @@ pub fn fold_event(items: &[TranscriptItem], entry: &SeqEvent) -> Vec<TranscriptI
     }
 }
 
+/// What became of one tool call -- every state a card has to be able to
+/// draw, including the two that are not answers.
+///
+/// The pair this enum exists for is [`ToolState::Succeeded`] against
+/// [`ToolState::NoResult`]. A call that finished having printed nothing
+/// and a call whose result never arrived both leave an empty `output`,
+/// and drawing them the same way states a verdict nobody reached: "it
+/// worked and said nothing" reads as a fact, where the truth is that the
+/// turn ended before anything came back.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum ToolState {
+    /// Started, no result yet, and the session is still working -- the
+    /// ordinary state of a call in flight.
+    Running,
+    /// Stopped on the reader: a permission or question this call carries
+    /// has not been answered, so nothing is happening until somebody
+    /// answers it. Distinct from [`Self::Running`] because whose move it
+    /// is differs, which is the Compose card's "your turn".
+    Deciding,
+    /// A result arrived and the tool did not report a failure.
+    Succeeded,
+    /// A result arrived and the tool reported that the call failed
+    /// (`is_error`).
+    Failed,
+    /// No result ever arrived and the session is not working any more --
+    /// the turn was interrupted, or the process went away. Not a verdict
+    /// on the call: it says only that nobody found out.
+    NoResult,
+}
+
+impl ToolState {
+    /// The state of one call. `session_working` is
+    /// [`session_working`]'s answer for the session this call is in --
+    /// the only thing here that is not a property of the call itself, and
+    /// what separates "still running" from "never came back".
+    ///
+    /// Written once, over the fields rather than per call site, because
+    /// the five states are decided by four conditions and every place
+    /// that re-derived a subset of them got a different subset.
+    pub fn of(item: &TranscriptItem, session_working: bool) -> Option<Self> {
+        let TranscriptItem::ToolRun {
+            done, failed, asks, ..
+        } = item
+        else {
+            return None;
+        };
+        debug_assert!(
+            !failed || *done,
+            "a call cannot have failed before its result arrived"
+        );
+        Some(if asks.iter().any(|ask| ask.answers.is_empty()) {
+            // Ahead of `done`: a call waiting on permission has not
+            // finished either, and which of the two the reader is being
+            // told about is the one they can act on.
+            Self::Deciding
+        } else if !*done {
+            match session_working {
+                true => Self::Running,
+                false => Self::NoResult,
+            }
+        } else if *failed {
+            Self::Failed
+        } else {
+            Self::Succeeded
+        })
+    }
+}
+
 /// One row as the transcript draws it: a run of consecutive tool calls, or
 /// anything else. Ported from `ToolRows.kt`'s `TranscriptRow` and
 /// `groupToolRuns` -- the Compose card rendering in that file is not part
@@ -606,6 +902,37 @@ pub fn group_tool_runs(items: &[TranscriptItem]) -> Vec<TranscriptRow> {
     rows
 }
 
+/// Folds a page of raw transcript lines (`ApiClient::fetch_transcript_page`'s
+/// `Vec<Value>`) into the flat item list this module works over. A line
+/// this build can't parse fails the whole page rather than being skipped --
+/// CODE_RULES's "an enumeration must be able to say 'it broke'" -- since
+/// silently dropping one event could hide, say, a user message that then
+/// looks like it was never sent. Moved here from `desktop-app`'s `app.rs`
+/// (RUST.md's E4) when the Android transcript client (I5) needed the same
+/// fold: "write the logic once" applies to any caller embedding
+/// `transcript-ui` against a live server, not just the first one.
+pub fn fold_page(values: &[serde_json::Value]) -> Result<Vec<TranscriptItem>, String> {
+    let mut items = Vec::new();
+    for value in values {
+        let event: SeqEvent = serde_json::from_value(value.clone()).map_err(|e| {
+            format!("the server sent a transcript line this build couldn't parse: {e}")
+        })?;
+        items = fold_event(&items, &event);
+    }
+    Ok(items)
+}
+
+/// The wire `seq` a raw transcript line carries -- the live-stream resume
+/// cursor after loading a page must be this, not a folded item's `seq()`.
+/// A folded `AssistantMsg` keeps the seq of the *first* delta it
+/// accumulated (`fold_event`'s own doc), so resuming from that seq would
+/// re-deliver every delta already folded into it, duplicating the tail of
+/// a reply that was mid-stream when the page was fetched -- found via a
+/// real screenshot in E4 (RUST.md), where the assistant's line doubled.
+pub fn raw_seq(value: &serde_json::Value) -> Option<u64> {
+    value.get("seq")?.as_u64()
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -749,6 +1076,7 @@ mod tests {
             Event::ToolEnd {
                 id: "x".to_string(),
                 output: "done".to_string(),
+                is_error: false,
             },
         )]);
         assert_eq!(
@@ -761,6 +1089,7 @@ mod tests {
                 input: String::new(),
                 output: "done".to_string(),
                 done: true,
+                failed: false,
                 asks: Vec::new(),
                 images: Vec::new(),
             }]
@@ -824,4 +1153,368 @@ mod tests {
             other => panic!("expected a QuestionCard, got {other:?}"),
         }
     }
+
+    fn line(seq: u64, json: serde_json::Value) -> serde_json::Value {
+        let mut obj = json;
+        obj["seq"] = serde_json::json!(seq);
+        obj["ts"] = serde_json::json!(1.0);
+        obj
+    }
+
+    /// The regression for a bug a real `run-headless.sh` screenshot found
+    /// in `desktop-app` (E4, RUST.md): resuming the live stream from the
+    /// last *item's* seq re-delivers the deltas already folded into a
+    /// still-open assistant message, doubling its tail. `raw_seq` of the
+    /// last wire line must be the true high-water mark instead, which for a
+    /// run of deltas is higher than every item's own `seq()`.
+    #[test]
+    fn the_resume_cursor_is_the_last_wire_seq_not_the_last_items_seq() {
+        let values = vec![
+            line(1, serde_json::json!({"type": "userMessage", "text": "hi"})),
+            line(
+                2,
+                serde_json::json!({"type": "assistantText", "delta": "a"}),
+            ),
+            line(
+                3,
+                serde_json::json!({"type": "assistantText", "delta": "b"}),
+            ),
+            line(
+                4,
+                serde_json::json!({"type": "assistantText", "delta": "c"}),
+            ),
+        ];
+        let after = raw_seq(values.last().unwrap()).unwrap();
+        assert_eq!(after, 4);
+
+        let items = fold_page(&values).unwrap();
+        let assistant_seq = items
+            .iter()
+            .find(|i| matches!(i, TranscriptItem::AssistantMsg { .. }))
+            .unwrap()
+            .seq();
+        assert_eq!(assistant_seq, 2);
+        assert_ne!(
+            after, assistant_seq,
+            "the fixed bug: these must differ here"
+        );
+    }
+
+    #[test]
+    fn a_page_folds_into_one_settled_assistant_message() {
+        let values = vec![
+            line(1, serde_json::json!({"type": "userMessage", "text": "hi"})),
+            line(
+                2,
+                serde_json::json!({"type": "assistantText", "delta": "hel"}),
+            ),
+            line(
+                3,
+                serde_json::json!({"type": "assistantText", "delta": "lo"}),
+            ),
+        ];
+        let items = fold_page(&values).unwrap();
+        assert_eq!(
+            items,
+            vec![
+                TranscriptItem::UserMsg {
+                    seq: 1,
+                    text: "hi".to_string(),
+                    attachments: Vec::new(),
+                },
+                TranscriptItem::AssistantMsg {
+                    seq: 2,
+                    text: "hello".to_string(),
+                    settled: false,
+                },
+            ]
+        );
+    }
+
+    #[test]
+    fn an_unparseable_line_fails_the_whole_page() {
+        let values = vec![serde_json::json!({"seq": 1, "ts": 1.0, "type": "not-a-real-type"})];
+        let err = fold_page(&values).unwrap_err();
+        assert!(err.contains("couldn't parse"));
+    }
+
+    fn tool_start(seq: u64, id: &str, tool: &str) -> SeqEvent {
+        event(
+            seq,
+            Event::ToolStart {
+                id: id.to_string(),
+                tool: tool.to_string(),
+                input: serde_json::json!({}),
+            },
+        )
+    }
+
+    fn tool_end(seq: u64, id: &str, output: &str) -> SeqEvent {
+        event(
+            seq,
+            Event::ToolEnd {
+                id: id.to_string(),
+                output: output.to_string(),
+                is_error: false,
+            },
+        )
+    }
+
+    /// AGENTS.md's "things that have bitten": `joinPages` used to run
+    /// `adoptRun` only on the path where a *split* call was found, so a
+    /// boundary landing cleanly between two already-finished calls -- most
+    /// of them -- left the older page's calls under the run name they were
+    /// folded with, drawing one run of tool calls as two groups. Two
+    /// finished, unrelated calls (no id in common) must still end up under
+    /// one run name after the join.
+    #[test]
+    fn a_clean_boundary_between_two_finished_runs_is_still_healed_into_one_run() {
+        let older = fold_all(&[tool_start(1, "a", "Bash"), tool_end(2, "a", "old output")]);
+        let newer = fold_all(&[tool_start(3, "b", "Bash"), tool_end(4, "b", "new output")]);
+        let joined = join_pages(&older, &newer);
+        let run_ids: Vec<_> = joined
+            .iter()
+            .map(|item| match item {
+                TranscriptItem::ToolRun { run_id, .. } => run_id.as_str(),
+                other => panic!("expected only ToolRun items, got {other:?}"),
+            })
+            .collect();
+        assert_eq!(
+            run_ids,
+            vec!["b", "b"],
+            "the older call must adopt the newer, already-on-screen run's name"
+        );
+    }
+
+    #[test]
+    fn a_call_split_across_the_boundary_merges_into_one_row() {
+        let older = fold_all(&[tool_start(1, "x", "Bash")]);
+        let newer = fold_all(&[tool_end(2, "x", "the result")]);
+        let joined = join_pages(&older, &newer);
+        assert_eq!(
+            joined,
+            vec![TranscriptItem::ToolRun {
+                seq: 1,
+                id: "x".to_string(),
+                run_id: "x".to_string(),
+                tool: "Bash".to_string(),
+                input: "{}".to_string(),
+                output: "the result".to_string(),
+                done: true,
+                failed: false,
+                asks: Vec::new(),
+                images: Vec::new(),
+            }],
+            "the older half's tool/input and the newer half's output/done must both survive"
+        );
+    }
+
+    #[test]
+    fn a_message_split_across_the_boundary_is_rejoined_with_the_newer_halfs_identity() {
+        let older = vec![TranscriptItem::AssistantMsg {
+            seq: 1,
+            text: "Hel".to_string(),
+            settled: false,
+        }];
+        let newer = vec![
+            TranscriptItem::AssistantMsg {
+                seq: 2,
+                text: "lo".to_string(),
+                settled: true,
+            },
+            TranscriptItem::UserMsg {
+                seq: 3,
+                text: "next".to_string(),
+                attachments: Vec::new(),
+            },
+        ];
+        let joined = join_pages(&older, &newer);
+        assert_eq!(
+            joined,
+            vec![
+                TranscriptItem::AssistantMsg {
+                    seq: 2,
+                    text: "Hello".to_string(),
+                    settled: true,
+                },
+                TranscriptItem::UserMsg {
+                    seq: 3,
+                    text: "next".to_string(),
+                    attachments: Vec::new(),
+                },
+            ]
+        );
+    }
+
+    /// A question is in a run of its own on both sides of a join -- healing
+    /// must never rename the run of calls the reader was asked something
+    /// on, the same rule `splitRun` enforces for a live turn boundary.
+    #[test]
+    fn adopt_run_never_renames_into_a_question_row() {
+        let older = fold_all(&[tool_start(1, "a", "Bash"), tool_end(2, "a", "done")]);
+        let newer = vec![TranscriptItem::ToolRun {
+            seq: 3,
+            id: "q".to_string(),
+            run_id: "q".to_string(),
+            tool: ASK_USER_QUESTION.to_string(),
+            input: "{}".to_string(),
+            output: String::new(),
+            done: false,
+            failed: false,
+            asks: Vec::new(),
+            images: Vec::new(),
+        }];
+        let joined = join_pages(&older, &newer);
+        match &joined[0] {
+            TranscriptItem::ToolRun { run_id, .. } => assert_eq!(run_id, "a"),
+            other => panic!("expected a ToolRun, got {other:?}"),
+        }
+    }
+}
+
+/// [`ToolState`] is what a card colours itself by, so each of its five
+/// states is asserted from the events that actually produce it rather than
+/// from a hand-built item -- a mapping that agreed with a fixture and
+/// disagreed with the fold would be invisible until it was on screen.
+#[cfg(test)]
+mod tool_state_tests {
+    use super::*;
+
+    fn event(seq: u64, e: Event) -> SeqEvent {
+        SeqEvent {
+            seq,
+            ts: 0.0,
+            event: e,
+        }
+    }
+
+    fn fold_all(events: &[SeqEvent]) -> Vec<TranscriptItem> {
+        events
+            .iter()
+            .fold(Vec::new(), |items, e| fold_event(&items, e))
+    }
+
+    fn start(id: &str) -> SeqEvent {
+        event(
+            1,
+            Event::ToolStart {
+                id: id.to_string(),
+                tool: "Bash".to_string(),
+                input: serde_json::json!({"command": "ls"}),
+            },
+        )
+    }
+
+    fn end(id: &str, output: &str, is_error: bool) -> SeqEvent {
+        event(
+            2,
+            Event::ToolEnd {
+                id: id.to_string(),
+                output: output.to_string(),
+                is_error,
+            },
+        )
+    }
+
+    fn state_of(events: &[SeqEvent], session_working: bool) -> ToolState {
+        let items = fold_all(events);
+        ToolState::of(&items[0], session_working).expect("the fixture's first item is a tool call")
+    }
+
+    #[test]
+    fn a_result_that_arrived_is_read_from_is_error() {
+        assert_eq!(
+            state_of(&[start("a"), end("a", "ok", false)], false),
+            ToolState::Succeeded
+        );
+        assert_eq!(
+            state_of(&[start("a"), end("a", "No such file", true)], false),
+            ToolState::Failed
+        );
+    }
+
+    /// The pair this enum exists for. Both calls have an empty `output`
+    /// and nothing else distinguishes them, so a card that only looked at
+    /// the text would draw the interrupted one as a call that ran fine and
+    /// printed nothing.
+    #[test]
+    fn a_call_that_printed_nothing_is_not_a_call_that_never_answered() {
+        assert_eq!(
+            state_of(&[start("a"), end("a", "", false)], false),
+            ToolState::Succeeded,
+            "a result arrived; it was empty"
+        );
+        assert_eq!(
+            state_of(&[start("a")], false),
+            ToolState::NoResult,
+            "no result, and the session is not working any more"
+        );
+    }
+
+    /// The same call, mid-turn: still running rather than abandoned. The
+    /// only thing separating the two is the session's own status, which is
+    /// why `of` takes it.
+    #[test]
+    fn no_result_while_the_session_works_is_still_running() {
+        assert_eq!(state_of(&[start("a")], true), ToolState::Running);
+    }
+
+    #[test]
+    fn an_unanswered_ask_is_the_readers_move_whatever_else_is_true() {
+        let asking = event(
+            3,
+            Event::Question {
+                id: "q1".to_string(),
+                prompt: "Allow?".to_string(),
+                header: None,
+                options: vec![QuestionOption {
+                    label: "Allow".to_string(),
+                    description: None,
+                    preview: None,
+                }],
+                multi_select: false,
+                about: Some("a".to_string()),
+            },
+        );
+        let answered = event(
+            4,
+            Event::Answered {
+                id: "q1".to_string(),
+                answers: vec!["Allow".to_string()],
+            },
+        );
+        // Ahead of both "still running" and "no result": the reader can
+        // act on this one, and cannot act on either of those.
+        assert_eq!(
+            state_of(&[start("a"), asking.clone()], true),
+            ToolState::Deciding
+        );
+        assert_eq!(
+            state_of(&[start("a"), asking.clone()], false),
+            ToolState::Deciding
+        );
+        assert_eq!(
+            state_of(
+                &[start("a"), asking, answered, end("a", "ok", false)],
+                false
+            ),
+            ToolState::Succeeded,
+            "once it is answered the call is an ordinary one again"
+        );
+    }
+
+    #[test]
+    fn nothing_but_a_tool_call_has_a_tool_state() {
+        assert_eq!(
+            ToolState::of(
+                &TranscriptItem::UserMsg {
+                    seq: 1,
+                    text: "hi".to_string(),
+                    attachments: Vec::new(),
+                },
+                true
+            ),
+            None
+        );
+    }
 }

client-core/src/transcript_source.rs

@@ -0,0 +1,588 @@
+//! Where a session screen gets a transcript from: this phone's copy first,
+//! the server for the rest. Ported from `app/.../TranscriptSource.kt`; see
+//! `docs/TRANSCRIPT_CACHE.md` for the design this implements and
+//! `docs/CLIENT_CORE.md` for how this file corresponds to the Kotlin.
+//!
+//! One seam rather than a cache the screen has to remember to consult.
+//! Everything fetched before is asked of this, and everything the server
+//! sends is written into the cache on the way past, so a caller never
+//! learns which side answered. The one rule worth keeping in mind: the
+//! cache is never load-bearing. Every read here has a network path beside
+//! it producing the same result.
+//!
+//! **Not ported**: `EventStream.kt`'s reconnect-with-backoff loop and the
+//! ability to close a live stream from another thread. Both are wall-clock
+//! and thread-lifetime concerns that belong to whatever runtime the caller
+//! embeds this crate in (a Tokio task, an iris timer, a Kotlin coroutine
+//! scope) rather than to this pure logic -- `follow` below is the same
+//! decorator shape `iris/desktop-app/src/app.rs` and
+//! `iris/android-app/src/transcript_client.rs` already hand-wrote around
+//! `event_stream::follow_session_events`, just with the cache write built
+//! in so a future caller does not have to repeat it a third time.
+
+use event_model::SeqEvent;
+
+use crate::api::{ApiClient, ApiError, Transport};
+use crate::event_stream::{self, StreamItem};
+use crate::transcript_cache::SessionCache;
+
+/// How many events a session screen opens with, cached or fetched.
+///
+/// The server's own default page size, named here because the cached
+/// opening has to be the same size as the fetched one -- a reader must not
+/// get a shorter first screen for having been here before (`OPENING_WINDOW`
+/// in the Kotlin original).
+pub const OPENING_WINDOW: u32 = 80;
+
+/// A transcript-line parse failure, told apart from [`ApiError`] so a
+/// caller can tell "the server is unreachable" from "the server (or this
+/// phone's own disk) sent something this build cannot read" -- the two
+/// mean different things to a reader (retry, versus a build that is
+/// behind).
+#[derive(Debug, Clone)]
+pub struct ParseError(pub String);
+
+impl std::fmt::Display for ParseError {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.write_str(&self.0)
+    }
+}
+impl std::error::Error for ParseError {}
+
+/// Either half of what can go wrong asking for a page: the network, or a
+/// line neither the cache's nor the server's copy of `parseSeqEvent` could
+/// read.
+#[derive(Debug, Clone)]
+pub enum PageError {
+    Api(ApiError),
+    Parse(ParseError),
+}
+
+impl From<ApiError> for PageError {
+    fn from(e: ApiError) -> Self {
+        Self::Api(e)
+    }
+}
+
+impl From<ParseError> for PageError {
+    fn from(e: ParseError) -> Self {
+        Self::Parse(e)
+    }
+}
+
+/// What [`TranscriptSource::page`] found, kept as two states rather than
+/// one possibly-empty list.
+///
+/// The difference is the whole of AGENTS.md's `loadOlderPage` incident: an
+/// empty [`Self::Events`] means "this conversation has no more history",
+/// which a caller is meant to latch, and [`Self::NothingLoaded`] means the
+/// question could not be asked yet, which it must not. Collapsing the two
+/// into an empty `Vec` puts the bug back, because the caller cannot tell
+/// them apart -- and `unwrap_or_default()` on an `Option` would do the
+/// same silently.
+#[derive(Debug, Clone, PartialEq)]
+pub enum OlderPage {
+    /// The events before the cursor, oldest first. Empty means the start
+    /// of the conversation has been reached.
+    Events(Vec<SeqEvent>),
+    /// Nothing is loaded, so there was no cursor to page back from
+    /// (`before == 0`). Not an answer about the conversation at all.
+    NothingLoaded,
+}
+
+fn parse_line(line: &str) -> Result<SeqEvent, ParseError> {
+    serde_json::from_str(line).map_err(|e| ParseError(format!("{e}")))
+}
+
+/// This phone's copy of one session's transcript, plus the server it
+/// falls back to. Ported from the Kotlin `TranscriptSource` class.
+pub struct TranscriptSource<T: Transport> {
+    api: ApiClient<T>,
+    session_id: String,
+    pub cache: SessionCache,
+}
+
+impl<T: Transport> TranscriptSource<T> {
+    pub fn new(api: ApiClient<T>, session_id: impl Into<String>, cache: SessionCache) -> Self {
+        Self {
+            api,
+            session_id: session_id.into(),
+            cache,
+        }
+    }
+
+    /// The cached opening window, or `None` when there is nothing usable
+    /// to draw.
+    ///
+    /// Meant to be drawn *before* [`Self::probe`] returns, which is the
+    /// whole point of the feature: the rows are on screen while the check
+    /// that they are still the server's rows is in flight, and a failed
+    /// check replaces them exactly as a reset does.
+    pub fn cached_opening(&self, limit: usize) -> Option<Vec<SeqEvent>> {
+        self.cache.tail()?;
+        let lines = self.cache.newest(limit);
+        if lines.is_empty() {
+            return None;
+        }
+        match lines.iter().map(|l| parse_line(l)).collect() {
+            Ok(events) => Some(events),
+            // A line this build cannot read at all, which the cache's own checks cannot
+            // see: it reads a seq off a line, not an event. Nothing to serve, so a cold
+            // open.
+            Err(ParseError(_)) => {
+                self.cache.purge();
+                None
+            }
+        }
+    }
+
+    /// Whether the server's event at the cached cursor is still the cached
+    /// one.
+    ///
+    /// A caller must not resume a live stream from a cached seq unless it
+    /// is the same conversation: a transcript is append-only in ordinary
+    /// use, but the file backing it can be replaced or truncated (a
+    /// sandbox re-seeded with the same ids, a backup restored, a session
+    /// re-imported), and the server's catch-up on such a file would hand
+    /// this phone a continuation of a *different* conversation, spliced
+    /// onto the cached one with no seam. Caught with one request of a few
+    /// hundred bytes.
+    ///
+    /// `Ok(false)` purges the cache and means "open cold". `Err` is the
+    /// server not being askable, which is neither: the cached rows stay
+    /// on screen and the caller tries again on its own reconnect schedule.
+    ///
+    /// What this cannot see is a line changed in the middle of the file
+    /// with the tail intact -- that is what a full reload is for.
+    pub fn probe(&self) -> Result<bool, ApiError> {
+        let Some(tail) = self.cache.tail() else {
+            return Ok(false);
+        };
+        // `before = seq + 1` is the newest event with seq <= the cursor, which is the
+        // event *at* the cursor when the server still has one there.
+        let page = self.api.fetch_transcript_lines(
+            &self.session_id,
+            Some(tail.seq + 1),
+            1,
+            false,
+            None,
+        )?;
+        let matches = page.len() == 1
+            && parse_line(&tail.line)
+                .map(|cached| cached == page[0].1)
+                .unwrap_or(false);
+        if !matches {
+            self.cache.purge();
+        }
+        Ok(matches)
+    }
+
+    /// Today's opening fetch, kept as the start of the live run. Only
+    /// called when the cache has nothing to open with, or when
+    /// [`Self::probe`] said what it had was not the server's.
+    pub fn fetch_opening(&self) -> Result<Vec<SeqEvent>, ApiError> {
+        let page =
+            self.api
+                .fetch_transcript_lines(&self.session_id, None, OPENING_WINDOW, false, None)?;
+        for (line, event) in &page {
+            self.cache.append(line, event.seq);
+        }
+        self.cache.flush();
+        Ok(page.into_iter().map(|(_, event)| event).collect())
+    }
+
+    /// The page before `before`: from the cache when it holds it,
+    /// otherwise from the server bounded by what the cache already has.
+    ///
+    /// The server bound (`after`) is what keeps the cache worth having. A
+    /// coalesced page reaches back as far as its row count takes it -- a
+    /// single reply is hundreds of lines -- so a page fetched after the
+    /// reader has been away could run straight past the cached run and
+    /// overlap it, and an overlapping page cannot be stored. Told where
+    /// this phone's copy starts, the server stops there instead.
+    ///
+    /// `before == 0` answers [`OlderPage::NothingLoaded`] without asking
+    /// the cache or the server anything -- see AGENTS.md's "things that
+    /// have bitten": there is no event before the first one, so the
+    /// request is not a harmless no-op, and its empty answer is
+    /// indistinguishable from having reached the start of history.
+    /// Guarded here rather than left to every caller, because it is a fact
+    /// about the question, not about who is asking it.
+    pub fn page(&self, before: u64, limit: u32, coalesce: bool) -> Result<OlderPage, PageError> {
+        if before == 0 {
+            return Ok(OlderPage::NothingLoaded);
+        }
+        if let Some(lines) = self.cache.page(before, limit as usize, coalesce) {
+            let events: Vec<SeqEvent> = lines
+                .iter()
+                .map(|l| parse_line(l).map_err(PageError::from))
+                .collect::<Result<_, _>>()?;
+            return Ok(OlderPage::Events(events));
+        }
+        let after = self.cache.covered_up_to(before).map(|v| v - 1);
+        let page = self.api.fetch_transcript_lines(
+            &self.session_id,
+            Some(before),
+            limit,
+            coalesce,
+            after,
+        )?;
+        if let Some((_, first_event)) = page.first() {
+            // `before` rather than the newest line's seq: a coalesced page covers
+            // everything up to the cursor it was asked with, and nothing in its lines
+            // says so.
+            let lines: Vec<String> = page.iter().map(|(line, _)| line.clone()).collect();
+            self.cache
+                .store_page(&lines, first_event.seq, before, coalesce);
+        }
+        Ok(OlderPage::Events(
+            page.into_iter().map(|(_, event)| event).collect(),
+        ))
+    }
+
+    /// [`event_stream::follow_session_events`], with every frame written to
+    /// the cache before `on_item` sees it.
+    ///
+    /// Before, so that an event held back for a reader who is scrolled
+    /// away is already on disk -- what the cache holds is what the server
+    /// sent, not what a screen has got round to drawing. Flushed on each
+    /// status change, which is a turn's boundary and the granularity a
+    /// crash may as well lose, and once more when the stream ends.
+    pub fn follow(
+        &self,
+        after: u64,
+        mut on_item: impl FnMut(StreamItem) -> bool,
+    ) -> Result<(), ApiError> {
+        let cache = &self.cache;
+        let result = event_stream::follow_session_events(
+            self.api.transport(),
+            &self.session_id,
+            after,
+            |item| {
+                if let StreamItem::Event { raw, event } = &item {
+                    cache.append(raw, event.seq);
+                    if matches!(event.event, event_model::Event::Status { .. }) {
+                        cache.flush();
+                    }
+                }
+                on_item(item)
+            },
+        );
+        cache.flush();
+        result
+    }
+
+    /// Leaves the cache with everything it was given -- called once a
+    /// caller is done with this source, mirroring the Kotlin `close`'s
+    /// final flush (that method's stream cancellation itself is the
+    /// runtime concern the module doc says is not ported here).
+    pub fn close(&self) {
+        self.cache.flush();
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::api::{Body, RawResponse};
+    use std::collections::VecDeque;
+    use std::io::Read;
+    use std::sync::Mutex;
+
+    /// A transport that answers fixed bodies in call order, and records
+    /// every path it was asked for -- so a test can assert *how many*
+    /// requests a method made, which is the point for the `before == 0`
+    /// guard (AGENTS.md's regression: the guard must stop the request
+    /// before it happens, not merely tolerate the empty answer).
+    #[derive(Default)]
+    struct ScriptedTransport {
+        responses: Mutex<VecDeque<(u16, String)>>,
+        calls: Mutex<Vec<String>>,
+    }
+
+    impl ScriptedTransport {
+        fn respond(&self, status: u16, body: impl Into<String>) {
+            self.responses
+                .lock()
+                .unwrap()
+                .push_back((status, body.into()));
+        }
+
+        fn call_count(&self) -> usize {
+            self.calls.lock().unwrap().len()
+        }
+    }
+
+    impl Transport for ScriptedTransport {
+        fn request(
+            &self,
+            _method: &str,
+            path: &str,
+            _body: Option<Body>,
+        ) -> Result<RawResponse, ApiError> {
+            self.calls.lock().unwrap().push(path.to_string());
+            let (status, body) = self
+                .responses
+                .lock()
+                .unwrap()
+                .pop_front()
+                .unwrap_or_else(|| panic!("ScriptedTransport got an unscripted request: {path}"));
+            Ok(RawResponse {
+                status,
+                body: body.into_bytes(),
+            })
+        }
+
+        fn stream(&self, path: &str) -> Result<Box<dyn Read + Send>, ApiError> {
+            self.calls.lock().unwrap().push(path.to_string());
+            let (_, body) = self
+                .responses
+                .lock()
+                .unwrap()
+                .pop_front()
+                .unwrap_or_else(|| {
+                    panic!("ScriptedTransport got an unscripted stream request: {path}")
+                });
+            Ok(Box::new(std::io::Cursor::new(body.into_bytes())))
+        }
+    }
+
+    fn source(
+        transport: ScriptedTransport,
+        cache_root: &std::path::Path,
+    ) -> TranscriptSource<ScriptedTransport> {
+        let api = ApiClient::new(transport);
+        let cache = crate::transcript_cache::TranscriptCache::new(cache_root).session("s1");
+        TranscriptSource::new(api, "s1", cache)
+    }
+
+    fn status_line(seq: u64) -> String {
+        format!(r#"{{"seq":{seq},"ts":1.0,"type":"status","state":"idle"}}"#)
+    }
+
+    #[test]
+    fn a_cold_cache_has_no_opening_and_fetches_from_the_server() {
+        let dir = tempfile::tempdir().unwrap();
+        let transport = ScriptedTransport::default();
+        transport.respond(200, format!("[{}]", status_line(1)));
+        let source = source(transport, dir.path());
+
+        assert_eq!(source.cached_opening(80), None);
+        let opening = source.fetch_opening().unwrap();
+        assert_eq!(opening.len(), 1);
+        assert_eq!(opening[0].seq, 1);
+        // The fetch wrote through: reopening the same cache now has something to show.
+        assert!(source.cache.tail().is_some());
+    }
+
+    #[test]
+    fn probe_matching_the_cached_tail_leaves_the_cache_alone() {
+        let dir = tempfile::tempdir().unwrap();
+        let transport = ScriptedTransport::default();
+        transport.respond(200, format!("[{}]", status_line(1)));
+        let source = source(transport, dir.path());
+        source.fetch_opening().unwrap();
+
+        let transport2 = ScriptedTransport::default();
+        transport2.respond(200, format!("[{}]", status_line(1)));
+        let cache = crate::transcript_cache::TranscriptCache::new(dir.path()).session("s1");
+        let source2 = TranscriptSource::new(ApiClient::new(transport2), "s1", cache);
+        assert!(source2.probe().unwrap());
+        assert!(source2.cache.tail().is_some());
+    }
+
+    #[test]
+    fn probe_mismatching_the_cached_tail_purges_the_cache() {
+        let dir = tempfile::tempdir().unwrap();
+        let transport = ScriptedTransport::default();
+        transport.respond(200, format!("[{}]", status_line(1)));
+        let source = source(transport, dir.path());
+        source.fetch_opening().unwrap();
+
+        // The server now answers with a different event at the same seq -- the file
+        // behind this session was replaced.
+        let transport2 = ScriptedTransport::default();
+        let different = r#"{"seq":1,"ts":1.0,"type":"status","state":"running"}"#.to_string();
+        transport2.respond(200, format!("[{different}]"));
+        let cache = crate::transcript_cache::TranscriptCache::new(dir.path()).session("s1");
+        let source2 = TranscriptSource::new(ApiClient::new(transport2), "s1", cache);
+        assert!(!source2.probe().unwrap());
+        assert!(source2.cache.tail().is_none());
+    }
+
+    #[test]
+    fn probe_finding_no_server_leaves_the_cache_untouched() {
+        let dir = tempfile::tempdir().unwrap();
+        let transport = ScriptedTransport::default();
+        transport.respond(200, format!("[{}]", status_line(1)));
+        let source = source(transport, dir.path());
+        source.fetch_opening().unwrap();
+
+        let transport2 = ScriptedTransport::default();
+        transport2.respond(500, "server on fire");
+        let cache = crate::transcript_cache::TranscriptCache::new(dir.path()).session("s1");
+        let source2 = TranscriptSource::new(ApiClient::new(transport2), "s1", cache);
+        assert!(source2.probe().is_err());
+        assert!(
+            source2.cache.tail().is_some(),
+            "an unreachable server must not be treated as a mismatch"
+        );
+    }
+
+    /// The regression this module exists to close: `before == 0` must
+    /// never reach the network or the cache, because an empty answer there
+    /// is indistinguishable from "there is genuinely no more history" --
+    /// AGENTS.md's `loadOlderPage` incident.
+    #[test]
+    fn paging_before_the_first_event_makes_no_request_at_all() {
+        let dir = tempfile::tempdir().unwrap();
+        let transport = ScriptedTransport::default();
+        let source = source(transport, dir.path());
+        assert_eq!(source.page(0, 80, true).unwrap(), OlderPage::NothingLoaded);
+        assert_eq!(source.api.transport().call_count(), 0);
+    }
+
+    #[test]
+    fn a_page_already_covered_by_the_cache_never_reaches_the_server() {
+        let dir = tempfile::tempdir().unwrap();
+        let transport = ScriptedTransport::default();
+        transport.respond(200, format!("[{},{}]", status_line(1), status_line(2)));
+        let source = source(transport, dir.path());
+        source.fetch_opening().unwrap();
+
+        let calls_before = source.api.transport().call_count();
+        let OlderPage::Events(page) = source.page(2, 10, true).unwrap() else {
+            panic!("a cursor of 2 is a real question about the conversation");
+        };
+        assert_eq!(page.len(), 1);
+        assert_eq!(page[0].seq, 1);
+        assert_eq!(
+            source.api.transport().call_count(),
+            calls_before,
+            "a cache hit must not touch the network"
+        );
+    }
+
+    /// With nothing older cached there is no floor to give the server, so
+    /// the request carries no `after` at all.
+    #[test]
+    fn a_server_page_with_nothing_older_cached_carries_no_bound() {
+        let dir = tempfile::tempdir().unwrap();
+        let transport = ScriptedTransport::default();
+        transport.respond(200, format!("[{}]", status_line(5)));
+        let source = source(transport, dir.path());
+        source.fetch_opening().unwrap();
+
+        let transport2 = ScriptedTransport::default();
+        transport2.respond(200, format!("[{}]", status_line(3)));
+        let cache = crate::transcript_cache::TranscriptCache::new(dir.path()).session("s1");
+        let source2 = TranscriptSource::new(ApiClient::new(transport2), "s1", cache);
+        source2.page(5, 10, true).unwrap();
+        assert_eq!(
+            source2.api.transport().calls.lock().unwrap()[0],
+            "/sessions/s1/transcript?limit=10&before=5&coalesce=true"
+        );
+    }
+
+    /// The half the test above cannot show: when the cache *does* hold an
+    /// older run, the fetch is floored at its end, or the page would run
+    /// straight past it and overlap -- which `store_page` then refuses,
+    /// silently costing the phone the page it just paid for.
+    #[test]
+    fn a_server_page_is_floored_at_the_end_of_the_cached_run() {
+        let dir = tempfile::tempdir().unwrap();
+        let cache = crate::transcript_cache::TranscriptCache::new(dir.path()).session("s1");
+        // A stored page covering [3, 6) and two live events above it, so the run this
+        // phone holds is [3, 8) -- the newest chunk has to be an appended one, or the
+        // cache reads the directory as damaged and discards it.
+        let lines: Vec<String> = (3..6).map(status_line).collect();
+        assert!(cache.store_page(&lines, 3, 6, true));
+        cache.append(&status_line(6), 6);
+        cache.append(&status_line(7), 7);
+        cache.flush();
+
+        let transport = ScriptedTransport::default();
+        transport.respond(200, format!("[{}]", status_line(9)));
+        let source = TranscriptSource::new(ApiClient::new(transport), "s1", cache);
+        source.page(10, 10, true).unwrap();
+        assert_eq!(
+            source.api.transport().calls.lock().unwrap()[0],
+            "/sessions/s1/transcript?limit=10&before=10&coalesce=true&after=7",
+            "the fetch must stop one seq below where this phone's copy ends"
+        );
+    }
+
+    /// A page the server could not answer is an error, never an empty
+    /// page: the caller would read the second as "this conversation has no
+    /// more history" and stop paging for good.
+    #[test]
+    fn a_failing_server_page_is_an_error_rather_than_an_empty_one() {
+        let dir = tempfile::tempdir().unwrap();
+        let transport = ScriptedTransport::default();
+        transport.respond(500, "server on fire");
+        let source = source(transport, dir.path());
+        assert!(matches!(source.page(9, 10, true), Err(PageError::Api(_)),));
+    }
+
+    /// A cached line this build cannot read is told apart from the network
+    /// failing, for the same reason: neither is "no more history".
+    #[test]
+    fn an_unreadable_cached_page_is_a_parse_error_rather_than_an_empty_one() {
+        let dir = tempfile::tempdir().unwrap();
+        let cache = crate::transcript_cache::TranscriptCache::new(dir.path()).session("s1");
+        cache.store_page(
+            &[r#"{"seq":3,"but":"not an event"}"#.to_string()],
+            3,
+            4,
+            true,
+        );
+        cache.append(&status_line(4), 4);
+        cache.flush();
+        let transport = ScriptedTransport::default();
+        let source = TranscriptSource::new(ApiClient::new(transport), "s1", cache);
+        assert!(matches!(source.page(4, 10, true), Err(PageError::Parse(_)),));
+        assert_eq!(
+            source.api.transport().call_count(),
+            0,
+            "a cache hit that cannot be read must not fall through to the server unnoticed"
+        );
+    }
+
+    #[test]
+    fn a_bad_cached_opening_line_purges_rather_than_panicking() {
+        let dir = tempfile::tempdir().unwrap();
+        let cache = crate::transcript_cache::TranscriptCache::new(dir.path()).session("s1");
+        cache.append("not json at all", 1);
+        cache.flush();
+        let transport = ScriptedTransport::default();
+        let source = TranscriptSource::new(ApiClient::new(transport), "s1", cache);
+        assert_eq!(source.cached_opening(80), None);
+        assert!(
+            source.cache.tail().is_none(),
+            "a damaged line purges the cache"
+        );
+    }
+
+    #[test]
+    fn follow_writes_events_to_the_cache_before_the_caller_sees_them() {
+        let dir = tempfile::tempdir().unwrap();
+        let transport = ScriptedTransport::default();
+        transport.respond(200, format!("{}\n\n", sse_frame(&status_line(1))));
+        let source = source(transport, dir.path());
+        let mut seen = Vec::new();
+        source
+            .follow(0, |item| {
+                if let StreamItem::Event { event, .. } = item {
+                    seen.push(event.seq);
+                }
+                true
+            })
+            .unwrap();
+        assert_eq!(seen, vec![1]);
+        assert_eq!(source.cache.tail().unwrap().seq, 1);
+    }
+
+    fn sse_frame(data: &str) -> String {
+        format!("data:{data}")
+    }
+}

docs/CLIENT_CORE.md

@@ -0,0 +1,216 @@
+# client-core
+
+`client-core/` is the app's pure logic held once instead of twice, per
+RUST.md's recommendation item 1. It is a plain Rust library crate with no UI
+framework dependency of any kind, so it can outlive whichever one the app
+ends up drawing with (Masonry, iris, or something else -- see RUST.md).
+`event-model/` is its sibling: the wire shape both this crate and `server/`
+share, extracted from `server/src/session/driver.rs` and
+`session/transcript.rs` on 2026-09-04.
+
+Neither crate is wired into anything yet. `server/` re-exports `event-model`
+so its own behaviour is unchanged (`./run-tests.sh` covers it); `client-core`
+has no caller -- it exists for whichever experiment in RUST.md picks it up
+next (a Masonry or iris transcript screen, most likely).
+
+## What's here, and what Kotlin file it replaces
+
+| `client-core/src/…`                     | Kotlin original                          | Status |
+|------------------------------------------|-------------------------------------------|--------|
+| `event-model/src/lib.rs` (shared crate)   | `Events.kt` (the enum mirror)             | Done   |
+| `ansi.rs`                                 | `Ansi.kt`                                 | Done, ported test-for-test |
+| `highlight/mod.rs`, `languages.rs`        | `Highlighter.kt`, `Languages.kt`          | Done, ported test-for-test |
+| `highlight/markdown.rs`                   | `MarkdownSyntax.kt`                       | Done, ported test-for-test |
+| `transcript_cache.rs`                     | `TranscriptCache.kt`                      | Done, ported test-for-test |
+| `sse.rs`                                  | `Sse.kt` (the framing half)               | Done, new tests (Kotlin had none of its own beyond integration) |
+| `api.rs`                                  | `Api.kt`                                  | Partial -- see below |
+| `event_stream.rs`                         | `EventStream.kt`                          | Done |
+| `transcript_fold.rs`                      | `TranscriptItems.kt`, `ToolRows.kt`       | Done -- see below |
+| `config.rs`                                | `ServerConfig.kt`'s `handleEnrollment`   | New, desktop-only so far -- see below |
+| `transcript_source.rs`                    | `TranscriptSource.kt`                     | Done -- see below |
+| *(not ported, and may never be)*          | `TranscriptUnits.kt`                      | Out of scope -- see below |
+
+Every file above whose Kotlin counterpart had a JVM unit test (`AnsiTest`,
+`HighlighterTest`, `TranscriptCacheTest`) has had every one of those test
+cases ported alongside it, plus new tests for the pieces that had none
+(`sse.rs`, `api.rs`, `event_stream.rs`, `transcript_fold.rs`,
+`transcript_source.rs` -- the Kotlin `TranscriptSource.kt`/`TranscriptItems.kt`
+had no JVM unit tests of their own, so these were written fresh against the
+Kotlin source and AGENTS.md's paging incidents as the spec). Test count by
+crate as of this writing: **109 in `client-core`**, 0 in `event-model` (its
+types carry no logic of their own to test -- `server/`'s own tests exercise
+them via `session::transcript`'s round-trip coverage).
+
+## Correspondence notes worth knowing before touching either side
+
+- **`ansi.rs`'s `StyledText`/`Style`/`Rgb`** stand in for Compose's
+  `AnnotatedString`/`SpanStyle`/`Color`, since this crate has no Compose.
+  `StyledText` is plain text plus a `Vec<(Range<usize>, Style)>` of
+  non-overlapping spans. Whatever UI framework ends up consuming this
+  crate maps `Style` onto its own text-styling type; nothing here should
+  change to accommodate a particular one.
+- **`highlight`'s `Span`/`Kind`** use **char indices, not byte offsets**
+  (`Vec<char>` internally), mirroring the Kotlin original's `Char`-indexed
+  strings. `highlight::span_text` turns a `Span` back into text for a
+  caller working the same way; a caller that wants byte offsets into a
+  `&str` has to convert.
+- **`transcript_cache.rs`'s `SessionCache::guard`** found a real
+  translation bug while it was being written: an early draft let a
+  *damaged* chunk (one file unreadable, discard just this session) and a
+  genuine I/O failure (disk gone, disable the whole cache) both surface as
+  the same `Err` from one closure, which would have disabled every
+  session's cache over a single corrupt chunk. Fixed by checking a
+  thread-local "was this damage" flag before deciding which failure mode
+  it was -- see the comment on `guard` and the commit message for
+  `transcript_cache.rs`.
+
+## What `api.rs` covers, and what it does not yet
+
+`ApiClient` wraps a `Transport` trait (network I/O kept out from behind, so
+`ApiClient` and `event_stream::follow_session_events` are tested with a
+fake transport and no server). `UreqTransport` is the only real
+implementation, backed by `ureq` -- see its Cargo.toml comment for why
+(blocking, already a project dependency, no extra TLS crate needed since
+`ureq::tls::Certificate::from_pem` reads the pinned CA directly).
+
+Covered: session list/read, message send, unqueue, answer, interrupt,
+stop, start, rename, cwd, model, permission-mode, notify, command,
+compact, delete, and one transcript page.
+
+**Not covered, and each is real work rather than a stub to fill in:**
+setups (`/setups*`, machine and provider discovery), the file explorer
+(`/setups/{id}/dir|file`), usage (`/usage`), models
+(`/models*`, HuggingFace browsing and downloads), attachments
+(`/sessions/{id}/attachments`), importing (`/setups/{id}/importable*`),
+and the `/notifications` stream. `server/src/routes.rs`'s module doc is
+the full table to work from when one of these is next.
+
+## What `transcript_fold.rs` covers, and what it does not yet
+
+`fold_event` covers every `Event` variant server/ can produce today,
+including tool-call/question/image attachment and peer-message placement.
+`group_tool_runs` groups adjacent calls into `TranscriptRow::Tools`.
+
+`join_pages` (with `heal_split_message` and `adopt_run`, both private) is
+now ported too, 2026-09-06 -- the page-boundary healing that merges a tool
+call split across two fetched pages, rejoins a message a boundary cut
+through, and renames a run of tool calls onto whichever name is already on
+screen. Ported with AGENTS.md's "things that have bitten" incidents as the
+spec rather than a JVM test file (`TranscriptItems.kt` had none of its
+own): `a_clean_boundary_between_two_finished_runs_is_still_healed_into_one_run`
+is the regression test for the bug that shipped -- `adopt_run` must run on
+*every* join, not only the one where a split call was found, or a boundary
+landing cleanly between two already-finished calls (most of them) leaves
+one run drawn as two. `a_call_split_across_the_boundary_merges_into_one_row`,
+`a_message_split_across_the_boundary_is_rejoined_with_the_newer_halfs_identity`,
+and `adopt_run_never_renames_into_a_question_row` cover the other three
+edges the Kotlin doc calls out. `join_pages` ends in a `debug_assert!`
+that no tool id survives in both halves -- the duplicate row it exists to
+prevent, checked rather than assumed. What it deliberately does *not*
+assert is seq ordering across the boundary: a peer note carries the seq
+its turn began at (`place_peer_note`), which can be older than the page
+it arrived in, so the two pages' seqs legitimately interleave there. An
+earlier draft asserted it and would have panicked in debug builds on an
+ordinary transcript.
+
+**Known gap, and a decision for whoever closes it:** `event_model::Event`
+has no `Unknown`/catch-all variant, unlike `Events.kt`'s hand-kept mirror.
+A server newer than this build that adds an event type will fail to parse
+that line rather than degrading to a placeholder row. Closing this means
+deciding how `event_model` itself represents "a shape I don't recognise"
+-- a shared-model decision affecting `server/` too, not a `client-core`-only
+fix, so it is recorded here rather than silently worked around.
+
+## `config.rs`: `EnrolledServer`
+
+`EnrolledServer` (host, port, bearer token) plus `parse_link`, which reads
+the exact `aiapp://enroll?host=H&port=P&token=T` deep link
+`wg-app-link`'s `enroll` mints and `ServerConfig.kt`'s `handleEnrollment`
+parses on the phone -- so any Rust client enrols from the same text a
+phone would scan as a QR, with no second format invented for it (RUST.md's
+E4, DECISIONS.md 2026-09-05). Deliberately does not decide where it is
+persisted or under what file permissions -- a phone seals its token in the
+Android Keystore, `iris/desktop-app/src/config.rs` writes it to
+`$XDG_CONFIG_HOME/ai-app-desktop/enrollment.json` at 0600 -- since that is
+caller-specific (the code rules' "ask for the least you need"). Its only
+caller today is `desktop-app`; a future Android build of this crate would
+be a second one, not a reason to move the type.
+
+## What `transcript_source.rs` covers, and what it does not
+
+`TranscriptSource<T: Transport>` is the seam a session screen asks for a
+page, ported test-for-test against the Kotlin doc rather than a JVM test
+file (there wasn't one): `cached_opening`, `probe`, `fetch_opening`,
+`page` and `follow`, each matching its Kotlin namesake's contract --
+including `probe`'s three-way outcome (matches / cache purged /
+unreachable, told apart so a caller never treats "couldn't ask" as "was
+wrong") and `page`'s cache-vs-server split bounded by `covered_up_to`.
+
+Two additions beyond a literal port, both load-bearing:
+
+- **`page(before, ..)` refuses `before == 0` before touching the cache or
+  the network**, answering `OlderPage::NothingLoaded`. This is AGENTS.md's
+  `loadOlderPage` incident (`before = 0` is "no event before the first
+  one," indistinguishable from "reached the start of history" if a caller
+  ever asks it) moved out of the Kotlin screen and into this layer, so
+  every future caller gets the guard rather than having to remember it.
+  **The return type is `OlderPage`, not a `Vec`, and that is the guard.**
+  The Kotlin's two falses are different answers -- `oldestSeq == 0`
+  returns without touching `moreHistory`, an empty page latches it false
+  -- so a port that answered both with an empty list would have moved the
+  bug rather than fixed it, one layer down and out of sight of the screen
+  that used to hold the check. `OlderPage::Events(vec![])` means the start
+  of the conversation; `OlderPage::NothingLoaded` is not an answer about
+  the conversation at all. Reviewed 2026-09-06.
+  `paging_before_the_first_event_makes_no_request_at_all` asserts zero
+  transport calls, not just the variant, since a request that happens to
+  answer empty is exactly what caused the original bug, and
+  `a_failing_server_page_is_an_error_rather_than_an_empty_one` plus
+  `an_unreadable_cached_page_is_a_parse_error_rather_than_an_empty_one`
+  are the same rule for the two ways a page can fail.
+- **`fetch_transcript_lines`** (new in `api.rs`) hands back each line
+  paired with the exact server bytes it came from, via
+  `serde_json::value::RawValue` rather than re-serializing a parsed
+  `Value` -- the cache and a live SSE frame for the same event have to
+  agree byte-for-byte, which is exactly what the `serde_json`
+  float-rounding bug (AGENTS.md) was about. The existing
+  `fetch_transcript_page` is untouched (other callers under `iris/`
+  depend on its signature); the two share a `transcript_path` helper so
+  the query string is written in one place.
+
+**Not ported:** `EventStream.kt`'s reconnect-with-backoff loop, and
+`TranscriptSource.close`'s ability to cancel a live stream from another
+thread. Both are wall-clock/thread-lifetime policy that belongs to
+whichever runtime embeds this crate (iris's own timers, a Tokio task, a
+Kotlin coroutine scope), not to this pure logic -- `follow` is the same
+"write to the cache, then hand the frame to the caller" decorator
+`iris/desktop-app/src/app.rs` and `iris/android-app/src/transcript_client.rs`
+already hand-wrote around `event_stream::follow_session_events` before this
+existed; the cache write moved into one shared place so a third caller
+does not repeat it again by hand.
+
+## What is not started at all
+
+- **A full markdown AST.** `markdown_blocks` (2026-09-06) splits a message
+  into its *top-level* blocks -- heading, paragraph, fence, list, table,
+  quote -- with each block's own source, which is what a renderer needs to
+  lay out prose versus code and what lets a streamed delta re-lay out one
+  block instead of the message (docs/RUST.md's Task B). What it
+  deliberately does **not** build is the tree below that: nested list
+  items, table cells, inline spans. Inline styling is still the renderer's
+  own job per block (`iris/transcript-ui/src/markdown.rs`), and nothing
+  has needed the rest yet. `CodeFence.kt`'s use of `org.intellij.markdown`
+  for a full CommonMark AST is Compose rendering plumbing, not something
+  to port as-is.
+- **`TranscriptUnits.kt`** (see above) -- deliberately out of scope, since
+  it flattens a row into bounded units for a *specific* lazy-list
+  framework's composition cost, which is a fact about that framework
+  rather than about the transcript.
+
+## Verifying
+
+`./run-tests.sh` from the repo root now runs `event-model`, `client-core`
+and `server` in that order (each `cargo test`, forwarding arguments the
+same way it always has). From `client-core/` directly: `cargo test`
+(119 tests), `cargo clippy --all-targets`, `cargo fmt` -- all clean as of
+this writing (2026-09-06).

docs/DECISIONS.md

@@ -0,0 +1,444 @@
+# Decisions taken for Iris to review
+
+Short list of design choices made by the design agent without asking, so
+they can be judged and reversed later. Detail lives in RUST.md (and IRIS.md
+for iris API changes); this file is only the summary. Newest first. Items
+marked **DEFERRED** are ones the agent chose not to decide alone.
+
+## 2026-09-06 (how a tool call looks, P1b)
+
+- **A card that never got a result says "no result", in yellow, and it is
+  a state Compose cannot say.** A call that finished having printed
+  nothing and a call whose turn was interrupted before anything came back
+  both leave an empty output. Compose draws both as an ordinary finished
+  call, which reads as a fact somebody established. There are five states
+  now, each with a word and a colour: nothing at all for a call that
+  worked, "running" (grey), "your turn" (peach, Compose's own wording and
+  colour), "failed" (red), "no result" (yellow).
+
+- **A failed call is drawn as failed, which needed a field on the wire.**
+  `is_error` is on the CLI's `tool_result` and was being dropped; the
+  server now carries it to the phone. Reversible, but the alternative is a
+  card that says a call succeeded because it cannot tell.
+
+- **A group's cards do not each carry their own surface.** Compose gives
+  each card a fill and squares the corners where it faces a neighbour, so
+  a run reads as one object broken into parts. iris has no per-corner
+  radius, and -- more to the point -- a group built the way Compose builds
+  it hit a framework layout defect that drew every card's text a card
+  below its own box. So a group is one surface with its cards on it,
+  separated by a small gap, and the 4dp inset Compose holds them off the
+  edge by is gone. Worth revisiting once the layout defect is fixed
+  (docs/IRIS_TODO.md).
+
+- **A long tool output is capped at 80 lines or 4 kB with a "Show all N
+  lines".** Compose draws the whole thing, and gets away with it because
+  its `Text` inside a `LazyColumn` lays out lazily; here the output is one
+  text widget and shaping a hundred kilobytes of it costs what the file
+  editor's 32 kB limit was measured against. If iris's text gets cheaper,
+  this is the number to move.
+
+- **A card's command is clipped, not pannable, and its summary line is
+  clipped rather than ellipsised.** Both are framework gaps rather than
+  choices (`scrollable_on` on a non-editable text draws nothing; there is
+  no overflow ellipsis), and both are worse than Compose today. Named here
+  because they are visible.
+
+## 2026-09-06 (how a markdown block looks, P1a)
+
+- **A table is drawn as padded monospace columns, not as a grid.** Your
+  call to reverse. Compose draws a real grid: cells on a tint, each
+  column with a 136dp floor, scrolling sideways when there are too many.
+  iris has no grid widget, and building one would be a widget per
+  markdown feature -- which is the thing the block model exists to avoid.
+  In a monospace face a character count *is* a pixel width, so padding
+  each cell to its column's width is alignment, the widths are still
+  measured from the cells, and a table that is too wide pans sideways
+  through the same mechanism a code fence already uses. The header is
+  bold with a rule under it, and a long cell wraps inside its column
+  (capped at 28 characters, which is what fits three columns across a
+  phone). **What it trades:** no cell borders, and a table looks like
+  code rather than like a table. If you want the grid, it is a new widget
+  and it is a day's work.
+- **Three block frames, and only three.** A heading, paragraph and list
+  are plain text with spans; a fence and a table are a rounded panel that
+  does not wrap; a quote is a bar with the text padded past it.
+  Everything else markdown says is expressed in span styles, which cost
+  no widgets and no layout nodes. So a new markdown feature is a span,
+  not a widget.
+- **A list's marker is part of the text, so a wrapped item's second line
+  returns to the left margin.** Compose keeps it indented by giving the
+  marker its own column. Doing the same here needs per-line indent in
+  iris's text attributes; it is written down rather than done, because
+  the list items in a real reply are usually one line.
+- **A link opens on a tap and not on the end of a drag.** A press that
+  panned the transcript past a link, or that held long enough to start a
+  selection, does not follow it -- decided by the same gesture machine
+  that decides pan-versus-select, so there is one rule rather than two
+  that can disagree.
+
+## 2026-09-06 (composer scroll and the streaming block model)
+
+- **A streamed message becomes a column of per-block widgets.** Decided by
+  the design agent; recorded here because it is the shape of every message
+  on screen. A transcript row is one `TextEdit` today, so a streamed delta
+  re-shapes the entire message through parley on every event -- the stream
+  phase is the one place iris is behind Compose on your phone (p50 18.2ms
+  vs 13.4ms). A row becomes a column of one widget per markdown block
+  (paragraph, heading, fence, list, table) and a delta replaces only the
+  last block, keeping every earlier block's layout. **Rejected:** splitting
+  parley's layout at block boundaries inside one text widget (couples
+  iris's text widget to markdown structure, and parley has no incremental
+  API), and caching shaped runs per paragraph inside `TextEdit` (a second
+  cache with its own invalidation beside the glyph cache). Chosen because
+  P1's markdown block model is needed anyway, so the split happens once, in
+  `client-core`, and iris stays a text renderer. **Status: designed, not
+  built** -- this pass spent its budget on the composer's three layout
+  defects; docs/RUST.md has the design and the pass conditions.
+- **The composer's overflowing text now scrolls on a finger**, capped at
+  six lines and clipped to the bar. Reverses the "still does not scroll"
+  item below.
+- **A widget may not report a `dp` length** (see IRIS.md). A rule for
+  widget authors, enforced by a `debug_assert!`; nothing changes for app
+  code.
+
+## 2026-09-06 (stale-primitives and touch-scroll pass)
+
+- **A vertical drag inside a focused composer now scrolls rather than
+  selects.** Android's own `EditText` does this -- a vertical drag scrolls
+  the field, and only a long press starts a selection -- so the platform
+  decided it. What it costs: you can no longer drag straight down inside
+  the composer to select several lines of what you typed; use a long press
+  and then drag, or drag sideways. Say if that trade is wrong for you.
+- **`Scroll` gets a finger pan but no fling.** `List` flings; a scroll area
+  does not, because it has no per-frame tick to animate one and the areas
+  it wraps are at most a screenful (Android does not fling a six-line text
+  box either). Easy to add later if a scroll area ever wraps something long.
+- **The composer still does not scroll its overflowed text**, though the
+  mechanism it needs is now in place. Wrapping the field in `.scrollable()`
+  was tried and reverted the same day: `Scroll` measures its content and
+  container against the *window*, so inside the `MaxSize` that caps the
+  composer at six lines the two are in different spaces and the field pans
+  itself entirely out of the bar (measured on the emulator with 474
+  characters in it -- the bar collapsed to its padding). Fixing that means
+  `Scroll` measuring against its own offered box, which is a change to a
+  widget the transcript and the bench shell both use, so it is its own
+  piece of work rather than a rider on this one.
+
+## 2026-09-06 (defect pass)
+
+- **The keyboard-open diagnostics overlay is gone; the capture only
+  logs now.** It was added when `on_insets_changed` was not firing at all
+  and there was no way to get a report off the phone. It fires reliably
+  since the activity went edge-to-edge -- and what that looks like in
+  use is a full-screen report covering the app **every time the keyboard
+  opens**, with its own Copy/Close buttons sitting underneath the
+  keyboard, so it cannot be dismissed (reproduced on the emulator this
+  pass: two `tap 'CLOSE'` runs left it up). An interruption for something
+  nobody asked for, over the app you are trying to type into. The named
+  `Diagnostics` button still shows the same text on demand, and the new
+  `iris surface:`/`iris insets:` log lines carry the lifecycle a `logcat`
+  pull needs. Reversible: `capture_keyboard_diagnostics` is still the one
+  place this is decided, and `PlatformHandle::show_diagnostics_overlay`
+  is still there.
+
+- **The bench shell's report pane is sized to its report, not to a share
+  of the window.** It held `.height(rest(1))` beside the transcript's
+  `rest(2)`, so an *empty* `TextEdit` reserved a third of every screen --
+  which is what Iris's "the app does not start with keyboard spacing
+  correct" screenshot was showing, with the composer two thirds down and
+  black below it. It is `.max_height(dp(260))` now and sits above the
+  transcript rather than under the composer, where it was eating the
+  navigation-bar clearance. Cost: a filled report is clipped at 260dp
+  rather than scrolling (a `Scroll` there drew itself off the top of the
+  screen, since `Scroll` pins to the end of its content and reports its
+  content's full length to the parent -- worth fixing in `Scroll`, not
+  worked around here). "Copy report" and `logcat` still have the whole
+  thing.
+
+## 2026-09-05
+
+- **iris no longer asks every device for compute-shader limits it never
+  uses.** `adapter.request_device` (both `iris/src/android/render.rs` and
+  `iris/src/default/render.rs`) used `Limits::default()` plus an override
+  for `max_buffer_size`, and `Limits::default()` unconditionally requests
+  desktop-tier compute limits (`max_compute_workgroups_per_dimension:
+  65535`, per `wgpu_types`) even though nothing in `iris`/`iris-core`
+  creates a `ComputePipeline` or writes a `@compute` shader stage —
+  confirmed by grepping the whole tree, not assumed. That crashed
+  `request_device` outright on the Android emulator's software GL path
+  (`EMU_GPU=software`, `--features force-gles`): SwiftShader's GL reports
+  itself as OpenGL ES 3.0, which has no compute shaders at all, so the
+  adapter's real limit is 0 against the unconditional request for 65535 —
+  `RUST.md`'s "Software mode ... crashes for a third, different reason,"
+  2026-09-05, earlier today. The same would happen on any real
+  GLES-3.0-only Android device, not just the emulator. Fixed by a new
+  `iris_core::device_limits()` (`iris/core/src/render/mod.rs`), shared by
+  both platform backends so the two requests cannot drift, that zeros the
+  six `max_compute_*` fields explicitly rather than switching to a
+  downlevel `Limits` preset — `Limits::downlevel_webgl2_defaults()` was
+  considered and rejected: it also zeros
+  `max_storage_buffers_per_shader_stage`, and `shader.wgsl`'s vertex stage
+  reads four `var<storage>` buffers (rects, glyphs, masks, move_offsets),
+  so that preset would trade the compute crash for a bind-group-layout
+  one on the same downlevel hardware this is meant to support. No
+  capability check or fallback path was needed since nothing is being
+  disabled — the request is simply narrowed to what the pipeline actually
+  uses. `rigs/gpu-probe`'s own mirrored limits (it is deliberately its own
+  crate, not a workspace member, so it cannot call `device_limits()`
+  directly) were updated to match, and confirm `IRIS DEVICE: ok` against
+  this VM's own Vulkan and GL adapters. **Not verified this pass**: the
+  specific SwiftShader-ES-3.0 crash this fixes, on-device — the
+  `EMU_GPU=software` cold boot this needs would have force-restarted this
+  checkout's emulator while another session was actively running its own
+  app on it (`com.example.aiapp` had window focus at the time), so it was
+  left for a pass when the emulator is free rather than disrupting that
+  session. Everything reachable without the emulator is clean: `cargo
+  fmt`/`clippy --workspace --all-targets`/`test --workspace`, `cargo ndk
+  build`/`clippy` for `iris-android-app` with `force-gles`, and
+  `gpu-probe` against this VM's own Vulkan and GL(ES 3.2, which still has
+  compute and so would not have reproduced the crash even before this
+  fix — not a substitute for the real ES-3.0 test).
+
+- **P0's Compose half is built and smoke-tested on the emulator** — the
+  `bench` build type, the shared `app/bench-fixture/` transcript, and an
+  in-process fake backend (`BenchFixture.kt`/`BenchNetwork.kt`) that
+  answers `TranscriptSource`/`EventStream` from an in-memory event log
+  instead of a real server, so the fold and paging under test are the real
+  ones. Full account, the smoke run's report, and what is deliberately
+  left (the iris half, the real on-phone runs) are in RUST.md's P0 box.
+  Not a decision to review so much as the gate itself now being runnable —
+  flagged here because it is the first half of something Iris explicitly
+  asked to see before P1.
+
+- **P0's iris half is also built and smoke-tested on the emulator,
+  2026-09-05.** A new `bench` Cargo feature on `iris-android-app`, on top
+  of `transcript-screen`: the same checked-in fixture (`include_str!`, no
+  asset pipeline needed), the same 24-swipe scroll loop animated through
+  `List::scroll` and the same 400-event/20s streaming phase through
+  `fold_event`, "Run benchmark"/"Copy report" as named accessible
+  controls, and the same three added report fields (process CPU time,
+  peak RSS, battery current) via direct JNI calls
+  (`bench_jni.rs::PlatformHandle`) since `android_view` has no
+  `BatteryManager`/`ClipboardManager` wrapper of its own. One small public
+  API addition to get there: `AndroidAppState::platform_ready` (`IRIS.md`),
+  a default-no-op lifecycle hook handing an implementor a `JavaVM` +
+  `GlobalRef` it can call Java through from any thread. Packaged with a
+  new `release` build type on `iris-android-app`'s own Gradle project
+  (there was previously only `debug`), signed with the same key
+  `app/build-apk.sh` generates. Smoke run and the full report are in
+  RUST.md's P0 box; not attempted this pass: the real on-phone runs and
+  Iris's pass/fail call, which is the actual gate.
+
+- **The intermittent touch-scroll dropout is root-caused and fixed: a
+  missed `ACTION_DOWN` hit-test, not the previously-suspected coalesced
+  first `ACTION_MOVE`.** Diagnosed by temporary logcat tracing of every
+  touch event, `DragArbiter` state transition and `Selection::drag`
+  dispatch (removed once confirmed), reproduced on this checkout's own
+  emulator against a real sandbox session. The trace showed the actual
+  mechanism: a gesture's `ACTION_DOWN` lands wherever the finger actually
+  is, which is not guaranteed to fall inside the same row-local sensor
+  region a later `ACTION_MOVE` in the same gesture lands in (a row's own
+  padding/gap, or its non-selectable sender-name header, is
+  pointer-transparent to `iris::sense::CursorSense`). When that happens,
+  the widget that ends up handling the gesture never saw `PressStart`, so
+  `DragArbiter` sits in `Idle` — which answers every subsequent frame with
+  `Undecided` and has no way to tell "no press is happening" from "a press
+  is happening but I missed its start," so it never recovers on its own
+  for the rest of that gesture. One real trace showed exactly this: touch
+  `Down`/`Move`/`Up` all delivered correctly, but zero `PressStart`
+  reaching the arbiter, `state=Idle` unchanged from first frame to last.
+  Fixed at the call site that has the context to recover
+  (`iris::transcript_ui::selection::Selection::drag`,
+  `iris/transcript-ui/src/selection.rs`): a new `DragArbiter::is_idle()`
+  (`iris/src/sense.rs`) lets it notice a `Pressing` frame arriving with the
+  arbiter still `Idle` — which can only mean a missed `PressStart`, since a
+  `Pressing` sense requires the button to genuinely be down — and start the
+  press there instead of where it was missed. Three new unit tests in
+  `sense.rs`'s `drag_arbiter_tests` and one in `transcript-ui`'s
+  `selection::tests` (the latter fails on the code before this fix).
+  Commit follows. Not the same failure the earlier pass's `DECISIONS.md`
+  DEFERRED item speculated about (a coalesced first `ACTION_MOVE` skipping
+  slop detection) — that hypothesis is now ruled out; the arbiter's own
+  slop/long-press logic was never wrong. RUST.md's I5 box,
+  "Touch-scroll dropout root-caused, 2026-09-05" has the full trace.
+- **P0, a phone benchmark gate before any porting, asked for by Iris
+  2026-09-05**: "before P1 I'd like to see benchmarks & also maybe stress
+  test on my own phone ... If it doesn't match compose reasonably well then
+  I don't think I'd wanna continue." Design (RUST.md's P0 box has the
+  detail): the same embedded synthetic fixture in both apps with no server
+  needed; the same scripted scroll loop then a streaming phase, run
+  programmatically since the phone has no usable system tracing and no
+  agent can drive it; the same report from both (frames, janky %, p50/p90/
+  p99, process CPU time, peak RSS, battery current where readable) with a
+  copy button; the iris app under its own id and the Compose one as a new
+  `bench` build type with an id suffix, so neither replaces her production
+  install; two arm64 APKs plus instructions delivered under `~/host/bench/`.
+  The gate is hers: iris within a reasonable margin of Compose release on
+  p50, p99 and CPU time, no crashes, no visible stutter. If it fails, the
+  port stops.
+- **The rest of the port is one UI crate, `iris/app-ui`, grown out of
+  `iris/transcript-ui` rather than started beside it.** It holds a
+  `Screen` enum plus a back stack — the Rust equivalent of `AppRoot.kt`'s
+  `when` — and `iris/desktop-app`/`iris/android-app` become thin entry
+  points over it. Chosen over a fresh crate because `transcript-ui`
+  already has the right generic shape (`Rsc: HasEvents` +
+  `Rsc::State: FocusHost`) and the `client-core`/`event-model` path
+  dependencies every later screen needs, so growing it in place is the
+  smaller diff. Platform-only code (notification service, share target,
+  QR scanner, Keystore token, deep-link enrolment) stays in the E3/E5
+  Java shell (`android-shell/` + `app/shellApp`) rather than moving into
+  this crate, since none of it is a screen. The Android APK is built by
+  `cargo xtask apk` (E5), merging the app-ui cdylib into the E3 shell so
+  there is one app rather than a demo shell plus a service shell.
+  `app/androidApp` (the Compose app) stays untouched and is the baseline
+  every step is measured against, until parity is reached (P7 decides
+  the switch, and is itself a load-bearing decision left to Iris). Order
+  is by risk to the daily-use path: session screen first (P1, where
+  every hard behaviour already lives), then the shell merge and a real
+  phone install (P2), then root tabs (P3), the explorer (P4),
+  settings/enrolment (P5), desktop parity (P6), and the cutover itself
+  (P7). Full plan: RUST.md's "The port, in order (decided 2026-09-05)".
+- **iris gets its own measured frame report, rather than waiting on a
+  `dumpsys`/`gfxinfo` answer that cannot see a `SurfaceView`'s GPU-drawn
+  frames.** `iris_core::FrameReport` (`iris/core/src/render/frame_report.rs`)
+  times each frame's wall clock from the same point `render()`'s redraw
+  starts to just after `queue.submit` + `present()` — the span Compose's
+  own render report and `gfxinfo` both count — into a fixed 4096-entry
+  ring (no allocation per frame; `report()` is the only place that
+  allocates, and only on a button tap). The report gives total frames,
+  janky % over the same 16.7ms budget `gfxinfo` uses, P50/P90/P99 and the
+  worst, plus a reset. Exposed the way the Compose app's copy-button
+  report already is: two named controls ("Frame report", "Reset frame
+  report") on the transcript screen, tappable by accessibility name via
+  `ui-trace`, logging under this crate's fixed `android_logger` tag
+  (`iris-android-app`) so a script can grep `"iris frame report"` the way
+  `transcript-bench.sh` greps `"ai-app render report"`. The report's own
+  `Display` line says plainly that it measures up to the `present()` call
+  returning, not GPU/compositor completion — wgpu's `present()` is not
+  fenced against either, so presenting that span as "time to reach the
+  screen" would be a measured-looking number that is actually inferred,
+  which the standing UI rule forbids.
+- **`ui-trace` gains a hold-then-drag gesture, additive, in
+  `emulator-tools`.** Neither of its two existing actions can produce
+  "hold stationary for `LONG_PRESS`, then move without lifting" — `tap`
+  has no hold and `swipe X1 Y1 X2 Y2 MS` interpolates motion across its
+  whole duration from t=0. A new action presses, waits, then moves to a
+  second point and releases as one continuous touch (raw
+  `sendevent`/`MotionEvent` injection, extending whatever mechanism the
+  existing `swipe` already uses), so `DragArbiter`'s pan-vs-select rule
+  (`iris/src/sense.rs`, already covered by 8 unit tests against a
+  synthetic clock) can finally be driven on a real device instead of only
+  in a test harness.
+- **Touch drag on a transcript row follows Android's own rule**: a vertical
+  drag pans the list immediately; a stationary press held 500 ms starts a
+  text selection which further dragging extends; a horizontal drag while
+  something is already selected extends that selection without the wait.
+  One `DragArbiter` per list decides it (`iris/src/sense.rs`). Chosen over a
+  "text layer always wins" or "list always wins" rule because either loses
+  one of the two gestures a reader expects.
+- **E4's desktop shape is a new `iris/desktop-app` crate**: a winit window
+  holding `transcript-ui`'s screen beside a session list, talking to a real
+  `ai-server` through `client-core`. It enrols by pasting the same
+  `aiapp://enroll?…` link a phone scans (`client-core::config::EnrolledServer`)
+  and keeps it owner-only under `$XDG_CONFIG_HOME/ai-app-desktop/`. The
+  pinned CA is a path given on the command line, not baked in. Chosen so
+  the phone and desktop share one enrolment format and no second one is
+  invented.
+- **I5's Android integration extends `iris-android-app` (I2's shell)
+  behind a Cargo feature (`transcript-screen`), rather than a third
+  shell crate.** That project already has the Gradle module, the
+  `IrisView`/`MainActivity` Java, and the JNI registration; the only
+  thing a second screen needs on top is a different `AndroidAppState`,
+  the same axis `tabs_ui::build`/`transcript_ui::build` already vary
+  along on the winit side. `tabs-screen`/`transcript-screen` are
+  mutually exclusive and each pulls in only its own deps, so the plain
+  tabs build (I2/I4) is untouched.
+- **Order of remaining work, updated 2026-09-05**: the two in-flight
+  pieces and I5's Android integration are all done; next is giving iris
+  its own frame-timing report so item 3 below can be decided by a number.
+- **DECIDED by Iris, 2026-09-05: iris is the app's framework; Masonry was
+  the calibration.** Her words: "I think iris definitely makes more sense
+  based on the limitations we've found." The limitations: Masonry has no
+  touch scroll on Android (E2), no per-span rich text and no cross-row
+  selection on the pinned commit (E2), and its keyboard bridge is a TODO
+  (E1); iris carries the same screen under the Compose baseline on the
+  host GPU (p50 15.0 ms against Compose's 20.0 ms, RUST.md's I5 box). What
+  follows: the E-steps are closed as calibration, and the port proceeds
+  on iris — screens, the shell (E3/E5), and `client-core` underneath.
+  The item below is kept as the record of what she decided from.
+- **Was DEFERRED — whether to commit to iris over Masonry for `ai-app`.**
+  Updated 2026-09-05 with the clean comparison the recommendation wanted:
+  same sandbox session content, same emulator, `EMU_GPU=software`, one
+  session. Headline numbers (RUST.md's I5 box, "Clean scroll comparison,
+  2026-09-05," has the full table and every caveat):
+
+  | app | build | frames | janky % | p50 | p90 | p99 | worst |
+  |---|---|---|---|---|---|---|---|
+  | Compose (in-app report) | debug | 1102 | 99.0% late | 33.8ms | 50.6ms | 79.5ms | -- |
+  | Compose (`dumpsys gfxinfo`) | debug | 1499 | 21.15% (95.66% legacy) | 32ms | 48ms | 150ms (p99) | -- |
+  | iris (`FrameReport`) | **release** | 299 | 94.65% | 79.1ms | 98.6ms | 117.8ms | 212.6ms |
+  | iris (`FrameReport`, repeat) | **release** | 233 | 94.42% | 109.3ms | 130.8ms | 147.1ms | 150.5ms |
+
+  **Not a clean apples-to-apples reading, stated plainly rather than
+  smoothed over**: iris had to be built **release** (debug `SIGSEGV`s on
+  this emulator's Vulkan loader, I4's finding) against Compose's mandated
+  **debug** build, so this asymmetry likely *understates* iris's gap
+  rather than the reverse; the three frame-time sources measure different
+  things (Compose's own phase accounting vs. Android's HWUI deadline-miss
+  definition vs. iris's redraw-start-to-present window, the last of which
+  `dumpsys gfxinfo` cannot see at all for iris's `SurfaceView`); and both
+  figures are emulator numbers under software rasterisation, which
+  Compose's *own* in-app report shows already costs 20-34ms/frame in
+  `swap`+`gpu` alone under this GPU mode, so a same-mode iris number well
+  above 16.7ms was expected going in for either app. A second pair under
+  `-gpu host` was not taken this pass. The earlier session's suspected
+  intermittent touch-delivery dropout was **not reproduced** this pass —
+  the zero-frame results this time traced to this pass's own script bug
+  (a `cd` that changed which emulator `ui-trace` targeted), not the
+  emulator; a CPU-load rise during the gesture was observed by a sampler
+  running throughout, but did not correlate with any failure, so the
+  original candidate is neither confirmed nor ruled out.
+  The choice in front of Iris, updated: decide now on the
+  structural-plus-functional case already made (iris works end-to-end
+  where Masonry's scroll gesture doesn't exist at all on Android) plus
+  this table — reading the two build profiles and three jank definitions
+  with the caveats above rather than as a single number — or ask for a
+  same-profile, same-GPU-mode rerun first. RUST.md's I5 box has the full
+  account.
+
+  **Updated 2026-09-05, the `-gpu host` pair taken.** Real GPU rendering
+  (`force-gles` -- the default Vulkan backend has no adapter at all under
+  plain host-GPU boot, confirmed by the exact `wgpu` error) reverses the
+  software-mode shape:
+
+  | app | build | GPU mode | frames | janky % | p50 | p90 | p99 | worst | cpu p50 | gpu-wait p50 |
+  |---|---|---|---|---|---|---|---|---|---|---|
+  | Compose (in-app report) | debug | host (virgl) | 1268 | 96.4% late | 20.0ms | 28.4ms | 37.7ms | -- | -- | -- |
+  | iris (`FrameReport`), **best of three, 2026-09-05** | release, `force-gles` | host (virgl) | 439 | 46.24% | 15.7ms | 23.3ms | 31.2ms | 57.4ms | 1.2ms | 13.2ms |
+
+  Under real GPU rendering iris's median frame is *faster* than
+  Compose's, not the 2-3x-slower shape the software-mode table shows. A
+  new split inside `FrameReport` (redraw-to-submit vs. submit-to-present,
+  commit `e2a1fad`) says why: iris's own CPU work per frame is a median
+  ~1ms -- almost the entire frame is time spent handing the frame to the
+  driver, not in iris's layout/text/primitive code. This is consistent
+  with the earlier software-mode gap being mostly SwiftShader's CPU
+  rasterisation cost rather than an iris-specific slowness. **Still not
+  proof, and now closed as unanswerable rather than merely untaken**: a
+  same-mode software `force-gles` run to isolate the backend was retried
+  2026-09-05 after fixing the compute-limit crash the first attempt hit,
+  and hit a second, structural wall instead — SwiftShader's ES 3.0 GL
+  path has no storage-buffer capacity at all, and `shader.wgsl` reads
+  `var<storage>` buffers unconditionally, so reaching that path needs a
+  shader rewrite, not a limits fix (RUST.md's I5 box, "The three
+  remaining I5 verifications, closed 2026-09-05," item 2). The
+  intermittent touch-scroll dropout this pass also reproduced is
+  root-caused and fixed as of the same date (a missed `ACTION_DOWN` on a
+  row's padding/header left `DragArbiter` stuck in `Idle`); three clean
+  `iris-scroll.sh` runs post-fix each scrolled all 24/24 swipes, replacing
+  the single-attempt 62-frame reading this table used to carry. RUST.md's
+  I5 box, "Where iris's frame time goes, 2026-09-05, the `-gpu host`
+  pass," and "The three remaining I5 verifications, closed 2026-09-05,"
+  have the full account. The iris-vs-Masonry choice itself is still
+  Iris's to make.

docs/IRIS_TODO.md

@@ -0,0 +1,957 @@
+# iris: known problems and things still to build
+
+Iris's own list for the library, recorded 2026-09-04 in her words where it
+matters, so the agents working through RUST.md pick these up in a sensible
+order rather than rediscovering them. Each item says where it sits in the
+order and what "done" looks like. Tick and date them in place.
+
+## Fix
+
+- [x] **`request_device` asked for compute-shader limits it never uses
+  (2026-09-05).** `Limits::default()` (both `iris/src/android/render.rs`
+  and `iris/src/default/render.rs`) requests desktop-tier compute limits
+  unconditionally, even though nothing in `iris`/`iris-core` creates a
+  `ComputePipeline` or writes a `@compute` shader stage — confirmed by
+  grepping the whole tree, not assumed. That crashed device creation
+  outright on the Android emulator's software GL path (`EMU_GPU=software`,
+  `--features force-gles`): SwiftShader's GL reports itself as OpenGL ES
+  3.0, which has no compute shaders, so the adapter's real limit is 0
+  against the unconditional request for 65535 — the same would happen on
+  any real GLES-3.0-only Android device. Fixed by a new, shared
+  `iris_core::device_limits()` (`iris/core/src/render/mod.rs`) that zeros
+  exactly the six `max_compute_*` fields rather than switching to a
+  downlevel `Limits` preset — `downlevel_webgl2_defaults()` also zeros
+  `max_storage_buffers_per_shader_stage`, which `shader.wgsl`'s vertex
+  stage needs (four `var<storage>` buffers), so that preset would trade
+  this crash for a bind-group-layout one on the same hardware.
+  `rigs/gpu-probe`'s own hand-mirrored `Limits` (it is deliberately its
+  own crate, not able to call `device_limits()` directly) was updated to
+  match. See `DECISIONS.md` and RUST.md's I5 box for the account,
+  including what could not be re-verified on-device this pass (the
+  emulator was in concurrent use by another session).
+
+- [x] **Input does not fall through by input type (2026-09-04).**
+  `SensorUi::run_sensors` (`src/default/sense.rs`) used to set "consumed,
+  stop checking lower layers" from mere hover — a widget registered for
+  nothing but `click()` blocked a `Scroll` meant for whatever was behind
+  it, since "the cursor is over this widget" and "this widget handled the
+  event" were the same check. Fixed by judging consumption per input
+  kind: with no button transition and no scroll happening this frame
+  ("momentary" activity), the topmost hovered widget still wins, same as
+  before; when something momentary *is* happening, only a widget whose
+  registered senses actually include a matching non-hover one (checked
+  via a new `TypeEventManager::registered`, which lists what a widget
+  registered without running anything) consumes it, so a widget with only
+  `Hovering`/click handlers can no longer block a scroll from reaching a
+  list underneath. `iris/src/sense_tests.rs` builds a button-over-a-list
+  `Stack` with a plain `HasEvents` impl (no GPU or window) and checks both
+  directions: a scroll over the button reaches the list, and a real click
+  still reaches the button — confirmed to fail on the pre-fix code and
+  pass after.
+
+- [x] **Appending one image to an already-loaded list rebuilds every other
+  image's bind group (2026-09-05, fixed 2026-09-05).** Found by the
+  benchmark below: `GpuTextures::update` (`core/src/render/texture.rs`)
+  triggered `rebuild_image_bind_groups` — a loop over *every live
+  standalone image*, rebuilding its `BindGroup` — whenever the shared
+  `masks` or `move_offsets` GPU buffer was resized (`masks_resized ||
+  moves_resized` in `UiRenderNode::update`, `core/src/render/mod.rs`), and
+  a widget getting its *first* move-offset slot (LAYOUT.md section 2 —
+  every widget gets one on first draw) could be exactly what grows that
+  buffer. So one new message with one new image, appended to a transcript
+  that already has N images loaded, did not cost O(1): it cost one
+  `create_image` for the new image plus one `make_image_bind_group` per
+  *existing* image, because the new widget's own move slot pushed the
+  arena past its capacity. Measured directly in
+  `iris/examples/bench_images.rs`: appending a 1,001st image to 1,000
+  already-settled ones reported **1,001** bind-group creates for that one
+  frame, not 1 (`./run-bench.sh images`, frame 5 in the transcript below).
+
+  **Fix**: `masks`/`move_offsets` never belonged in a standalone image's own
+  bind group (group 2) in the first place — the group also holds that
+  image's own texture view, which is the only thing that is genuinely
+  per-image, so a buffer shared by *everything* forced a rebuild of
+  *every* group the moment it moved. Gave masks/move_offsets their own
+  bind group (group 3 in `shader.wgsl` and `UiRenderNode`: `masks_layout`/
+  `masks_group`), bound once per frame in `UiRenderNode::draw` rather than
+  once per draw call, instead of duplicating them into every per-image
+  group. `GpuTextures` and its image bind groups now know nothing about
+  either buffer — `rebuild_image_bind_groups` is called only from
+  `grow_array` (the atlas array texture growing, which genuinely does
+  change what every image's own bind group must reference) — so a
+  masks/move_offsets resize now touches exactly one bind group, ever,
+  regardless of how many images are live. Numbers after the fix, same
+  benchmark and command:
+
+      ./run-bench.sh images
+      frame=1 bind_group_creates=1000   (cold load, unchanged)
+      frame=2 bind_group_creates=0      (was 1000 -- see the item below)
+      frame=3 bind_group_creates=0
+      frame=4 bind_group_creates=0
+      (append one image here)
+      frame=5 bind_group_creates=1      (was 1001)
+      frame=6 bind_group_creates=0
+
+  `run-headless.sh tabs --shot` still 27266 bytes, byte-for-byte unchanged,
+  confirming the bind-group restructuring changed nothing about what is
+  drawn.
+- [x] **Bind-group creation takes two frames to reach the steady state, not
+  one (2026-09-05, closed by the fix above, 2026-09-05).** Same benchmark:
+  loading 1,000 images cold used to report 1,000 creates on frame 1
+  (expected — `create_image`, one per new image) *and again* 1,000 on
+  frame 2, before settling to 0 from frame 3. This was `rebuild_image_bind_groups`
+  firing a second time for the same masks/move-offsets buffer-growth
+  reason as the item above, confirming the guess recorded here — the two
+  were exactly the same root cause measured two different ways. Frame 2
+  now reports 0 (see the numbers above); not a separate fix.
+
+- [ ] **A read-only text display has no widget of its own — P0's bench
+  report area is a `TextEdit` standing in for one (2026-09-05).** The only
+  way to get selectable text on screen today is `.editable(...)` plus
+  `.attr::<Selectable>(())` (`Selectable` is only implemented for
+  `TextEdit`, `iris/src/attr.rs`), which also makes the field focusable —
+  tapping the bench report opens the soft keyboard over text nothing lets
+  you type into. Harmless for a bench-only debug screen (not fixed this
+  pass), but a real "selectable, not editable" text primitive would
+  remove the keyboard side effect and is worth having before another
+  screen wants the same thing (P1's own transcript rows already read
+  their content from a `TextEdit` for the same reason).
+
+## From the phone, 2026-09-06
+
+Found on Iris's own phone while working RUST.md's P0 box's phone-report
+follow-ups. Recorded here rather than fixed in that pass, so a follow-up
+agent takes them without colliding with that pass's `bench_client.rs`/
+`android/view.rs`/`android/sense.rs` changes.
+
+- [x] **Swiping has no momentum, fixed 2026-09-06.** `List::fling`/
+  `VelocityTracker`/`FlingCalculator` (`iris/src/widget/list.rs`,
+  `iris/src/sense.rs`) -- IRIS.md's 2026-09-06 entry has the full account.
+  Wired through `Selection::drag`'s release path, cancelled by the next
+  touch-down, clamped at the loaded content's start/end. Verified by unit
+  test (fling distance against the closed-form spline result, cancel-on-
+  touch, the clamp), not yet by an on-device or emulator feel-check --
+  that is still open.
+- [x] **Scrolling down sometimes jitters the text, fixed 2026-09-06.**
+  Root-caused by reading `DragArbiter::update`'s `Undecided`-to-`Panning`
+  transition rather than by an on-device trace (no emulator was used this
+  pass): it was the first named suspect, not the second. `self.last` stays
+  at the press origin for every `Undecided` frame (nothing pans while the
+  gesture might still be a selection), so the frame that finally crosses
+  `DRAG_SLOP` returned `Pan(dy)` with `dy` measured from `press_start` --
+  the *whole* pre-threshold drag, applied to the list in one step, however
+  many frames it had taken to get there. Fixed by applying only the
+  excess past `DRAG_SLOP` on that one frame (`dy - DRAG_SLOP.copysign
+  (dy)`), the same "consume the slop, don't replay it" rule Android's own
+  touch handling follows. New regression test,
+  `crossing_the_slop_by_a_little_pans_by_a_little` (`iris/src/sense.rs`).
+  **Not yet done**: an emulator trace of the real per-frame offset
+  confirming this was the whole story on real touch input rather than
+  only the arbiter's own unit tests -- worth a follow-up pass before
+  calling it fully closed.
+- [x] **Composing text held back until a space, caret not moving, fixed
+  2026-09-06.** `InputMethodManager.updateSelection` was never called --
+  see IRIS.md's 2026-09-06 entry and RUST.md's P0 box, item 1, for the
+  full account and the emulator evidence.
+- [x] **Swipe over the composer summons the keyboard, fixed 2026-09-06.**
+  `Selector`/`Selectable` now wait for a completed tap -- see IRIS.md's
+  2026-09-06 entry and RUST.md's P0 box, item 5. Verified via `dumpsys
+  input_method`'s `mInputShown` on the emulator, not yet on the phone.
+- [x] **Text disappears again after leaving and returning to the app,
+  fixed 2026-09-06.** `GlyphAtlas::clear`/`Textures::reset` on a
+  genuinely new renderer -- see IRIS.md's 2026-09-06 entry and RUST.md's
+  P0 box, item 4. Verified on the emulator (home, reopen, screenshot);
+  not yet on the phone.
+- [x] **Composed/typed text never becomes visible at all -- root-caused
+  and fixed 2026-09-06.** Not the renderer at all: **the composer's buffer
+  was empty the whole time.** `TextEditCtx::select` (`iris/src/widget/
+  text/edit.rs`) compared the tap against the *laid-out text's* box and
+  set `selection = None` for anything outside it -- and an empty field's
+  layout is a zero-width box, so tapping an empty composer granted focus
+  and opened the keyboard while leaving no caret; `insert_str` returns
+  early with no caret, so every keystroke after that was dropped in
+  silence. Gboard's suggestion strip is its own composing state, not a
+  read of our buffer, which is what made the earlier pass conclude the
+  buffer held the text. Fixed by letting parley clamp a tap outside the
+  layout to the nearest cursor position (a press that reaches `select`
+  has already been hit-tested to the widget, so there is no "outside"),
+  plus a `debug_assert!` in `insert_str` so an insert with no caret fails
+  at the mistake instead of dropping input -- it immediately caught
+  `layout_tests::composing_text_after_a_keyboard_resize_...` typing into
+  an unfocused field. Three new tests in `edit.rs`
+  (`tapping_an_empty_field_places_a_caret_so_typing_lands`,
+  `tapping_past_the_end_of_the_text_clamps_to_the_end`,
+  `dragging_without_a_previous_selection_selects_nothing`); the first
+  fails on the pre-fix code. Emulator evidence: `adb shell input text`
+  after `tap 'Message'` now shows the text in the bar
+  (`/tmp/final-typing.png`) and logs `iris text render: chars=5 ...
+  glyphs=5`, against `glyphs=0` on every keystroke before.
+
+  **The old, superseded diagnosis, kept because it was wrong in an
+  instructive way:** The composer bar stays empty even once the
+  buffer genuinely holds the typed text (confirmed indirectly: Gboard's
+  own suggestion strip reacts correctly to each keystroke). A new unit
+  test proves the widget tree's own layout math resolves the field's
+  region correctly across a keyboard resize, so the bug is downstream of
+  that -- most likely `UiRenderState::redraw`'s single-widget redraw path,
+  or specific to this emulator's forced `force-gles` backend (untested on
+  Vulkan or the real phone). RUST.md's P0 box, item 2, has the full
+  writeup, what was ruled out, and where to look next. **Also unverified
+  because of this**: item 3's composer rebuild (one `Stack`-based widget,
+  a capped/scrollable height, bottom padding tied to the IME/nav-bar
+  inset) -- structurally in place and unit-tested, but its own visual
+  correctness cannot be screenshotted until text actually renders.
+- [x] **The composer has no touch-drag scroll for overflowing text.**
+  **Done 2026-09-06.** `field.scrollable().masked()` in
+  `transcript-ui/src/composer.rs`: a finger drag inside the bar pans the
+  message, the bar stays capped at six lines, and a vertical drag in the
+  focused field no longer extends a selection (Android `EditText`'s own
+  behaviour). Verified on this checkout's emulator with the
+  `transcript-screen bench force-gles` debug build -- six repetitions of a
+  13-word sentence typed in, then
+  `ui-trace record --do "swipe 540 1200 540 1460 300"`: the field's
+  `Message` box moved `31,1041..1048,1509` -> `31,1131..1048,1651` (the
+  content panned down with the finger) with its **height unchanged at
+  468px** (the bar did not grow), and the two screenshots either side show
+  different text in the same band.
+  Three real defects had to be fixed first, each with a headless
+  regression test in `iris/src/layout_tests.rs` and each confirmed to fail
+  without its fix (docs/RUST.md's plan box has the measurements):
+  a `MaxSize` reporting its cap as an unresolved `dp` (`Len::fold_dp`), a
+  `Masked` allocating a fresh mask slot per draw (`ActiveData::own_mask`),
+  and a panned widget's own hit box moving twice (`move_applied`).
+  `Scroll` itself turned out to measure the right number by a misleading
+  route -- it is written against `painter.px_size()` now, and the claim
+  below that it "measures against the window" was wrong.
+  **The grey background was not missing** -- that note (written here on
+  2026-09-06 and repeated as still open) is withdrawn. Re-measured the
+  same day on the same AVD by decoding the screencap rather than reading
+  it: the bar is `rgb(41,40,49)`, the declared `UiColor::new(40, 40, 46)`
+  after sRGB rounding, **full width and y2245..y2365** on 1080x2424, with
+  the field at `31,2277..1048,2329` and the 63px nav strip below it. It
+  is dark by design and sits on black, which is very likely what the
+  earlier reading was: at a glance the band and the background are hard
+  to tell apart. If it should read as a bar rather than as a slightly
+  different black, the colour is the thing to change, not the tree.
+
+## From the phone, 2026-09-06, 11:39 (build delivered 02:07, commit 543f6d9)
+
+Iris's report on the build with the composing-text, tap-vs-swipe and
+atlas-reset fixes, with a screenshot, verbatim. Each is open until an
+agent ticks it here with the evidence.
+
+- [x] **"The app definitely does not start with keyboard spacing
+  correct. This is how it looks without me doing anything initially."**
+  **Not an inset bug at all -- fixed 2026-09-06.** The black third is the
+  bench shell's own empty *benchmark report* pane: `bench_client.rs`'s
+  root tree gave it `.height(rest(1))` beside `content.height(rest(2))`,
+  so an empty `TextEdit` reserved a third of the window at every launch
+  and pushed the composer up by exactly that. Measured on this checkout's
+  emulator at the phone's own size (1080x2424, density 420, gesture nav),
+  which reproduced Iris's screenshot exactly: new `iris insets:` log line
+  reported `bottom=63 ime_bottom=0` at launch (a nav bar, no keyboard --
+  so the inset the composer was fed was never large), while `ui-trace
+  show -m Message --field box` put the field at `31,1488..1048,1540` on a
+  2282px-tall surface, 789px clear of the bottom -- that pane's third.
+  **Unit mixing checked explicitly and cleared**: `set_bottom_inset` takes
+  physical px and stores `Len::abs`, `MainActivity.java`'s `1`/`0`
+  `ime_bottom` only ever reaches `insets.bottom.max(ime_bottom)` and
+  `> 0.0`, and every `dp` in the composer resolves at layout time. Fix:
+  the report pane is sized to its content (`.max_height(dp(260))
+  .scrollable()`), and moved above the transcript so it cannot eat the
+  composer's nav-bar clearance. After: field box `31,2277..1048,2329`,
+  grey bar ending at device y2361 with the 63px nav strip below it
+  (`/tmp/fix1.png` this pass).
+  The screenshot shows the composer bar (the grey band) sitting about
+  two thirds of the way down a 704x1568 screen, with black below it to
+  the bottom, and the transcript ending at "Claude / Results" just above
+  it -- at launch, no keyboard. So the composer's bottom padding, which
+  the 2026-09-06 rebuild tied to the IME/nav-bar inset, is being fed a
+  large value at start on the phone. Suspects, in order: the initial
+  inset delivery on the phone (GrapheneOS, gesture navigation) versus
+  the emulator; `ime_bottom` now carrying a `1`/`0` boolean through a
+  field the composer may still read as pixels or dp; a stale value from
+  before the first `on_insets_changed`. Reproduce with the phone's
+  screen size and density on the emulator before guessing.
+- [~] **"Swiping still gets caught by the grey bar but keeps working
+  after I go past it."** Improved 2026-09-06 by the focused-field rule
+  below, still needs her phone to close. `attr.rs`'s `on_press` treated an
+  already-focused composer as the plain drag-to-select case, so a swipe
+  starting inside it dragged a highlight through the typed text for the
+  whole gesture; it now abandons that the moment the press passes
+  `DRAG_SLOP` vertically (Android `EditText`'s own rule), which removes one
+  of the two things that made the bar feel like it caught the swipe. The
+  residual `DRAG_SLOP` measured from the boundary crossing, described
+  below, is unchanged. Original note follows.
+  Not closeable from the emulator, annotated
+  2026-09-06 after the `DragGesture` merge. `attr.rs`'s `on_press` never
+  calls `capture_pointer` and never consumes a `Pressing` frame past
+  `DRAG_SLOP` (it just stops watching), so once the finger's *current*
+  position leaves the composer's box and enters the list's, `List`
+  starts receiving ordinary hit-tested `Pressing` frames there --
+  `DragArbiter::is_idle()`'s 2026-09-05 recovery (a missed `PressStart`)
+  picks it up rather than leaving it stuck. What this does **not** do is
+  what "wherever it began" implies literally: `DragArbiter::press_start`
+  restarts from the *boundary-crossing* position, not from the original
+  touch-down inside the composer, so the pan still needs a fresh
+  `DRAG_SLOP` of travel measured from the boundary rather than from the
+  start of the gesture -- composer and list are adjacent, non-overlapping
+  widgets (`lib.rs`'s `(list, composer_bar).span(Dir::DOWN)`), and only
+  the composer forwarding its own drag to the list would remove that
+  residual slop entirely, which is more than this pass's merge changes.
+  RUST.md's merge-pass box has the reasoning in full and an emulator
+  swipe confirming the composer's own box never moves/resizes during it;
+  whether the residual slop is still perceptible as "caught" needs Iris's
+  phone, since the emulator's per-widget boundary is a few dp wide and
+  easy to cross without noticing on a real screen too.
+- [ ] **"Flinging still does not work."** No longer expected to reproduce
+  after the `DragGesture` merge (`e12c708`, pointer capture +
+  `CursorSense::Drop`), 2026-09-06. Emulator evidence (RUST.md's
+  merge-pass box, check (b)): a real `ui-trace` finger swipe followed by
+  screenshot-hash sampling caught a post-release frame distinct from the
+  drag's own last frame in one run, and every run showed 28-32
+  `render()` frames per gesture against an idle baseline of 0 and ~8
+  expected from the drag alone -- redraw kept being requested well past
+  the finger lifting, which only happens while a fling is still
+  animating. Left unticked in spirit until Iris's phone confirms it,
+  since only she can say whether it *feels* like a fling now; the
+  emulator's screenshot timing could not always catch the tail of a
+  fast-settling one visually (same caveat noted in RUST.md).
+- [~] **"Text still disappears if I leave and come back to the app."**
+  **Instrumented 2026-09-06 so the phone can answer it**, since no
+  emulator here has a Vulkan adapter. `iris/src/android/view.rs` now logs
+  one `log::info!` line per surface event with the glyph/atlas counts:
+  `iris surface: surface_destroyed, tearing the renderer down
+  (glyphs_cached=387 atlas_pages=1)`, `iris surface: surface_changed
+  1080x2424 already_live=false glyphs_cached=387 atlas_pages=1`, `iris
+  surface: new renderer built (Gl), clearing glyph atlas: glyphs=387
+  pages=1`, plus `iris insets: ... window=(1080, 2424)` on every insets
+  change. That is the emulator's own healthy app-switch cycle, verified
+  this pass (home, reopen, screenshot: all text intact,
+  `/tmp/appswitch.png`). **The one line to look for on the phone is
+  `already_live=`**: `true` on the return from backgrounding would mean
+  the surface came back *without* a `surface_destroyed`, so
+  `surface_changed` reconfigured a renderer whose Vulkan swapchain and
+  atlas textures belong to a window that is gone -- the reuse branch
+  never clears the atlas, by design. `false` with no `new renderer built`
+  line after it would mean the renderer failed to rebuild. Either answer
+  names the fix; guessing between them from here does not.
+  The `GlyphAtlas::clear`/`Textures::reset` fix was verified on the
+  emulator under `force-gles` only; the phone runs Vulkan. So either the
+  reset is not reached on the phone's path (a different surface-
+  lifecycle sequence -- `surface_destroyed`/`surface_created` ordering,
+  or the renderer not being rebuilt but its textures lost), or the CPU
+  glyph cache and the GPU atlas still disagree after it. Needs logging
+  of the renderer lifecycle on the phone build, readable from `adb
+  logcat` when Iris next runs it, since no emulator here has a Vulkan
+  adapter under host GPU.
+
+## From the phone, 2026-09-06, 22:16 (build from 20303e0, delivered via ai-app-bench 95e25fe)
+
+Iris's report, verbatim, with a screenshot. Phone: Mali-G715 (Vulkan),
+`content_scale: 2.55`, 120Hz. Open until ticked with phone-side evidence.
+
+- [ ] **"Fling still doesn't work."** -> on `ed04d4c`, 2026-09-07:
+  *"flinging now does technically do something, but it seems to just be
+  linear velocity with an abrupt stop."* **It was exactly that, and the
+  arithmetic said so.** `distance_fraction(t)` returned `t` for every `t`
+  -- a constant-speed slide for the whole duration, then a stop at full
+  distance -- because two halves of AOSP's spline build loop were
+  transposed, which made `SPLINE_POSITION` and `SPLINE_TIME` identical, and
+  the lookup bracketed `t` between `SPLINE_TIME` entries rather than
+  between even time steps. The two cancelled to the identity. Ported
+  exactly now from `OverScroller.java` and
+  `androidx.compose.animation:animation:1.12.0`'s `SplineBasedDecay.kt`
+  (they agree line for line), with `iris/benches/fling_spline_reference.py`
+  as an independent transcription supplying the numbers the tests assert
+  on. Emulator, 2026-09-07: a released `v=3750` decelerates
+  `3746 -> 2624 -> 1834 -> 1144 -> 752 -> 449 -> 243 -> 83px/s` across 32
+  frames; a flick into the end of the list stops there in one tick with no
+  overshoot; a tap 200ms into a fling ends it at 11 ticks instead of 32.
+  **Open until the phone says so** -- a flick should now visibly slow
+  before it stops. Its earlier three defects (the velocity, the missing
+  animation registration, the 56x coefficient) are all still fixed and were
+  never the linear part.* Second report; the emulator's
+  `ui-trace` swipe flings (verified 2026-09-06 with `render()` counts),
+  a finger on the phone does not. What differs: a real flick at 120Hz is
+  batched by Android into few `MotionEvent`s with *historical* samples
+  (`getHistoricalX/Y/EventTime`), and can be DOWN, one or two MOVEs, UP
+  inside `DRAG_SLOP`'s worth of frames; a `ui-trace` swipe is many
+  evenly-spaced MOVEs. Suspects, in order: `android/sense.rs` reading
+  only each event's final position (the velocity tracker sees two
+  samples, or one); the release path starting a fling only from a
+  gesture already in `Panning`, so a flick that crosses the slop on its
+  last sample is treated as a tap; `ACTION_CANCEL`/pointer-capture
+  delivering no `Drop`. Log the release decision (samples, span,
+  velocity, outcome) at `info` so the next logcat settles it.
+- [x] **"I can't reopen keyboard by tapping on message box after it
+  already happened once."** *(Fixed 2026-09-07: `attr.rs`'s already-
+  focused branch calls `focus_gained` on a tap inside `DRAG_SLOP`.
+  Emulator: first tap `mInputShown=true`, back gesture, second tap
+  `mInputShown=true`. Negative control with that one call removed leaves
+  the second tap at `false`; a horizontal and a vertical swipe over the
+  focused field both leave it at `false`, so the earlier "swiping over
+  the input bar brings up the keyboard" has not returned.)* The field stays focused after the keyboard
+  is dismissed (back gesture, or the IME's own hide), so `on_press`'s
+  already-focused branch never requests the IME again. Android's
+  `EditText` shows the IME on every tap of a focused field; do the same
+  (`FocusHost`: a tap on a focused field requests the IME, idempotent
+  when it is already shown).
+- [ ] **"Message box does not push up the scroll area."**
+  **Reopened by the phone on 2026-09-07** -- *"similarly, the keyboard
+  raising up does not push things upwards"* -- after being ticked on
+  emulator evidence the day before (`ime_bottom=883`, composer box
+  `31,2277..1048,2329` -> `31,1457..1048,1509`). The JNI half was right;
+  what was wrong is one line of `iris/android-app/app/build.gradle`:
+  **`targetSdk = 34`** against `compileSdk = 37`, while the Compose app in
+  `app/` targets 37 and *does* push up on her phone. Below target 35 the
+  window keeps the legacy behaviour, where `adjustResize` shrinks it for
+  the IME and `getInsets(ime()).bottom` therefore measures zero;
+  `setDecorFitsSystemWindows(false)` opts out of that and still takes on
+  the API 36 emulator here, which is why every test run passed. Now
+  `targetSdk = 37`, plus a `WindowInsetsAnimation.Callback` for the devices
+  where only the animation path carries the height -- which also makes the
+  push-up animate (`ime_bottom=509, 663, 833, 881, 883` instead of one
+  jump). **This is a reading, not a measurement**: no Android 17 device is
+  reachable from here. So the Diagnostics pane now prints
+  `insets: dispatches=N left=… ime_bottom=… ime_visible=…` --
+  **screenshot that line with the keyboard open.** `ime_bottom` in the
+  hundreds and the composer risen means fixed; `dispatches` climbing with
+  `ime_bottom=0` means the reading was wrong and the window is still being
+  resized; `dispatches=0` means the listener is not firing at all, which is
+  a third thing again.* Since
+  `MainActivity` went edge-to-edge (`e12c708`), `adjustResize` no
+  longer resizes the window, so the app owns the IME inset -- but
+  `ime_bottom` is passed through JNI as the boolean `1`/`0` (the
+  2026-09-06 "(b)" fix), so nothing has the inset's *height* to pad the
+  transcript and composer with. Pass both: `isVisible(ime())` and
+  `getInsets(ime()).bottom` in px; the list's bottom padding and the
+  composer's position follow the height, the visibility drives the
+  boolean the `imePadding` rule in AGENTS.md's "Things that have bitten"
+  describes.
+- [x] **"Picture is what happens if I leave the app and come back,
+  which completely removes text, and then I tap on the debug info. The
+  textures are definitely getting cooked for some reason after leaving
+  the app and resuming."** Screenshot: every glyph drawn *before* the
+  resume is fragments; the diagnostics text drawn *after* is perfect;
+  the report says `atlas format: Rgba8Unorm, views live: 0`. Reading:
+  `Textures::reset`/`GlyphAtlas::clear` on the new renderer emptied the
+  GPU atlas, but the per-widget cached text primitives (`TextView`'s
+  render cache -- the one `c3cfc67`'s shape counter is keyed on) still
+  carry the old atlas coordinates and are re-submitted as-is; only
+  widgets drawn fresh after the resume shape and upload again. Fix: a
+  renderer rebuild invalidates every cached text render (one
+  generation counter on the atlas, checked at `TextView::render`, or
+  a full-tree redraw with caches dropped), with a `debug_assert!` that
+  no submitted glyph quad references an atlas generation older than the
+  live one. Reproducible on the emulator by forcing a renderer rebuild
+  (home + return, or `surface_destroyed`/`surface_created`) on a screen
+  with text already drawn -- the earlier "verified" home/reopen check
+  screenshotted the emulator's GLES path, where a resume may not
+  destroy the surface at all.
+
+  **Fixed in `ba2afba` and confirmed on the phone (Iris, 2026-09-07:
+  "the resume glyph corruption is fixed").** Closed. The emulator could
+  never have settled it -- no Vulkan adapter here, and the GLES path may
+  not destroy the surface at all -- so the phone was the only place this
+  could be answered, and it has been. `clearing_the_atlas_re_renders_
+  cached_text_instead_of_reusing_it` is what keeps it.
+
+  The reading above is right and the mechanism is one step narrower than
+  "cached text primitives". `IrisViewPeer::surface_changed`
+  (`iris/src/android/view.rs`) *does* already force a full-tree redraw
+  after a rebuild: it calls `render.resize(...)` unconditionally, which
+  sets `UiRenderState::resized`, which makes the next `update` take
+  `redraw_all` rather than `redraw_updates`. So every widget's `draw`
+  really does run again after the resume. What survives it is one cache
+  further in: `TextView::render` (`iris/src/widget/text/mod.rs`) returns
+  its cached `RenderedText` whenever the wrap width, buffer and attrs are
+  unchanged -- true of every pre-resume row -- so `TextData::place` is
+  never reached, nothing is re-rasterised into the fresh atlas, and the
+  *old* atlas's `uv_min`/`uv_max`/`layer` are re-submitted verbatim. Only
+  text whose content changed after the resume (the diagnostics pane Iris
+  tapped) re-shapes, which is exactly the split in her screenshot.
+  `Painter::glyphs` has one call site in the whole workspace, that one,
+  so there is no second holder of a `RenderedText` to fix.
+
+  The fix, in `ba2afba`: `GlyphAtlas::generation`, bumped by
+  `GlyphAtlas::clear`; `RenderedText::generation` recording which atlas
+  its glyphs were placed against; `Painter::atlas_generation()`;
+  `TextView::render`'s cache key gains it; and a `debug_assert_eq!` in
+  `Painter::glyphs` that a submitted quad's generation is the live one.
+  Headless test
+  `clearing_the_atlas_re_renders_cached_text_instead_of_reusing_it`
+  (`iris/src/widget/text/mod.rs`): draw, `atlas.clear()`, `resize`, draw
+  again, and assert the atlas holds the same glyph count again -- it
+  stays at 0 without the fix, because the cache short-circuits before
+  `place`.
+
+## Build
+
+- [x] **Benchmarks**, not unit tests, run on demand (2026-09-05; a
+  `benches/` or a script under `iris/`, never in `cargo test`). The
+  scenario that matters most is a **message list** — chat apps and this
+  app's transcript alike — stressed with many messages and many images.
+  One case in particular: **resizing an input box** (typing enough text to
+  grow it) that pushes a long list of messages above it must stay very
+  fast and recalculate almost nothing — a move of everything above, not a
+  re-layout. That is exactly the O(1) move chain in LAYOUT.md; the
+  benchmark is what proves it. Done when the numbers are in this file with
+  the command, and the input-box case reports draws re-run, not just frame
+  time.
+
+  **Built as two rigs**, chosen per scenario by whether a real `wgpu`
+  device is needed (`UiRenderState`/`Widgets` touch no GPU or window, so
+  most of this runs as an ordinary binary — the same property
+  `layout_tests.rs` relies on):
+
+  - `iris/benches/message_list.rs` — a plain `Instant`-timed binary
+    (`[[bench]] harness = false` in `iris/Cargo.toml`), not criterion: see
+    the file's own header for why (short version — every scenario here
+    reduces to a *count* `UiRenderState::take_counters` already produces,
+    which criterion's statistical machinery adds nothing to and which a
+    new dependency is not worth pulling in for). Covers (a) first-frame
+    cost of a message list of N wrapped-text rows (one in 20 also carrying
+    a small in-memory image) for N = 100/1,000/10,000; (b) per-frame cost
+    of scrolling that list, 200 ticks; (c) the input-box case — a
+    fixed-height field at the bottom of the screen growing by a line 40
+    times, with the message list above it filling the rest of the screen.
+    Run: `cd iris && cargo bench --bench message_list` (always release —
+    `cargo bench` builds the `bench` profile, which is optimized).
+  - `iris/examples/bench_images.rs` — needs a real device, so it runs
+    through `iris/run-headless.sh bench_images`, printing
+    `UiRenderNode::take_image_bind_group_creates()` (a new counter, added
+    in `core/src/render/texture.rs` and `core/src/render/mod.rs`,
+    mirroring `UiRenderState::take_counters`) each frame. Covers (d): 1,000
+    image rows, checked both cold (does bind-group creation reach zero
+    once loaded) and after appending one more image once settled (does
+    *that* stay cheap) — the second question is what actually matters for
+    a live transcript and is what turned up the two Fix items above.
+  - `iris/run-bench.sh [list|images]` runs either or both and is what to
+    run before/after touching `Scroll`, `Span`, `Sized`, the move-offset
+    chain, or `GpuTextures`.
+
+  **Numbers (2026-09-05, release, `cargo bench`/`run-headless.sh`, this
+  VM: AMD Ryzen 7 3800X, 8 cores, rustc 1.98.0 nightly-2026-09-03):**
+
+      cd iris && cargo bench --bench message_list
+      (a) first frame, N=100:    30.30ms  draws=227   rewrites=15   moves=0
+      (a) first frame, N=1000:  186.04ms  draws=2252  rewrites=150  moves=0
+      (a) first frame, N=10000:1770.36ms  draws=22502 rewrites=1500 moves=0
+      (b) scroll, N=100/1000/10000, 200 ticks each:
+          draws=200 rewrites=0 moves=200 (identical at every N)
+          per-tick average: 0.0002ms (identical at every N)
+      (c) input grows 40 lines, N=100/1000/10000 rows above it:
+          draws=320 rewrites=40 moves=160 (identical at every N)
+          per-line average: 0.0012-0.0013ms (identical at every N)
+
+      cd iris && ./run-bench.sh images        (2026-09-05, before the fix)
+      frame=1 bind_group_creates=1000   (cold load)
+      frame=2 bind_group_creates=1000   (see Fix item above)
+      frame=3 bind_group_creates=0
+      frame=4 bind_group_creates=0
+      (append one image here)
+      frame=5 bind_group_creates=1001   (see Fix item above)
+      frame=6 bind_group_creates=0
+
+      cd iris && ./run-bench.sh images        (2026-09-05, after the fix)
+      frame=1 bind_group_creates=1000   (cold load, unchanged -- genuine work)
+      frame=2 bind_group_creates=0
+      frame=3 bind_group_creates=0
+      frame=4 bind_group_creates=0
+      (append one image here)
+      frame=5 bind_group_creates=1      (one image's own create_image, O(1))
+      frame=6 bind_group_creates=0
+
+  **Reading it**: (a) is real, necessary work — shaping and laying out N
+  never-before-seen text rows — and scales with N as it must, ~10x cost
+  per 10x N. (b) and (c) are the pass conditions that matter: both are
+  **exactly flat across N = 100 to 10,000**, confirming LAYOUT.md's O(1)
+  move chain holds for both scrolling and for a growing input box pushing
+  the message list — draws/moves per tick or per line do not grow with
+  list size, and the per-operation cost (a fraction of a microsecond) is
+  nowhere near a frame budget. (d)'s cold-load and steady-state halves
+  behave as designed; its *append* half did not, until the fix above moved
+  masks/move_offsets out of the per-image bind group — now flat at O(1)
+  the same way (b) and (c) are.
+
+- **I5's transcript screen (`iris/transcript-ui/`, 2026-09-05) — what it
+  left, each recorded at the point in the code it would go rather than
+  silently dropped. See RUST.md's I5 box for the full account of what
+  *was* built (the screen, `SpanStyle`, cross-row selection, the growing
+  composer).**
+  - [x] **Android integration for this screen — done, 2026-09-05.**
+    `iris-android-app`'s `transcript-screen` Cargo feature
+    (`transcript_client.rs`) runs this screen against a real `ai-server`
+    through `client-core`, confirmed on-device: real scrolling, real
+    touch-drag panning, tap-by-name on the composer. Two real bugs found
+    and fixed along the way (a missing `INTERNET` permission; a
+    background-thread redraw request that crashed via a `Looper`
+    requirement, fixed by routing through `View::post_delayed` — see
+    `IRIS.md`'s `Tasks::redraw_handle` entry). See RUST.md's I5 box,
+    "The Android integration, done 2026-09-05" for the full account.
+  - [x] **A render-time number for iris, comparable to Compose's
+    `transcript-bench.sh` report — instrumentation done and a real number
+    obtained, 2026-09-05 (later the same day); the clean comparable loop
+    is not.** `iris_core::FrameReport` (`iris/core/src/render/
+    frame_report.rs`, `IRIS.md`'s new entry) times every frame from
+    `render()`'s redraw start to after `queue.submit`+`present()`, exposed
+    as two named on-screen controls ("Frame report", "Reset frame
+    report"). Driven against a real on-device touch-drag it read
+    `frames=34 janky%=61.76 p50=26.5ms p90=48.0ms p99=98.1ms
+    worst=98.1ms` — real, not inferred, but accumulated across several
+    gestures rather than one clean 24-swipe loop, because of the new
+    finding below. See RUST.md's I5 box, "Update, 2026-09-05, later the
+    same day" for the full account.
+  - [ ] **New, 2026-09-05: intermittent touch delivery to iris's
+    `SurfaceView` under this checkout's `EMU_GPU=software` emulator.**
+    The same swipe coordinates, confirmed (by scanning a screenshot
+    column for the first non-black pixel) to sit over real row text,
+    sometimes produced 30+ real frames and a screenshot diff and
+    sometimes produced zero of either, across otherwise-identical
+    `ui-trace` invocations. Not the already-understood "already at that
+    scroll edge" case (reproduced with content confirmed taller than the
+    viewport, in both directions). Leading candidate, not yet confirmed:
+    this checkout's emulator was independently observed at ~78% of one
+    CPU core, continuously, while idle on-screen — SwiftShader's software
+    rasterisation is CPU-bound by design, and a synthetic touch competing
+    with that load for delivery is plausible but unmeasured *during* a
+    failing gesture (the standing rule against diagnosing from
+    after-the-fact measurements applies here). Needs a sampler (load,
+    `dumpsys input`, a `-i 0` `ui-trace` capture) running while a failing
+    gesture is driven, and ideally a comparison under `-gpu host` (real
+    Vulkan) to see whether it is specific to software rendering. This is
+    what blocks the clean, comparable 24-swipe loop above.
+  - [x] **Long-press-then-drag-to-select — confirmed on-device, 2026-09-05
+    (later the same day).** `ui-trace` gained a `holddrag X1 Y1 X2 Y2
+    HOLD_MS MOVE_MS` action (`emulator-tools`, additive, extends the same
+    `MotionEvent`/`injectInputEvent` mechanism `swipe` already used):
+    press, hold past `LONG_PRESS`, move, release, as one continuous touch.
+    Driven against a real row (`holddrag 300 1850 300 2050 600 300`) it
+    produced `iris selection: begin at row ...` then a sequence of
+    `iris selection: extend to row ...` log lines
+    (`transcript-ui/src/selection.rs`, a new small `log` dependency since
+    selection has no accessibility label of its own yet — see the next
+    item), and a screenshot taken right after shows the expected
+    highlighted selection spanning multiple rows. `DragArbiter`'s own
+    unit tests already covered this sequence against a synthetic clock;
+    this is the first time it has been driven by a real device touch.
+  - [x] **Touch-drag panning over a row's own rendered text — done,
+    2026-09-05.** `row.rs` used to register `CursorSense::click_or_drag()`
+    on each row's `TextEdit` for cross-row selection; `TextEdit::draw`'s
+    `painter.child_layer()` (`iris/src/widget/text/edit.rs:87`) meant that
+    registration won `core/src/sense.rs::run_sensors`'s per-layer
+    arbitration on every frame it was pressed, not just the frame the
+    press started, so a list pan gesture registered on `List` itself never
+    got a turn while a row was under the finger. Fixed with
+    `iris::sense::DragArbiter` (recorded in `IRIS.md`), one small state
+    machine per list deciding pan vs. select the way Android does (a
+    vertical drag pans immediately; a stationary press held `LONG_PRESS`
+    (500ms) starts a selection which further drag extends; a horizontal
+    drag while something is already selected extends immediately) —
+    `transcript-ui/src/selection.rs`'s `Selection::drag` is the one place
+    every row's drag now routes through. 8 new unit tests
+    (`iris/src/sense.rs`'s `drag_arbiter_tests`); `cargo fmt/clippy/test
+    --workspace` and `cargo ndk` (both `iris` and `transcript-ui`) all
+    clean; `run-headless.sh` screenshot byte-identical to before the
+    change (38578 bytes). See RUST.md's I5 box, "Gap closed, 2026-09-05".
+  - [x] **Intermittent touch-scroll dropout — root-caused and fixed,
+    2026-09-05.** Not the coalesced-`ACTION_MOVE` hypothesis the earlier
+    pass suspected (ruled out): a gesture's `ACTION_DOWN` can land on a
+    row's own padding/gap or its header, which no `CursorSense` covers,
+    so `DragArbiter` never gets `press_start` and sits in `Idle`
+    (answers `Undecided` forever) for that whole gesture. Fixed via a new
+    `DragArbiter::is_idle()` that `Selection::drag`
+    (`transcript-ui/src/selection.rs`) checks to recover a missed press
+    on the next `Pressing` frame. Four new unit tests. See RUST.md's I5
+    box, "Touch-scroll dropout root-caused, 2026-09-05", for the trace and
+    what a peer session sharing this checkout's emulator mid-pass
+    prevented from being re-verified end-to-end (the aggregate
+    `iris-scroll.sh` three-run confirmation and a re-taken FrameReport
+    row) — a future pass should finish that once the emulator is free.
+  - [ ] **Row-level accessibility names.** The composer carries
+    `.label("Message")`; transcript rows do not carry a `.label()` of
+    their own yet, so `Widgets::named()` (I4) does not include them —
+    `row.rs`'s `build_text_row` is where one would go, keyed to something
+    stable per row (its sender + a short excerpt, matching what a screen
+    reader announcing a chat message would say).
+  - [x] **A tappable link** — done 2026-09-06 (P1a). `TextEditCtx::
+    byte_at(pos, size)` answers which byte a tap landed on without
+    handing out the parley layout, `GestureOutcome::Tapped` says the
+    press committed to neither a pan nor a selection, and
+    `iris::platform::OpenUrl` is the capability each backend implements
+    (`xdg-open`/`open`/`start`; an `ACTION_VIEW` intent on Android,
+    deferred to `after_input` the way `pending_show_keyboard` is).
+  - [ ] **A background chip behind inline code.** Still needs per-range
+    glyph *geometry* — a run's boxes, not one offset — which
+    `TextEditCtx` does not expose outside `iris::widget::text`
+    (`edit.rs`'s `layout()` helper is private). The same primitive
+    `TextEdit::draw`'s own selection highlight uses internally,
+    `iris/src/widget/text/edit.rs:99`. `byte_at` above deliberately did
+    not open that up: a tap needs one offset and a chip needs the run.
+  - [ ] **`Selection`'s anchor-row shortcut.** The row a drag started in
+    is selected in full (`select_all`) the moment the drag leaves it,
+    rather than "from the click point to whichever edge points away from
+    the drag" — needs the same private `layout()` access as the item
+    above. `selection.rs`'s module doc has the exact reasoning.
+  - [x] **Syntax highlighting inside a fenced code block** — done
+    2026-09-06 (P1a). `client_core::highlight::spans_of` by language,
+    converted from its char indices to `SpanStyle`'s byte offsets, in
+    the same Catppuccin palette `Theme.kt` uses. A language the scanner
+    has no rules for stays plain rather than being coloured by the
+    nearest one's.
+
+- [ ] **Masks defined relative to each other.** Wanted: mask A multiplies
+  by something *and also* applies mask B — a mask can reference a parent
+  mask, the way the move chain references a parent offset. Today masks
+  are independent regions. Design it beside the move chain (same shape:
+  a parent index and a bounded walk in the shader); do it when a real
+  widget needs it, not before.
+- [ ] **Positions as a single float per scroll.** Iris raised, and half
+  rejected, letting a scroll update one float rather than positions:
+  input handling cares about most elements in a list, so absolute
+  positions must be computed on the CPU anyway. LAYOUT.md's design
+  already lands here (GPU walks the chain, CPU resolves on demand for
+  hit tests). Keep the CPU resolution lazy and per query; do not
+  materialise every row's absolute position per frame.
+- [ ] **Animations, last.** Cosmetic, so after everything above. Must be
+  **modular — a piece of the library rather than a core part forced into
+  everything, the same way input is**. Whatever the mechanism, a widget
+  that does not animate must pay nothing and import nothing for it.
+
+## Found by P1a (2026-09-06)
+
+- [x] **`Rect` claimed to be size-independent, and it is not.** A `Rect`
+  fills whatever region it is handed, so `draw_inner`'s size-independent
+  fast path -- which rewrites primitives with
+  `r.outside(&from).within(&region)` rather than redrawing -- could not
+  reproduce its `draw`, and a `.background(rect(..))` kept the size of
+  the *provisional* full-region pass `Span` does in phase 1. One fenced
+  code block's panel covered every block below it and every row below
+  that. Fixed in `iris/src/widget/rect.rs`; the reason is written at the
+  definition. Suspect the same cause for anything else tinted with a
+  background rect.
+- [x] **A wrapped transcript row tripped `reposition`'s debug assert.**
+  Settled 2026-09-06 by giving the move slot one owner instead of two.
+  `mov` accumulates a delta on it, `reposition` overwrote it, and both
+  legitimately land on one widget in one frame: `List::place`'s
+  Bottom-known branch offers a row a same-size box that has *moved*
+  (`mov`), then corrects the placement inside it when the row's cached
+  height no longer matches what the row reports (`reposition`). The
+  slot now always means `move_applied + repositioned`
+  (`ActiveData::repositioned`, `iris/core/src/ui/render_state.rs`), so
+  `reposition` adds the move rather than dropping it -- the assert is
+  gone and the arithmetic is right. Test:
+  `a_widget_moved_by_its_parent_and_then_placed_inside_it_lands_at_the_placement`
+  in `layout_tests.rs`, which lands the child at the *offered* position
+  (-100px) instead of the placement (100px) without the fix, and a
+  `debug_assert_eq!` in `reposition` that nothing but those two ever
+  writes the slot. Verified with the `.wrap(true)` repro (draws, no
+  panic) and an emulator bench run with assertions live.
+- [ ] **Desktop colours are washed out: the winit surface is sRGB and
+  the shader writes the palette's bytes as linear.** Mocha Crust
+  (17,17,27) is drawn as (73,73,91), measured off
+  `run-headless.sh --shot`. Android is correct, so this is the surface
+  format rather than the palette -- but it makes the desktop build
+  useless as a colour reference, which is exactly what P1a needed it for
+  when the emulator could not draw glyphs.
+- [x] **Every glyph was a solid box on the GLES backend -- iris's bug,
+  not the emulator's.** Fixed 2026-09-06. The atlas is one
+  `texture_2d_array` and `GpuTextures::new` created it with **one
+  layer**; wgpu-hal picks the GL target from the descriptor
+  (`(false, 1) => TEXTURE_2D`), so under GLES that array was a
+  `GL_TEXTURE_2D` bound to the shader's `sampler2DArray`, the unit was
+  incomplete, every `textureSample` returned (0,0,0,1), and
+  `draw_glyph`'s `color.a *= texel.a` filled the quad. `MIN_ARRAY_LAYERS
+  = 2` in `iris/core/src/render/texture.rs`, with a `debug_assert!` at
+  `create_array_texture`. Vulkan (the phone, the desktop's default
+  backend) was never affected. Reproduce the class in seconds without an
+  emulator: `iris`'s `force-gles` feature now switches the **desktop**
+  backend too -- `./run-headless.sh transcript --shot /tmp/x.png -- -p
+  transcript-ui --features iris/force-gles`.
+
+- [ ] **The bench report pane draws over the transcript rows instead of
+  replacing them.** Visible on the emulator for the first time now that
+  glyphs render there (`/tmp/emu-final.png`, 2026-09-06): after a bench
+  run the report's lines and the transcript's occupy the same rows in the
+  top third of the screen, both legible, neither on top. Pre-existing --
+  the same overlap is in a screenshot taken before the move-slot fix -- so
+  it is its own item, most likely the report pane not masking or not
+  claiming its region.
+
+## Found by P1b (2026-09-06), all with a headless repro
+
+Each was found by looking at `iris/run-headless.sh transcript -- -p
+transcript-ui` rather than at a diff, and each is worked around in
+`transcript-ui/src/tool.rs` rather than fixed here. docs/RUST.md's P1b box
+has the fuller account.
+
+- [ ] **A `Span` of `Pad`ded children inside another `Span` places those
+  children a slot out of step.** Each child drew its content one sibling's
+  height below its own box. Repro: `IRIS_TOOLS_EXPANDED=1
+  iris/run-headless.sh transcript --shot /tmp/x.png -- -p transcript-ui`
+  with `tool.rs`'s group built as `Span(DOWN)[header, Pad(Span(DOWN)
+  [cards]), bar]` instead of the single `Span` it uses now. Bisected:
+  removing the inner `Span` fixes it, and so does removing the children's
+  own `Pad`; the background `Stack`, the `Sized` wrappers and the
+  `WidgetPtr` per child make no difference. **Not** the `mov`-vs-
+  `reposition` fault f5b8893 fixed -- it survives that commit. The
+  workaround costs the group the 4dp inset its Compose counterpart holds
+  its cards off the edge by, so this is worth fixing.
+- [ ] **`scrollable_on(Axis::X)` on a non-editable `Text` draws nothing.**
+  The panel is drawn and the text inside it is not. A markdown fence does
+  the same to a `TextEdit` and is fine, so it is the widget kind rather
+  than the chain. `tool.rs`'s `raw_block` is `masked()` only until this is
+  fixed, which means a long command is clipped rather than pannable.
+- [ ] **No overflow ellipsis.** `TextAttrs` can wrap or not wrap; there is
+  no "one line, ellipsised" the way `maxLines = 1` + `TextOverflow.
+  Ellipsis` gives Compose. A tool card's summary is clipped instead, so
+  nothing on screen says it was cut. Whichever end is cut has to be a
+  choice when this lands: a path is identified by its tail, a command by
+  its head.
+- [ ] **A drawn chevron.** `Chevron.kt` draws its own strokes precisely
+  because a chevron from a font is a glyph a system font may not have --
+  and the bundled `NotoSans-Regular.ttf` indeed has no U+25B8/25BE/25B4,
+  while `NotoSansMono-Regular.ttf` does. `tool.rs` sets the mark in the
+  monospace face as a result. A real fix needs a line/path primitive;
+  iris has rects, text and textures only.
+- [ ] **A tool card's text is not selectable.** `Selection` is keyed
+  `(RowKey, block index)` and a card has no markdown blocks, so nothing in
+  a card registers. Compose's `SelectionContainer` covers tool output,
+  which is the text people most want to copy. Needs a key for "the nth
+  text of this row" that a card can mint without colliding with a
+  message's blocks.
+
+## Build (for the port)
+
+Widgets `RUST.md`'s "The port, in order (decided 2026-09-05)" needs and
+iris does not have yet, one entry per gap, named against the P-step that
+first needs it. Move an entry up to "Fix" or tick it in place once built;
+do not duplicate it there.
+
+- [ ] **A history-paging cushion measured in on-screen viewports, not a
+  row count.** (**P1**.) `iris::widget::List` has no equivalent of the
+  Compose app's `HISTORY_SCREENS` — AGENTS.md's "Things that have
+  bitten" is explicit that a fixed row count under-fills a screen on a
+  tool-heavy transcript and over-fills one on a text-heavy one, so
+  whatever loads the next page has to ask the list how many viewports
+  are actually on screen, not assume a constant.
+- [ ] **A scaled thumbnail/image widget for an in-transcript image.**
+  (**P1**.) `SessionImage.kt`'s bitmap decode-and-downscale has no iris
+  counterpart; iris's own image widget (used by `bench_images.rs`) draws
+  a loaded texture but does nothing about sourcing or scaling one from a
+  server-produced attachment.
+- [ ] **A modal/dialog primitive.** (**P1**, reused by **P3** and
+  **P5**.) Needed for the session settings dialog, `UsageDialog`'s
+  equivalent, and the delete-with-`deleteForeign` confirmation with its
+  toggle switch. Build once, wherever it is first needed, rather than
+  once per screen that wants one.
+- [ ] **A horizontal gauge/bar widget.** (**P1**.) For
+  `SessionUsageBar`'s equivalent — a bounded fill reflecting a fraction,
+  nothing fancier.
+- [ ] **A `BusyItem` equivalent: a dimmed row carrying an operation
+  label that does not block its list's own scroll/drag.** (**P3**.) The
+  Compose version tried an overlay first and it swallowed the drag along
+  with the tap (AGENTS.md's "Shared appearance") — worth not repeating
+  that attempt in iris before building the row-level version directly.
+- [ ] **A toggle switch.** (**P3**.) For the delete dialog's
+  `deleteForeign` control; iris has no switch/checkbox widget yet as far
+  as this pass found.
+
+## Reconsider
+
+- [ ] **`WidgetView`.** Iris is unsure of it: what she wants is an easy way
+  to compose a widget from others (a button is the main case). With
+  sizing folded into `draw`, composing may be easy enough that `View` is
+  redundant. Decide after the layout change lands, by writing a button
+  both ways and keeping the one that is shorter to explain; delete the
+  other rather than keeping two ways.
+
+## Build (asked for by Iris, 2026-09-06): a density-independent length unit
+
+- [x] **A third length kind beside relative and pixels, so display scales
+  "just work".** Done 2026-09-06 — `Len::dp`/`len_fns::dp`, resolved
+  against `UiRenderState`/`Painter::density()` at `apply_rest` time; text
+  additionally rasterises at the resolved (physical) size instead of
+  scaling a low-resolution bitmap afterward, which was making text blurry.
+  `Span::gap`/`Padding` moved from `f32` to `Len` so they take `dp(...)`
+  too; transcript-ui's row/composer padding and one example migrated.
+  `em` was not added — nothing in this pass needed a text-relative unit,
+  and `dp`'s own doc says why it and physical pixels are kept as separate
+  fields rather than one the caller pre-multiplies. Not yet verified on
+  Iris's own phone at two densities (this pass had no device) — see
+  docs/RUST.md's P0 box and docs/IRIS.md's 2026-09-06 entry for what to
+  check. Iris's words: "another length type similar to absolute &
+  relative, so instead there would be relative, pixels, and another unit
+  like em or whatever is standard. That way different display scales
+  should just work." Today a length is either a fraction of the parent
+  (`rest`/relative) or physical pixels, and the phone drew 16 px text at
+  roughly a third of its intended size until the P0 fixes applied the
+  display's scale factor globally. That global scale is a stopgap for the
+  benchmark; the real shape is a unit resolved against the display's
+  density at layout time — Android's `dp` / CSS's reference pixel is the
+  standard (1 unit = 1/160 in), with `em` as the text-relative option —
+  so a widget author writes `16.dp()` once and never sees the scale.
+  Done when: `Length` (or whatever the enum is called) has the third
+  variant; every place that resolves a length takes the density; the
+  examples and `transcript-ui` use the new unit for text sizes, padding
+  and control sizes; the emulator at two densities and the phone draw the
+  same layout at the same physical size. After the bench setup is
+  finished, before P1 draws any new screen.
+
+## From the phone, bench v2 (2026-09-06): streaming re-lays out the whole message
+
+- [x] **Streaming a delta into a long message costs a full text layout of
+  that message.** **Done 2026-09-06** -- a row is a column of one
+  `TextEdit` per markdown block (`client_core::markdown_blocks`,
+  `row::RowBlocks::apply_delta`), so a delta re-shapes the last block and
+  keeps every earlier block's layout. A block is the selection unit now
+  (`Selection`'s `SelKey`); selection across blocks and rows still works,
+  checked on the emulator with a real long-press drag. Pass condition met
+  in `a_delta_into_a_long_reply_redraws_the_same_widgets_as_a_short_one`:
+  a delta into a 100-paragraph reply redraws the same widget count as one
+  into a one-paragraph reply (30 either way). Emulator stream phase, same
+  AVD before and after: **p50 61.5 -> 54.5ms, p90 211.7 -> 113.1ms, p99
+  342.6 -> 137.4ms, worst 403.6 -> 143.0ms**, 202 -> 293 frames in the same
+  21 seconds. docs/RUST.md's Task B box has the detail and the two dead
+  ends. **The phone is the measurement that decides it** -- these are
+  emulator numbers and only the ratio transfers.
+
+  The original entry, for the record: Iris's phone report (`docs/bench/iris-phone-v2-2026-09-06.md`):
+  the stream phase is the one place iris is behind Compose (p50 18.2 ms vs
+  13.4 ms; p99 level at ~43 ms). `TranscriptScreen::apply` replaces only
+  the last row, but that row is the growing message, and replacing it
+  re-renders its markdown and re-shapes the entire paragraph run through
+  parley on every event. Compose pays a reparse (8.6 ms mean) for the
+  same event. What "done" looks like: a streamed delta re-lays out only
+  the block it lands in (the last paragraph or code block), with earlier
+  blocks' layouts kept -- which needs a row to be a column of per-block
+  `Text`s rather than one `TextEdit` for the whole message, or parley's
+  layout to be split at block boundaries; measured by the stream phase's
+  p50 dropping below Compose's on the phone. Do this after the four bench
+  v2 defects (stale primitives, finger fling, decay curve, IME show) are
+  closed, since they are what make the run unrepresentative today.
+
+## From the phone, 2026-09-07 (build from ed04d4c)
+
+- [ ] **"Some transcript blocks will be hidden until I uncover enough of
+  them."** Two screenshots of the bench app's transcript at the top
+  edge, both wrong in opposite directions: in one, rows scrolled above
+  the viewport are still drawn and bleed *through* the header bar
+  (`version = "0.1.0"` and a paragraph visible behind "Run benchmark /
+  Copy report / Diagnostics"), so the list's mask is not clipping at
+  the header's bottom edge; in the other, scrolled a little further,
+  the row that straddles the top edge is not drawn at all -- black from
+  the header down to "You", where the previous shot showed a paragraph
+  -- so a row is culled as soon as its *top* leaves the viewport rather
+  than when its *bottom* does. Suspects: the list's visible-range test
+  (`iris/src/widget/list.rs`) comparing a row's top against the
+  viewport top; the mask region for the transcript set from the
+  window rather than from the area under the header; and the two-phase
+  provisional/real draw noted in `03c6be8`'s header-duplicate
+  investigation, which was never root-caused and has the same shape.
+  Reproduce at layer 1 of the test rig: a headless screen with a row
+  straddling the top edge must place that row, and a primitive above
+  the header's bottom must be masked. Fix both with one rule: a row is
+  drawn if any part of it intersects the viewport, and the viewport is
+  the list's own region.

docs/LAYOUT.md

@@ -861,6 +861,58 @@ unspecified rather than getting them wrong:
   conditions, so the remaining slack was accepted rather than chased
   further.
 
+## Density: `Len::dp`, resolved at `apply_rest` time (2026-09-06)
+
+Iris asked for a third length kind beside `abs` (physical pixels) and
+`rel`/`rest` (a fraction of the parent) — IRIS_TODO.md's "density-
+independent length unit" — after the P0 phone pass found 16px text
+drawing at roughly a third size on a real phone. The fix that shipped
+first (RUST.md's P0 box) was a global stopgap: divide the whole window
+into a "logical" coordinate space (physical ÷ `content_scale`) and let
+the shader's NDC mapping stretch it back up onto the real framebuffer.
+That fixed the *size* but not the *sharpness* — a glyph rasterised at the
+small, pre-stretch size and then stretched onto more physical pixels than
+it has texels for is blurry, which is exactly what Iris's next report
+said.
+
+**The fix**: `Len` gained a `dp` field, resolved against a `density: f32`
+(physical pixels per dp) at the one place a `Len` becomes a `UiScalar`
+(`Len::apply_rest`) — `abs + dp * density`. `density` lives on
+`UiRenderState` (`set_density`/`density()`) and `Painter` (`density()`),
+set once from `DisplayMetrics.density` in `android::view::new_peer`; the
+desktop backend has no per-monitor density wired up yet and stays at
+`1.0`. Every layout call site that used to call `.apply_rest()`/
+`.to_uivec2()` now passes `painter.density()` (nine call sites — `Span`,
+`Sized`, `MaxSize`, `Aligned`, `Scroll`, `List::place`, and
+`UiRenderState::reposition` itself). This also meant the Android
+boundary's global logical-space stopgap could come out entirely: window
+size, touch coordinates and insets are physical pixels again, matching
+`AndroidRenderer`'s own swapchain resolution, with `dp` doing the
+per-length work the global divide used to do for everything at once.
+
+**Text is the case that needed more than the `Len` plumbing.** A widget's
+`font_size`/`line_height` are plain `f32`, not routed through `Len` at
+all (there is no sensible `rel`/`rest` for a font size). `TextBuffer::
+shape` now takes `density` directly and multiplies `font_size`/
+`line_height` (and any span override) by it before handing them to
+parley — so the size that reaches both the line-breaker and the
+rasteriser (`TextData::place`, which reads back whatever `shape` set) is
+the display's *physical* size, and the glyph atlas holds a bitmap at the
+resolution it is actually shown at. `GlyphKey.size` already keys on the
+resolved size, so a cache entry is naturally per-physical-size with no
+further change. The one caller with no `Painter` to read density from
+(`TextEditCtx::layout`, cursor movement and hit-testing) reads a second
+copy kept directly on `TextData` (`TextData::density`) instead — an
+accepted duplication rather than threading a `Painter` into every input
+handler for one field, the same tradeoff `AndroidRenderer::content_scale`
+already makes for the Diagnostics page.
+
+**What did not change**: `rel`/`rest` are unaffected (already
+resolution-independent, a fraction of the parent). `Span::gap` and
+`Padding`'s four sides moved from bare `f32` to `Len` so `dp(...)` works
+on them the same as any other size; a bare number is still `abs`,
+physical pixels, unchanged.
+
 ## For IRIS.md
 
 When this lands, copy this entry into `IRIS.md` (newest first):
@@ -895,3 +947,95 @@ When this lands, copy this entry into `IRIS.md` (newest first):
 > `SizeCtx` and `Cache` are gone with it — see `LAYOUT.md` for the full
 > design, the move-offset mechanism this shipped alongside, and the file
 > list.
+
+## Masks with a shape (decided 2026-09-07, not yet built)
+
+Iris, on the code block's scrolling: "the code block scrolling currently
+masks in an inner rectangle. Ideally masks should have a shape
+associated with them, rounded rectangle being one of them, and/or
+another widget you can select, so that the mask becomes the parent
+container with rounded edges. Make sure alpha works properly with it,
+eg. on the corners where alpha should be decreased / multiplied."
+
+**What exists.** `Mask` in `shader.wgsl`/`data.rs` is two `UiSpan`s and
+a `move_idx`; `fs_main` resolves it and does `color *= 0.0` outside the
+rectangle -- a hard cut on a pixel boundary. `Masked` (`widget/mask.rs`)
+sets the painter's mask to its own region. Separately, `draw_rounded_rect`
+already produces an anti-aliased rounded edge from
+`distance_from_rect(pos, center, corner, radius)` with a half-pixel
+`smoothstep`, and the border variant multiplies a second coverage in.
+
+**Design** (revised the same day on Iris's two corrections: hit-testing
+applies the shape too, and a mask should reference a primitive rather
+than carry a copy of its shape).
+
+1. **A mask is a reference to a primitive already drawn, plus how to
+   use it.** `Mask { kind, idx, flags, parent }`: the primitive's
+   binding (`RECT`, `TEXTURE`, `GLYPH`) and slot, flags (today one:
+   *alpha only* -- take the primitive's coverage and ignore its colour,
+   which is the default and the only mode until a need for another
+   appears), and the enclosing mask's slot for nesting. The fragment
+   stage evaluates the referenced primitive *at the masked pixel* --
+   for a `Rect`, the same `draw_rounded_rect` coverage from the same
+   SDF; for a texture or glyph, the sampled alpha -- and does
+   `color.a *= coverage`. Nothing about the shape is copied: a rounded
+   container's corner and its children's clipped corner are the same
+   primitive's arithmetic, and a texture mask (an alpha image as the
+   clip) works with no new shader path.
+   What this needs from the data layout: evaluating a primitive at an
+   arbitrary pixel means its placement (its spans and `move_idx`, today
+   vertex attributes) has to be readable from a storage buffer in the
+   fragment stage. If it is not already there, put it there once, for
+   every primitive, rather than keeping a second copy for masks -- the
+   vertex stage can read the same buffer. Textures: the shader binds one
+   image at a time (see `masks_layout`'s comment on why an image's own
+   bind group must not name the masks buffer), so a texture mask is
+   limited to what the fragment can sample without a bind-group switch:
+   the atlas, and the primitive's own bound image when the masked
+   primitive is drawn in the same image's batch. Say so at the flag.
+2. **Nested masks chain and multiply, like moves.** `parent` walks up
+   the chain, bounded like `resolve_move` (`MOVE_CHAIN_LIMIT`'s sibling;
+   debug-assert on overflow and print the chain); coverages multiply,
+   so a pixel inside two feathered corners is dimmed by both, which is
+   what a compositor does and what "alpha should be multiplied" asks.
+3. **`.masked()` points the mask at the current widget's own
+   primitives.** `Masked` stops describing a region: it records which
+   primitive(s) the wrapping widget drew this frame (the painter knows
+   -- it just allocated the slots) and sets the mask to reference them.
+   So a rounded `Rect` widget's `.masked()` clips its children to
+   itself by pointing at the rect it already draws; an image widget's
+   `.masked()` clips to its alpha. No radius or shape argument exists to
+   fall out of sync. When a widget draws more than one primitive (a
+   bordered rect is one primitive; a card with a stripe is two), the
+   mask references the *first* and the doc says so; a widget that wants
+   another names it.
+4. **Hit-testing applies the shape.** A press is inside a masked
+   subtree only if the mask's coverage at that point is above one half.
+   For a `Rect` that is the same rounded-rect SDF evaluated on the CPU
+   -- one function in the shared crate, with the WGSL a transliteration
+   of it and a test that compares the two at a grid of points
+   (`headless` renders to a buffer and reads back, or the Rust version
+   is checked against the values the shader produced once and recorded).
+   For a texture, the CPU needs the alpha: keep the alpha channel of an
+   image used as a mask readable on the CPU (it was uploaded from CPU
+   memory; keeping the alpha plane is a quarter of the image), and read
+   it at the point. A masked corner that cannot be tapped and a masked
+   corner that is not drawn are then the same corner.
+
+**Rejected.** A stencil buffer (a second pass per mask level and no
+anti-aliasing); the scissor rectangle (rectangles only, no alpha);
+rendering a masked subtree to an offscreen texture and compositing
+(a texture allocation per mask, every frame it scrolls, on the phone).
+
+**Pass conditions.** A headless test draws a rounded container with a
+masked child that overhangs all four sides and asserts the child's
+coverage at a corner pixel equals the container's own coverage there
+(same primitive evaluated, so exactly equal, not approximately); a
+nested-mask test asserts the product at a pixel inside both feathers; a
+texture-mask test clips a rect to an alpha image and asserts a
+transparent texel masks fully; a hit-test asserts a press in a
+container's clipped corner misses and one just inside the curve hits,
+and that the CPU SDF and the shader agree at a grid of points; a
+`run-headless.sh --phone` screenshot of a scrolled code block shows
+rounded corners with no square pixels poking out at the top and bottom
+of the scrolled content. Record the commands in RUST.md when it lands.

docs/PLAN.md

@@ -147,12 +147,37 @@ turn.
 
 Spawn: `claude -p --verbose --input-format stream-json --output-format
 stream-json --permission-mode <mode>` in the chosen working directory, plus
-`--model`. Wire-format notes are pinned against CLI 2.1.237 in
-`session/claude.rs`'s module doc: permissions need the hidden
-`--permission-prompt-tool stdio` flag, AskUserQuestion answers ride
-`updatedInput.answers` keyed by question text, and `set_model`/`interrupt`
+`--model` and, where one has been chosen, `--effort`. Wire-format notes are
+pinned against CLI 2.1.237 in `session/claude.rs`'s module doc: permissions
+need the hidden `--permission-prompt-tool stdio` flag, AskUserQuestion answers
+ride `updatedInput.answers` keyed by question text, and `set_model`/`interrupt`
 are control requests.
 
+**The thinking level is settled at launch** (added 2026-09-04, because it is
+the largest saving available on a long session: output is about an eighth of
+what a session costs and thinking is the bulk of output, against the ~1.5% that
+is prose). The CLI's only two setting control requests are `set_model` and
+`set_permission_mode` -- checked against the 2.1.258 binary -- so there is no
+way to ask a running process to think differently. `set_session_effort` is
+therefore shaped like `set_session_cwd` rather than like `set_session_model`:
+it records the level and **stops the process**, and the next message or Start
+launches one that has it. It lives in the session settings dialog beside the
+working directory for that reason, not on the session bar beside the model and
+the mode, which do take effect mid-turn. `None` is a level in its own right --
+the CLI's own default -- so the picker can return to it; a level this app named
+as the default instead would be this app choosing one.
+
+**What a new session starts at is `Config::default_effort`**, applied in
+`spawn_session` rather than filled in by the spawn screen, so it holds for an
+import and a bare API call as well. It is set by the spawn screen's own
+picker, whose label says so: one control, where new sessions are made, rather
+than a settings page for a single value. It is not on a provider, because
+providers are discovered and the next rediscovery would erase it, and not on
+the phone, because a second device would then spawn at a level nobody there
+chose. `GET`/`POST /defaults` carry it, as a struct rather than a bare value
+so the permission mode -- still hardcoded to `auto` on the spawn screen -- can
+move there without a second route.
+
 **`--resume` only ever runs when nothing else has that session open.** That
 is the rule behind the import refusal, the single `ClaudeDriver::launch`
 entry point, and the `Exited` correction below; two CLIs on one session file
@@ -169,10 +194,33 @@ deliberate and easy to undo by accident:
   when the process restarts. That leaves the Claude driver as the odd one
   out rather than this one — the CLI's memory is a cache in front of the same
   transcript. Resolve any inconsistency in this direction.
-- **A llama session on an ssh host is refused.** The model is reached over
-  HTTP and forwarding that port is not built, so refusing beats silently
-  talking to the wrong machine. A transport is "run this" plus "reach this
-  port", and only the first half exists.
+- **A llama session runs on whatever machine its setup names** (2026-09-04,
+  the last of phase 5). A transport is "run this" plus "reach this port", and
+  the second half is `Transport::reserve_port` — the port the server binds
+  *there* and the port that reaches it *here*, the same number locally —
+  carried by `Launch::reaching` onto the connection that already runs the
+  command. `llama-server` binds loopback on the far machine, so nothing is
+  served to its network. The far port is a guess from a range below the
+  ephemeral one, because no portable way to ask a machine for a free port
+  avoids racing the bind anyway; a collision is not silent, since the server
+  fails to bind and the readiness poll reports what its log said.
+- **The model file lives on the machine that serves it** (2026-09-04). Each
+  setup names its own models directory (`SshConfig::models_dir`, default
+  `~/.local/share/ai-app/models` expanded *there*), and a spawn resolves the
+  key on that machine — one round trip answering "at /abs/path" or "missing",
+  so a model that is not there is refused at the spawn rather than becoming a
+  server that never becomes ready. The spawn screen offers
+  `GET /setups/{id}/models`, that machine's list, rather than `GET /models`,
+  which is this backend's downloads. Downloading *to* another machine is
+  deliberately not built: a multi-gigabyte transfer with no progress
+  anywhere, and the file gets there however anything else on that machine
+  did.
+- **The readiness poll watches the process, not only the port.** A model that
+  will not load, a port already taken, a flag an older build does not know:
+  all exit within a second and none will ever answer `/health`, so waiting
+  out the 300s timeout turned the server's own account of the problem into
+  "gave up". The failure carries the tail of `llama-server.log`, which on a
+  remote session is the only copy anybody reading the phone can see.
 
 ### Models (2026-08-28)
 
@@ -203,6 +251,13 @@ deliberate and easy to undo by accident:
   A driver says what to run; something above it turns that into a process.
   Otherwise transport knowledge sits inside a translator whose job is a wire
   format, and every future driver has to remember to do the same.
+- **A forwarded launch gets a pty and every other one does not** (measured
+  2026-09-04). Killing the ssh client ends a CLI because it closes the stdin
+  that CLI is reading; `llama-server` never reads its stdin, so the same kill
+  left it running on the far machine with the model loaded — one orphan per
+  stopped session. With `-tt` the far side takes SIGHUP when the connection
+  goes. Its log then arrives through a line discipline, which nothing parses.
+  `-T` stays everywhere else, where a pty would rewrite the JSONL.
 - **`command -v` follows ssh's non-login PATH**, which is narrower than an
   interactive shell's, so a binary somewhere unusual is invisible to
   discovery. Point `command` at an absolute path.
@@ -499,6 +554,25 @@ rate-limited bucket). Poll at ≥180 s, only while a Claude session exists or
 the usage screen is open, and cache the last answer. It is undocumented, so
 `usage.rs` treats every field as optional and degrades rather than erroring.
 
+**Per provider, not per machine (2026-09-04).** A machine is not what is
+metered; the provider a session runs is. One machine offers echo, the Claude
+CLI and a local model side by side, and only the second spends anything — so
+pairing a session with a snapshot by machine alone drew the CLI's five-hour
+window under every echo session on it, a quota that session cannot spend. A
+session now names its meter (`usageProvider`, from
+`DriverKind::usage_provider`, which `usage::providers_for` reads too, so the
+two lists cannot disagree) and `GET /usage` is matched on machine *and*
+provider. `None` is a session that meters nothing, and the phone draws
+nothing at all for it — not a zero, and not "unknown".
+
+`DriverKind::Echo` names a meter of its own that exists only when a test has
+asked for one: `/usage` in an echo session sets an invented answer
+(`usage::Fixture`), and with none set there is no snapshot and no bar. That
+is what makes those screens' states reachable — a number near the top, a
+window between blocks with no reset time, a machine nobody logged into, one
+that could not be reached — without spending real quota to arrange them,
+which is why none of them had ever been looked at.
+
 **Per machine, not per backend (2026-08-29).** The credential store that
 matters is the one on the machine the session runs on, because that is the
 account being billed — and in the layout this aims at, `ai-server` is on the
@@ -522,6 +596,68 @@ always running. So absent means **not running**, and only a timestamp that
 arrives and cannot be parsed is unknown. `WindowEnd` in `ResetCountdown.kt`
 is the one rule both readers go through.
 
+### Auto-resume (2026-09-05)
+
+**A session may pick itself back up when the account's usage limit lifts.**
+Off unless somebody switched that session to it, because it spends quota the
+moment quota exists and does so with nobody looking — that is not a thing a
+default may decide. It sends one message, `continue` unless another was
+typed, and then it is done; there is no retry loop around the conversation
+itself.
+
+**Running out of quota is a state, not an error.** `Event::LimitReached`
+carries the dialect's reset time where it gave one, and recognising it
+belongs to the driver — the Claude CLI ends the turn with `is_error` and
+`Claude AI usage limit reached|1788546972`, and nothing above the driver
+matches on a string. The transcript draws it as a divider, like a clear or a
+compaction: what a reader scrolling back wants from it is why the
+conversation stops at that line.
+
+**The schedule is a plan to ask, never a plan to send.** Every reset time
+available here is untrustworthy in the direction that matters: the dialect's
+is written when the turn fails, and the endpoint's moves when the window
+does. So the wait ends in a question to `usage.rs`, and only `ok` with no
+window at 100% sends anything. A window still spent reschedules to *its own*
+reset time — which is what makes a limit that lifts later than promised wait
+longer, and one that lifts sooner resume sooner. A meter that cannot be
+asked at all is a longer wait too, never a send: "we could not find out"
+must not be able to produce the same action as "there is room".
+
+Bounded, because something has to be: a day after the limit was hit the wait
+stops and says so in the session's own transcript. A machine that can never
+be asked would otherwise be retried for ever with nothing on screen saying
+so.
+
+The schedule is persisted on the session (`resume: Some(ScheduledResume)`),
+not held in memory: a five-hour window routinely outlasts a backend restart,
+and a wait forgotten across one is a session that silently never comes back.
+`resume.rs` is the top layer — it holds the manager and the monitor and
+neither holds it — which is what lets the decision be a pure function of a
+snapshot and a clock. The pump reports limits downward on a broadcast, for
+the reason `Shared` exists: the pump runs underneath the manager.
+
+**Exercised with echo, never with a real account.** `/limit [minutes]` in an
+echo session reports the same event a real driver does, and `/usage` sets
+what the meter answers — deliberately two commands, because the two
+disagreeing is the state the whole design is about. The loop was driven end
+to end that way on 2026-09-05: the wait moved from the dialect's two minutes
+to the meter's seven when the meter changed its mind, and the message went
+out on the first check after the meter came back under the limit.
+
+### Subagents (2026-09-05)
+
+**A subagent is a second transcript owned by a session, in the same event
+model, with no process and no controls of its own.** Full design and wire
+shape in `SUBAGENTS.md`, kept separate because the app half is being built
+against it in parallel and it is the shared contract between the two. The
+one-paragraph reason: a session's Task-tool helpers already speak the common
+event model on the parent's own stdout (each line carrying
+`parent_tool_use_id`), so giving each one its own small transcript — same
+file format, same paging routes, same SSE stream, reused by addressing rather
+than by copying — costs a routing step in the translator and a registry
+(`session/subagent.rs`) rather than a second session type with a driver, a
+process and a config entry it does not need.
+
 ### HTTP surface
 
 **`routes.rs`'s module doc comment is the table.** REST for actions, one SSE
@@ -800,8 +936,9 @@ Noticed and deliberately not fixed, so they are not re-found from scratch.
 
 Phases 1–3 (the skeleton pipe, the full Claude driver, the usage screen) done
 2026-08-24. Phase 4 (llama.cpp: model browsing, downloads, and `llama-server`
-through its OpenAI-compatible endpoint) and phase 5 (ssh) done 2026-08-28.
-The file explorer and the transcript cache followed in September. What is
+through its OpenAI-compatible endpoint) and phase 5 (ssh) done 2026-08-28,
+except for the remote `llama-server` and its port forward, which landed
+2026-09-04. The file explorer and the transcript cache followed in September. What is
 left is real-phone/WireGuard bring-up, which is operational rather than code.
 
 Each phase ended runnable and verified against the real thing. The backend

docs/REVIEW-2026-09-06.md

@@ -0,0 +1,214 @@
+# Review: iris changes since 0e46293
+
+Scope: `git diff 0e46293..HEAD -- iris/ client-core/` (58 files, +5224/-226).
+Read-only review; no source changed. Ordered likely-bug, then invariant
+guards, then rules, then tests/docs.
+
+## Likely bugs
+
+1. **`iris/transcript-ui/src/lib.rs:152-160` (`RowDiff::Rebuild` arm of
+   `TranscriptScreen::apply`) never unregisters the rows it drops from
+   `Selection`, so a stale `WeakWidget<TextEdit>` outlives the widget it
+   points to and the next touch on *any* row panics.**
+   `Selection::rows: BTreeMap<RowKey, WeakWidget<TextEdit>>` documents its
+   own contract at `selection.rs:69-71`: "every addition here needs its
+   removal ... called when `List` evicts the row." The `ReplaceLast` arm
+   above it honours this (`lib.rs:143-145`, `self.selection.borrow_mut()
+   .unregister(old_key)` when the key changes). The `Rebuild` arm calls
+   `(self.list)(rsc).clear()` and rebuilds every row from `new_rows`, but
+   never touches `self.selection` — any key present in `old_rows` and
+   *absent* from `new_rows` (exactly what `group_tool_runs` regrouping two
+   separate tool-call rows into one produces — see `diff_tests::
+   a_tool_run_closing_and_joining_an_earlier_call_is_a_regroup_fallback`,
+   which tests the diff decision but not `apply` itself) is left in
+   `self.rows` pointing at a widget `List::clear()` just freed.
+   `TextEditable::edit` (`iris/src/widget/text/edit.rs:582-587`) resolves
+   that handle with `ui.widgets.get_mut(self).unwrap()` — an unconditional
+   panic on the freed slot. `Selection::begin` (`selection.rs:88-101`)
+   iterates *every* registered row (`w.edit(ui).deselect()`) on an
+   ordinary fresh press, so the crash fires on the next tap anywhere in
+   the transcript after a regroup, not only on a tap targeting the
+   orphaned row.
+   Fix: give `Selection` a way to reconcile against the row set that
+   survived a rebuild (e.g. `Selection::retain(&self, keys: &BTreeSet<RowKey>)`
+   removing everything else, called from the `Rebuild` arm before
+   rebuilding), or simplest — call `self.selection.borrow_mut()` cleared
+   the same way `List::clear()` clears the list, then let the rebuild's
+   `push_row` calls re-`register` everything as they already do.
+
+## Guarded invariants missing
+
+2. **`iris/src/widget/list.rs:751` (`List::place`) indexes/expects on
+   `slot` with no assertion that it exists.** `slot_widget` (`:563-575`)
+   panics via `.expect(...)` for a sentinel with no widget set, and does
+   an unchecked `&self.items[s as usize]` for a real index — a bare
+   "index out of bounds" with no context if `place` is ever reached with a
+   stale slot. Every current caller happens to derive `slot` from
+   `repair_anchor`/`prev_slot`/`next_slot`, which already check existence,
+   but that invariant is enforced by convention across three call sites,
+   not by the function that depends on it. Add
+   `debug_assert!(self.slot_exists(slot), "place() called with a slot that doesn't exist: {slot:?}");`
+   at the top of `place`.
+3. **`iris/src/widget/list.rs:426` (`List::fling`) and `sense.rs`'s
+   `FlingCalculator::distance`/`duration`/`position_at` never check that
+   the incoming velocity is finite.** A `NaN`/`inf` velocity (a
+   `VelocityTracker::velocity()` divide-by-near-zero span, or a caller
+   passing a raw device value straight through) propagates through
+   `deceleration_for`'s `.ln()` silently — the fling either never settles
+   (`settled_on_schedule` compares against a `NaN` `duration()`, which is
+   always `false`) or jumps to `NaN` positions with nothing on screen
+   saying why. Add `debug_assert!(velocity_px_per_s.is_finite())` in
+   `List::fling` and `FlingCalculator::new`/`distance`.
+4. **`iris/src/sense.rs:592-604` (`VelocityTracker::velocity`) has no
+   assertion that samples are chronological.** `add_sample` trusts its
+   caller's `Instant` ordering; a caller that samples out of order (a
+   restored/replayed gesture, a test) would silently produce a negative
+   `span` handled only by the `span <= 0.0 => 0.0` catch-all, masking the
+   bug that produced it rather than surfacing it. Add
+   `debug_assert!(self.samples.back().is_none_or(|&(last, _)| at >= last))`
+   in `add_sample`.
+5. **`iris/core/src/render/frame_report.rs:247-252` (`mark_phase`) has no
+   assertion that phases are pushed in non-decreasing `start_index`
+   order.** `phase_stats`'s slicing (`:274`, `idx >= phase.start_index &&
+   idx < end_index`) silently produces an empty or nonsensical slice for
+   an out-of-order phase rather than surfacing the misuse — cheap to add
+   given `self.phases.last()` is already in scope:
+   `debug_assert!(self.phases.last().is_none_or(|p| self.total_frames >= p.start_index));`
+
+## Rules
+
+6. **Two mechanisms answer "what row selection points at, still valid?"**
+   `Selection` relies on callers remembering to `unregister` (finding 1);
+   `List` relies on callers deriving slots only from already-checked
+   sources (finding 2). Both are the same class of problem — a derived
+   handle that silently outlives what it points to — solved ad hoc twice
+   rather than once. Not asking for a shared abstraction here, but the two
+   should at minimum cross-reference each other's doc comment so the next
+   caller who adds a third handle-into-`List`-rows type (the code rules'
+   "a rule that governs a set belongs to the set") finds both existing
+   examples.
+7. **`iris/android-app/src/bench_client.rs:224-225` (`battery_line`)
+   calls `.min().unwrap()`/`.max().unwrap()` on `samples` guarded three
+   lines above by `if samples.is_empty()`, which is fine — but the guard
+   and the two unwraps are two statements apart with a `let mean = ...`
+   in between reading the same slice; a future edit reordering those
+   lines loses the guard's protection silently.** Low severity (this is
+   the bench tool, not the app), but worth a one-line comment tying the
+   unwraps back to the guard, or restructuring as
+   `let (Some(min), Some(max)) = (samples.iter().min(), samples.iter().max())`
+   pattern so the empty case can't be separated from the check by a future
+   edit.
+
+## Tests
+
+8. **No test exercises `TranscriptScreen::apply`'s `Rebuild` arm through
+   `Selection`.** `lib.rs`'s `diff_tests` module (`:284-379`) tests only
+   the pure `diff_rows` decision function, never `apply` itself wired to a
+   real `Selection`; `selection.rs`'s own tests (`a_missed_press_start_
+   recovers_on_the_next_pressing_frame`, `unregister_forgets_the_row_and_
+   clears_a_matching_anchor`) never go through `apply`/`List::clear`
+   either. This is exactly the gap that let finding 1 through: the two
+   pieces (`apply`'s fallback, `Selection`'s registration contract) are
+   each tested in isolation and never together. Add: build a
+   `TranscriptScreen`, force a `RowDiff::Rebuild` (two adjacent tool-call
+   rows regrouping, per the existing `diff_tests` case), then call
+   `selected_text`/simulate a fresh press on a surviving row and assert no
+   panic.
+9. **`iris/src/widget/list.rs`'s fling tests check total distance and the
+   start/end clamp but not the speed profile in between.**
+   `fling_moves_the_list_and_then_settles`/`fling_distance_is_positive_
+   toward_the_end` only assert the fling started, moved in the right
+   direction, and eventually stopped — none checks that
+   `tick_fling`'s per-tick delta is *monotonically decreasing* once past
+   the fling's peak (the property `fling_calculator_tests::position_at_
+   is_monotonic_and_clamped_past_the_end` already checks one level down,
+   for `FlingCalculator` alone, but never through `List::tick_fling`'s own
+   `scroll`/`anchor.offset` accumulation). A regression that made
+   `tick_fling` apply the *total* distance every tick instead of the
+   incremental one, for instance, would still pass both existing tests
+   (final position and direction are unaffected by how the interior ticks
+   split it up) while being wildly wrong every intermediate frame.
+10. **`iris/src/widget/list.rs::replacing_the_last_row_stays_pinned_to_
+    the_bottom` and its sibling test `replace_back`'s effect on the
+    displayed row, never that the row it evicted is actually gone from
+    `heights`/`extents`.** Both tests assert the *new* row's position;
+    neither asserts `old.key` is absent from `list_ref.heights`/`extents`
+    after the replace (the "stale primitive" class finding 1 is a
+    production instance of). A cheap addition: assert
+    `!list_ref.heights.contains_key(&old.key)` after `replace_back` in the
+    existing test, since `old.key` is already returned to the test as
+    `evicted`... (`lib.rs` calls it that way; the `list.rs` test would need
+    to capture the key from `old` similarly.)
+
+## Docs
+
+No missing `IRIS.md` entry found for a *public* API change in this diff —
+`List::fling`/`VelocityTracker`/`FlingCalculator`, `List::
+anchor_position_display`, `FrameReport::mark_phase`/`phase_stats`/
+`late_at_hz`, `UiRenderNode::new`'s `Result` change, `Len::dp`, and
+`List::replace_back`/`clear`/`TranscriptScreen::apply` all have entries.
+The `List::replace_back`/`clear`/`TranscriptScreen::apply` entry
+(`docs/IRIS.md:526`) predates this review's finding 1 and does not mention
+`Selection`'s registration contract at all — once finding 1 is fixed,
+that entry should gain a line noting what the fix requires of a caller
+that keeps its own row-keyed side table (the same shape `Selection` is),
+so the next such table doesn't reproduce the same gap.
+
+## Fixed, 2026-09-06
+
+All ten findings addressed after the `DragGesture` merge (`selection.rs`
+was rewritten by that merge, but finding 1's shape and location were
+unchanged — `TranscriptScreen::apply`'s `Rebuild` arm, `iris/transcript-ui/
+src/lib.rs`).
+
+1. **Fixed.** `Selection::clear()` (`selection.rs`) drops `rows` and
+   `anchor`, called from `apply`'s `Rebuild` arm right before
+   `List::clear()` — `push_row` re-`register`s whatever survives as it
+   rebuilds each row, the "simplest" fix option the finding named.
+2. **Fixed.** `debug_assert!(self.slot_exists(slot), ...)` at the top of
+   `List::place` (`iris/src/widget/list.rs`).
+3. **Fixed.** `debug_assert!(velocity_px_per_s.is_finite())` in
+   `List::fling`, and `debug_assert!(velocity.is_finite())` in
+   `FlingCalculator::distance`/`duration` (`iris/src/sense.rs`).
+   `position_at` calls both, so it inherits the guard rather than needing
+   its own.
+4. **Fixed.** `debug_assert!` on chronological sample order in
+   `VelocityTracker::add_sample` (`iris/src/sense.rs`).
+5. **Fixed.** `debug_assert!` on non-decreasing `start_index` in
+   `FrameReport::mark_phase` (`iris/core/src/render/frame_report.rs`).
+6. **Fixed (doc cross-reference only, as asked).** `Selection::register`'s
+   doc now points at `List::place`'s `slot_exists` assertion and vice
+   versa isn't needed since finding 2's fix already cites this file in
+   its own comment; both are grep-able on "docs/REVIEW-2026-09-06.md" and
+   on each other's type names.
+7. **Fixed.** `bench_client.rs::battery_line` restructured to
+   `let (Some(min), Some(max)) = (samples.iter().min(), samples.iter().max())`,
+   so the empty-guard and the two lookups can no longer be separated by a
+   future edit.
+8. **Fixed.** `transcript-ui`'s new `apply_tests::
+   a_row_dropped_by_a_regroup_does_not_outlive_itself_in_selection`
+   (`lib.rs`) builds a real `TranscriptScreen`, forces the same regroup
+   shape `diff_tests` already covers at the pure-diff level, calls `apply`,
+   and then `Selection::begin` on a surviving row — which panicked before
+   fix 1, resolving a `WeakWidget` `List::clear()` had just freed.
+9. **Fixed.** `list.rs`'s new `tick_fling_applies_shrinking_incremental_
+   deltas` flings toward the end from `jump_to_start` and asserts each
+   tick's `extents[&0]` delta is no larger than the previous one — would
+   fail against a `tick_fling` that applied the total spline distance
+   every tick instead of the incremental slice, which the two pre-existing
+   fling tests cannot catch.
+10. **Fixed.** `list.rs`'s new `replace_back_forgets_the_evicted_keys_own_
+    height` replaces row 4 with a row keyed `100` (the two existing
+    `replace_back` tests always reuse the same key, so neither actually
+    exercises the removal) and asserts `heights` no longer contains the
+    evicted key.
+
+Docs: `docs/IRIS.md`'s 2026-09-05 `List::replace_back`/`clear`/
+`TranscriptScreen::apply` entry now has a line on what the fix requires of
+a caller with its own row-keyed side table, naming `Selection` as the
+example and dating the fix.
+
+Verification run alongside the rest of this pass's checks: `cargo fmt
+--all`, `cargo clippy --workspace --all-targets`, `cargo test --workspace`
+from `iris/` — see docs/RUST.md's plan box for the pass/fail and any
+caveats from this same session.

docs/TODO.md

@@ -33,3 +33,14 @@ one in place when it turns out to need a decision.
       that would work today, for Claude sessions, and it is the option that was
       not chosen.
 
+
+## From Iris's phone log export, 2026-09-07 (Compose app)
+
+- [ ] **Crash on 2026-09-03 11:40, `IllegalArgumentException: Reversed
+  range is not supported`** at `ToolInput.kt:200` (`highlighted`, inside
+  `ToolInputView` -> `RawBlock` -> `ToolCard`). An `AnnotatedString`
+  range was built with end before start while highlighting a tool
+  input. Found in the per-package system log she exported; the tool
+  input that triggered it is not in the log. Reproduce by fuzzing
+  `highlighted` with inputs whose token boundaries collapse, and guard
+  the range construction.

docs/bench/compose-phone-2026-09-06.md

@@ -0,0 +1,73 @@
+# Compose bench report from Iris's phone, 2026-09-06
+
+The Compose half of P0 (RUST.md), run by Iris on her own phone and pasted
+back verbatim. The iris half's report goes beside it in this directory
+when it exists. Her caveat, worth keeping with the numbers: "I don't think
+this is entirely fair because the UI for iris is more minimal" -- the
+Compose screen also draws the usage bar, the status row and tool cards,
+which the iris bench screen does not yet. Her impression of the iris build
+before its first-touch bug: "it already feels very smooth so far".
+
+What to read first: the phone runs at 120 Hz, so the budget is 8.3 ms;
+`late` is measured against that. Compose's tail is the streaming phase --
+`markdown reparsed while streaming: 396, 8.5ms mean, 25.8ms worst` and
+`record: one block: 398, 6.3ms mean, 19.9ms worst` -- which is exactly the
+path iris's `TranscriptScreen::apply` (replace the last row only) is meant
+to beat. Process CPU over the run is 20.9 s of a 38.5 s run; peak RSS
+587 MB; battery current mean 419 mA.
+
+```
+ai-app render report
+device: Pixel 9 Pro XL (Google), Android 17
+build: release
+
+transcript:
+  43 events, 41 rows, 93 units loaded
+  viewport 1333px, 2 units visible
+  on screen: the list's own 0px, AssistantMsg 24520px
+  0 tool calls and 0 groups open
+
+frames:
+  1613 frames over 38.5s at 120Hz (8.3ms budget)
+  late: 742 (46.0%)
+  total  p50 7.7ms  p90 29.2ms  p99 41.1ms
+  waited p50 0.5ms  p90 12.9ms  p99 27.2ms
+  input  p50 0.0ms  p90 0.0ms  p99 0.0ms
+  anim   p50 1.1ms  p90 5.8ms  p99 9.5ms
+  layout p50 0.0ms  p90 0.1ms  p99 0.2ms
+  draw   p50 0.4ms  p90 15.9ms  p99 27.9ms
+  sync   p50 0.1ms  p90 0.5ms  p99 1.0ms
+  issue  p50 1.1ms  p90 1.7ms  p99 3.0ms
+  swap   p50 0.4ms  p90 0.5ms  p99 0.7ms
+  gpu    p50 1.8ms  p90 2.1ms  p99 6.6ms
+
+where the draw phase went:
+  draw phase 3.83ms per frame, of which:
+    the transcript: 0.25ms (measure 0.15, place 0.10, record 0.00)
+    everything else: 3.58ms (93%)
+
+work since this was last copied:
+  draw: the whole transcript: 12, 0.2ms total, 0.0ms mean, 0.0ms worst
+  grouped tool runs: 398, 10.3ms total, 0.0ms mean, 0.1ms worst
+  markdown cut into pieces: 1, 0.0ms total, 0.0ms mean, 0.0ms worst
+  markdown parsed while composing: 7, 3.0ms total, 0.4ms mean, 0.5ms worst
+  markdown ready: 46
+  markdown reparsed while streaming: 396, 3384.6ms total, 8.5ms mean, 25.8ms worst
+  markdown warmed: 1, 1.4ms total, 1.4ms mean, 1.4ms worst
+  measure: the whole transcript: 957, 248.7ms total, 0.3ms mean, 15.7ms worst
+  message composed: 403
+  message cut into parts: 1, 0.1ms total, 0.1ms mean, 0.1ms worst
+  place: the whole transcript: 1319, 162.4ms total, 0.1ms mean, 1.3ms worst
+  record: one block: 398, 2498.7ms total, 6.3ms mean, 19.9ms worst
+  session screen recomposed: 413
+  status row recomposed: 1
+  unit composed: 538
+  units flattened: 399, 44.3ms total, 0.1ms mean, 0.5ms worst
+  usage bar recomposed: 413
+
+bench:
+  scroll: 6 cycles (24 swipes), streamed 400/400 fixture events
+  process CPU time over this run: 20907ms
+  peak RSS: 587356kB
+  battery current: mean -418509µA over 39 samples (min -1988281, max -107812)
+```

docs/bench/compose-phone-v2-2026-09-06.md

@@ -0,0 +1,93 @@
+# Compose bench v2 report from Iris's phone, 2026-09-06
+
+Bench v2 (fling / stream / type / keyboard, RUST.md's P0 box) on the
+Compose `bench` build, run by Iris on her Pixel 9 Pro XL, verbatim. Note
+the display was at **60 Hz** for this run (16.7 ms budget) where the v1
+run was at 120 Hz -- the phone's adaptive refresh rate decides, and
+`late` is judged against whichever it was, so compare a run with a run at
+the same rate. The iris v2 report goes beside this when it exists.
+
+What it says: fling, type and keyboard are all essentially clean on
+Compose (0.1%, 0.9% and 0% late; fling p50 5.5 ms, p99 11.6 ms). The
+whole tail is the streaming phase again -- 41.9% late, p99 42.5 ms,
+driven by `markdown reparsed while streaming` (8.6 ms mean, 30.3 ms
+worst) and `record: one block` (6.3 ms mean, 25.5 ms worst). Process CPU
+69.6 s over the 125.5 s run; peak RSS 577 MB; battery current mean
+571 mA over 126 samples.
+
+```
+ai-app render report
+device: Pixel 9 Pro XL (Google), Android 17
+build: release
+
+transcript:
+  108 events, 26 rows, 58 units loaded
+  viewport 1531px, 2 units visible
+  on screen: the list's own 0px, AssistantMsg 24520px
+  0 tool calls and 0 groups open
+
+per phase:
+  fling: 3278 frames over 32.7s
+    late: 4 (0.1%)
+      total  p50 5.5ms  p90 8.7ms  p99 11.6ms
+    worst 49.0ms
+  stream: 1041 frames over 21.3s
+    late: 436 (41.9%)
+      total  p50 13.4ms  p90 31.7ms  p99 42.5ms
+    worst 52.5ms
+  type: 2446 frames over 61.5s
+    late: 23 (0.9%)
+      total  p50 7.3ms  p90 13.2ms  p99 16.5ms
+    worst 38.9ms
+  keyboard: 358 frames over 10.0s
+    late: 0 (0.0%)
+      total  p50 6.3ms  p90 8.6ms  p99 11.1ms
+    worst 12.0ms
+
+frames:
+  7122 frames over 125.5s at 60Hz (16.7ms budget)
+  late: 463 (6.5%)
+  total  p50 6.0ms  p90 13.8ms  p99 34.0ms
+  waited p50 0.5ms  p90 1.1ms  p99 19.9ms
+  input  p50 0.0ms  p90 0.0ms  p99 0.0ms
+  anim   p50 0.7ms  p90 4.5ms  p99 7.6ms
+  layout p50 0.1ms  p90 0.1ms  p99 0.2ms
+  draw   p50 0.7ms  p90 2.9ms  p99 21.9ms
+  sync   p50 0.1ms  p90 0.2ms  p99 0.6ms
+  issue  p50 1.4ms  p90 2.4ms  p99 3.2ms
+  swap   p50 0.4ms  p90 0.8ms  p99 1.2ms
+  gpu    p50 1.5ms  p90 2.1ms  p99 6.6ms
+
+where the draw phase went:
+  draw phase 1.74ms per frame, of which:
+    the transcript: 0.24ms (measure 0.10, place 0.14, record 0.00)
+    everything else: 1.51ms (86%)
+
+work since this was last copied:
+  draw: the whole transcript: 280, 1.9ms total, 0.0ms mean, 0.0ms worst
+  grouped tool runs: 407, 18.6ms total, 0.0ms mean, 0.2ms worst
+  markdown cut into pieces: 40, 0.4ms total, 0.0ms mean, 0.0ms worst
+  markdown parsed while composing: 2, 1.6ms total, 0.8ms mean, 1.1ms worst
+  markdown ready: 323
+  markdown reparsed while streaming: 395, 3406.9ms total, 8.6ms mean, 30.3ms worst
+  markdown warmed: 40, 38.2ms total, 1.0ms mean, 4.3ms worst
+  measure: the whole transcript: 1978, 683.9ms total, 0.3ms mean, 15.9ms worst
+  message composed: 397
+  message cut into parts: 40, 2.9ms total, 0.1ms mean, 0.2ms worst
+  place: the whole transcript: 4308, 996.0ms total, 0.2ms mean, 2.4ms worst
+  record: one block: 394, 2472.7ms total, 6.3ms mean, 25.5ms worst
+  session screen recomposed: 1630
+  status row recomposed: 1
+  transcript page from server: 10
+  unit composed: 927
+  units flattened: 408, 85.5ms total, 0.2ms mean, 1.8ms worst
+
+bench:
+  fling: 8 flings out + 8 back at 12000px/s, travel start=idx=0/off=0px outward=idx=188/off=182px end=idx=0/off=0px
+  scroll: 6 cycles (24 swipes, legacy tween), streamed 400/400 fixture events
+  type: 600 characters inserted then deleted, one per 50ms
+  keyboard: shown 5/5, hidden 5/5 (confirmed via isImeVisible)
+  process CPU time over this run: 69564ms
+  peak RSS: 577452kB
+  battery current: mean -571483µA over 126 samples (min -2361718, max -99218)
+```

docs/bench/iris-phone-2026-09-06-before-fixes.md

@@ -0,0 +1,31 @@
+# iris bench report from Iris's phone, 2026-09-06, before the phone fixes
+
+Build 46246ea (Vulkan, bench v1: 24-swipe scroll loop then 400 streamed
+events), run by Iris on her Pixel 9 Pro XL before the first-touch wipe,
+the missing bold faces, the density scale and the status-bar inset were
+fixed -- so the rows were drawn at roughly a third of their intended size
+and the run may have included frames after the wipe. Preliminary, kept
+because it is the first iris number from real hardware. Compare with
+`compose-phone-2026-09-06.md`, taken on the same phone with the same
+fixture and gesture loop.
+
+Reading it: the phone is 120 Hz (8.3 ms budget). `janky%` here counts
+frames over 16.7 ms, so it is not Compose's `late` (over 8.3 ms). Like for
+like: iris p50 6.2 ms vs Compose 7.7 ms; p90 32.0 vs 29.2; p99 42.1 vs
+41.1. `cpu_p50=4.7ms` is iris's own per-frame CPU work on the phone,
+against 0.2-0.4 ms on the emulator's x86 cores. Process CPU 15.6 s vs
+20.9 s, but over a shorter run (692 frames vs 1613 -- iris only renders on
+change and had no fling settle time), so per-second CPU is not directly
+comparable; peak RSS 365 MB vs 587 MB. Battery current mean 563 mA vs
+419 mA is the one figure that reads worse, and it is the least
+comparable: 22 samples vs 39, over runs of different length and different
+idle share. Bench v2's per-phase accounting is what makes these comparable.
+
+```
+iris bench report
+frames=692 janky%=32.37 p50=6.2ms p90=32.0ms p99=42.1ms worst=52.6ms (measures redraw-start to after present() is called, not GPU/compositor completion) cpu_p50=4.7ms gpu_wait_p50=1.3ms (redraw-start-to-submit vs. submit-to-after-present)
+  scroll: 6 cycles (24 swipes), streamed 400/400 fixture events
+  process CPU time over this run: 15554ms
+  peak RSS: 365328kB
+  battery current: mean -563493µA over 22 samples (min -1807812, max -132812)
+```

docs/bench/iris-phone-v2-2026-09-06.md

@@ -0,0 +1,66 @@
+# iris bench v2 report from Iris's phone, 2026-09-06
+
+Build 2e3f4ad (bench v2, fling physics, keyboard-wipe fix, dp unit), run
+by Iris on her Pixel 9 Pro XL, verbatim. The display was at **120 Hz**
+(8.3 ms budget) where `compose-phone-v2-2026-09-06.md` ran at 60 Hz, so
+compare the millisecond percentiles, not `late`.
+
+Side by side (Compose 60 Hz / iris 120 Hz, p50 / p90 / p99 ms): fling
+5.5/8.7/11.6 vs 3.8/6.9/12.6; stream 13.4/31.7/42.5 vs 18.2/35.8/43.1;
+type 7.3/13.2/16.5 vs 7.2/9.2/11.2; keyboard: iris could not show the IME
+(phase invalid). Process CPU 69.6 s over 125 s vs 40.6 s over 150 s; peak
+RSS 577 MB vs 379 MB; battery current mean 571 mA vs 452 mA.
+
+Iris's observations on the same run: "the scrolling is not similar at
+all. It does not fling for me yet [with a finger], and the test also seems
+to give it a constant velocity and abruptly stop it at some point. Also
+unsure what's going on in that image with the compaction" -- her
+screenshot shows the `Compacted: 180000 -> 20000 tokens.` row drawn twice
+overlapping, and once more below the composer bar: primitives of a
+replaced/removed row surviving in the GPU buffers, the same shape as the
+header drawn twice after a keyboard resize.
+
+**Root-caused and fixed 2026-09-06** (commit `76b1f99`): the diagnosis in
+that sentence was right and the location was not -- `UiRenderState::
+draw_inner` read the `needs_redraw` mark without consuming it and skipped
+the branch that frees a redrawn widget's old primitives. docs/RUST.md's
+"Stale primitives, the phone's half" box has the full account, the guard
+(`orphaned_primitives`, `debug_assert`ed every frame) and the emulator run
+that exercises it.
+
+```
+iris bench report
+per phase:
+  fling: 1783 frames over 53.2s
+    late: 104 (5.8%)
+      total  p50 3.8ms  p90 6.9ms  p99 12.6ms
+    worst 29.1ms
+  stream: 401 frames over 21.3s
+    late: 306 (76.3%)
+      total  p50 18.2ms  p90 35.8ms  p99 43.1ms
+    worst 43.8ms
+  type: 1202 frames over 65.7s
+    late: 309 (25.7%)
+      total  p50 7.2ms  p90 9.2ms  p99 11.2ms
+    worst 15.3ms
+  keyboard: 9 frames over 9.7s
+    late: 9 (100.0%)
+      total  p50 12.0ms  p90 12.9ms  p99 12.9ms
+    worst 12.9ms
+
+frames:
+  3395 frames over 149.9s at 120Hz (8.3ms budget)
+  late: 728 (21.4%)
+  total  p50 5.0ms  p90 10.9ms  p99 36.6ms
+  worst 43.8ms
+  cpu_p50 2.0ms  gpu_wait_p50 2.6ms
+
+bench:
+  fling: 8 flings out + 8 back at 12000px/s, travel start=idx=651/off=1217px outward=idx=651/off=101536px end=idx=651/off=1022px
+  scroll: 6 cycles (24 swipes, legacy tween), streamed 400/400 fixture events
+  type: 600 characters inserted then deleted, one per 50ms
+  keyboard: could not be shown (5 attempts, 0 confirmed visible)
+  process CPU time over this run: 40603ms
+  peak RSS: 379156kB
+  battery current: mean -452353µA over 149 samples (min -1753125, max -204687)
+```

event-model/src/lib.rs

@@ -150,6 +150,19 @@ pub enum Event {
     ToolEnd {
         id: String,
         output: String,
+        /// Whether the tool reported that the call *failed*, from the
+        /// CLI's own `is_error` on the `tool_result`.
+        ///
+        /// Added 2026-09-06 with the tool-call cards (RUST.md's P1b),
+        /// because without it a result is the only thing a card has and a
+        /// failed call is drawn as confidently as a successful one -- the
+        /// missing state, not a wrong one. `#[serde(default)]` so a
+        /// transcript written before this field, or a peer on an older
+        /// build, reads back as "not reported to have failed" rather than
+        /// failing to parse; that is the same claim the field's absence
+        /// used to make implicitly.
+        #[serde(default)]
+        is_error: bool,
     },
     /// An image the session produced or was sent, saved under the session
     /// dir and referenced by id; the phone fetches it by URL.
@@ -305,6 +318,25 @@ pub enum Event {
     /// it, which is why this is written down rather than left to be inferred
     /// from a second example that does not exist.
     Cleared,
+    /// The account behind this session has no quota left, so the turn stopped
+    /// without finishing.
+    ///
+    /// Its own event rather than an [`Event::Error`] carrying the dialect's
+    /// sentence, because two things act on it that cannot read English: the
+    /// transcript draws it as a state the session is in rather than as a
+    /// failure of something it did, and `crate::resume` schedules the message
+    /// that picks the work back up. Recognising it belongs to the driver, which
+    /// is the only layer that knows its dialect's wording -- above here nothing
+    /// matches on strings.
+    ///
+    /// `resets_at` is epoch seconds, and `None` is a real state: the dialect
+    /// said the limit was hit without saying when it lifts. Nothing here
+    /// invents one -- what the wait is actually decided against is the usage
+    /// endpoint, and this is the hint that starts the waiting.
+    LimitReached {
+        #[serde(default, skip_serializing_if = "Option::is_none")]
+        resets_at: Option<f64>,
+    },
     Error {
         message: String,
     },

iris/Cargo.toml

@@ -13,7 +13,13 @@ swash = { workspace = true }
 pollster = { workspace = true }
 wgpu = { workspace = true }
 image = { workspace = true }
+accesskit = { workspace = true }
 tokio = { workspace = true, features = ["sync", "rt", "rt-multi-thread"] }
+# For diagnostics visible through android_logger (or whatever logger the
+# app crate installs) -- this crate never installs one itself. Not in the
+# android-only block below any more: the lines that matter most are in
+# shared widget code, which the host backend compiles too.
+log = "0.4.28"
 
 # winit everywhere except Android; android-view (below) is what stands in
 # for it there. Both backends live in this crate (see `src/android/mod.rs`'s
@@ -27,6 +33,11 @@ tokio = { workspace = true, features = ["sync", "rt", "rt-multi-thread"] }
 [target.'cfg(not(target_os = "android"))'.dependencies]
 winit = { workspace = true }
 arboard = { workspace = true, features = ["wayland-data-control"] }
+# I4 (RUST.md): the desktop half of the AccessKit push, `winit`'s own
+# adapter over `accesskit`. No pin needed the way android-view's rev is
+# pinned -- this is an ordinary crates.io release with no local abort to
+# track (that finding is Android-only, see below).
+accesskit_winit = "0.34.0"
 
 # Pinned to the exact commit RUST.md's E1 (2026-09-04) measured on this
 # emulator -- real Vulkan rendering, a working `InputConnection`, and the
@@ -35,13 +46,32 @@ arboard = { workspace = true, features = ["wayland-data-control"] }
 # is dated rather than floating.
 [target.'cfg(target_os = "android")'.dependencies]
 android-view = { git = "https://github.com/rust-mobile/android-view.git", rev = "bec6c62a96cef8239b0fd7fedeef9b184d02e3a1" }
+# I4 (RUST.md): the Android half of the AccessKit push, over android-view's
+# `AccessibilityNodeProvider`. **0.8.0 carries the same detach-abort E1
+# found on 0.4.0** (the `State` enum still never returns to `Inactive`,
+# and `send_completed_event` still unwraps a Java exception) -- advancing
+# the version is not the fix, so pinning to a specific rev buys nothing
+# here the way it does for android-view itself. `android/view.rs`'s
+# `raise_if_enabled` is the mitigation, carried from E1.
+accesskit_android = "0.8.0"
 # Not re-exported by android-view (only `jni` and `ndk` are), and needed
 # for `android/insets.rs`'s own id -> state map -- the same reason
 # android-view's own `PEER_MAP` carries one.
 send_wrapper = "0.6.0"
-# For diagnostics visible through android_logger, wherever the app crate
-# installs it -- this crate never installs a logger itself.
-log = "0.4.28"
+
+[features]
+# RUST.md's I5 "Where iris's frame time goes" diagnosis: forces the Android
+# `wgpu::Instance` to `Backends::GL` instead of `Backends::PRIMARY`, so the
+# same build can be measured against SwiftShader's software Vulkan ICD (the
+# default) or virgl's GLES path, without a second env-var plumbing path that
+# nothing on this machine can hand to an already-launched Android process
+# (there is no `am start` environment and no system-property reader here to
+# add one). Read by `android/render.rs` and, so the GLES path can be
+# reproduced on a machine with a real GPU rather than only in the emulator,
+# by `default/render.rs`:
+#     ./run-headless.sh transcript --shot /tmp/x.png -- -p transcript-ui \
+#         --features iris/force-gles
+force-gles = []
 
 [dev-dependencies]
 tokio = { workspace = true, features = ["sync", "rt", "rt-multi-thread", "time"] }
@@ -50,14 +80,30 @@ tokio = { workspace = true, features = ["sync", "rt", "rt-multi-thread", "time"]
 # to build the library itself, so this only matters for `--examples`.
 tabs-ui = { path = "tabs-ui" }
 
+# Plain Instant-timed binaries, not criterion -- see benches/message_list.rs's
+# header for why. `harness = false` opts out of the unstable `#[bench]`
+# test-crate harness cargo would otherwise want, in favour of an ordinary
+# `fn main()`.
+[[bench]]
+name = "message_list"
+harness = false
+
 [workspace]
-members = ["core", "macro", "tabs-ui"]
+members = [
+    "core",
+    "macro",
+    "tabs-ui",
+    "transcript-ui",
+    "transcript-fixture",
+    "rig-input",
+    "desktop-app",
+]
 # android-app pulls in android-view, which needs the NDK sysroot to link
 # -- excluded so `cargo build --workspace --all-targets` on the host stays
 # buildable. Cross-compile it from its own directory (its own single-crate
 # workspace, since it has no `[workspace]` table of its own and this
 # exclusion stops it inheriting this one): `cd android-app && cargo ndk
-# -t x86_64 -P 26 build`.
+# -t x86_64 -P 29 build`.
 exclude = ["android-app"]
 
 [workspace.package]
@@ -74,6 +120,12 @@ parley = "0.11.1"
 swash = "0.2.10"
 fxhash = "0.2.1"
 arboard = "3.6.1"
+accesskit = "0.25.0"
 iris-core = { path = "core" }
 iris-macro = { path = "macro" }
 tokio = "1.49.0"
+# Current stable as of 2026-09-05 (`cargo search`) -- I5's markdown block
+# model, the same crate E2's uncommitted `e2-transcript` experiment used for
+# the identical job (RUST.md), rather than reimplementing a CommonMark
+# parser.
+pulldown-cmark = "0.13.4"

iris/android-app/Cargo.toml

@@ -17,10 +17,59 @@ crate-type = ["cdylib"]
 
 [dependencies]
 iris = { path = "../" }
-tabs-ui = { path = "../tabs-ui" }
 android-view = { git = "https://github.com/rust-mobile/android-view.git", rev = "bec6c62a96cef8239b0fd7fedeef9b184d02e3a1" }
 android_logger = "0.15.0"
 log = "0.4.28"
+# `tabs-screen` (default, I2/I4's demo) and `transcript-screen` (I5's
+# Android integration) are mutually exclusive -- one `ActiveClient` type is
+# compiled in, never both (`lib.rs`'s doc comment) -- so both sets of deps
+# are optional and each screen's feature pulls in only its own. Without
+# this, building `--features transcript-screen` alone (default features
+# still on) left `tabs-ui` linked but never referenced under that cfg,
+# which Cargo's `unused_dependencies` lint (on by default) correctly flags.
+tabs-ui = { path = "../tabs-ui", optional = true }
+transcript-ui = { path = "../transcript-ui", optional = true }
+# P0's bench build only: the fixture and the folded screen both bench
+# clients open, shared with the headless harness and the desktop window
+# (docs/RUST.md's "Three test layers").
+transcript-fixture = { path = "../transcript-fixture", optional = true }
+client-core = { path = "../../client-core", optional = true }
+event-model = { path = "../../event-model", optional = true }
+serde_json = { version = "1", features = ["float_roundtrip"], optional = true }
+# P0's bench build only (docs/RUST.md): `getrusage(RUSAGE_SELF)` for
+# process CPU time, matching `libc::getrusage`'s mention in that box over
+# parsing `/proc/self/stat` by hand and assuming `USER_HZ`. Already in the
+# workspace's own dependency tree transitively (`iris/Cargo.lock`, pinned
+# at 0.2.179) -- this makes it a direct dependency at the same version
+# rather than a second, possibly-drifting resolution.
+libc = { version = "0.2.179", optional = true }
+# P0's bench build only: the scroll animation and the streaming phase are
+# both a sequence of `sleep`s inside the async task `rsc.spawn_task` already
+# runs on iris's own tokio runtime (`iris/src/task.rs`'s `Tasks::init`), and
+# the battery sampler is a second, concurrent task on that same runtime
+# (`tokio::spawn`) -- so this crate needs `tokio` directly rather than only
+# through `iris`. `rt`+`time` only: no I/O, no macros, nothing this crate
+# doesn't call. Version matches the one `iris`'s own dependency tree already
+# resolves to (`iris/Cargo.lock`), so there is one copy of the runtime, not
+# two.
+tokio = { version = "1.53.1", features = ["rt", "time"], optional = true }
+
+[features]
+default = ["tabs-screen"]
+tabs-screen = ["dep:tabs-ui"]
+transcript-screen = ["dep:transcript-ui", "dep:client-core", "dep:event-model", "dep:serde_json"]
+# RUST.md's I5 "Where iris's frame time goes": forces the GLES backend
+# instead of SwiftShader's software Vulkan. See `iris/Cargo.toml`'s own doc
+# on the feature this forwards to.
+force-gles = ["iris/force-gles"]
+# P0's iris half (docs/RUST.md, docs/AGENTS.md's "The rigs"): the same
+# checked-in fixture, scroll loop and streaming phase the Compose `bench`
+# build type drives, run here against `transcript-ui`'s real screen with no
+# server. Depends on `transcript-screen` for `transcript-ui`/`client-core`/
+# `event-model` -- `lib.rs`'s `ActiveClient` selection gives this feature
+# priority over `transcript-screen`'s own `TranscriptClient` when both are
+# listed, which is how this crate's build command names both explicitly.
+bench = ["transcript-screen", "dep:transcript-fixture", "dep:libc", "dep:tokio"]
 
 [profile.release]
 panic = "abort"

iris/android-app/app/build.gradle

@@ -13,15 +13,74 @@ android {
 
     defaultConfig {
         applicationId = "dev.iris.android.demo"
-        minSdk = 26
-        targetSdk = 34
+        // 29, not 26: `iris::android::view`'s touch handler dates each
+        // sample with `MotionEvent.getEventTimeNanos` and
+        // `getHistoricalEventTimeNanos`, both API 29, and a missing JNI
+        // method there is a hard crash on the first touch rather than a
+        // degraded fling. Raised deliberately rather than guarded at
+        // runtime: nothing this app is built for runs below 29, and an
+        // untested fallback path is its own defect. `build-apk.sh`'s
+        // `cargo ndk -P` is kept at the same number.
+        minSdk = 29
+        // 37, matching `compileSdk` and the Compose app in `app/` -- which
+        // is the one part of this that is measured rather than reasoned:
+        // that app targets 37 and its keyboard does push the transcript up
+        // on Iris's phone, and this one targeted 34 and does not
+        // (2026-09-07). The emulator here is API 36 and the push-up works
+        // there at either target, so the target is the only difference the
+        // two devices do not share.
+        //
+        // The mechanism, stated as the reading it is: below targetSdk 35
+        // a window keeps the legacy behaviour, where `adjustResize` shrinks
+        // the window for the IME and `getInsets(ime()).bottom` therefore
+        // measures the overlap with an already-shrunk window -- zero, with
+        // nothing left to push up. `MainActivity`'s
+        // `setDecorFitsSystemWindows(false)` opts out of that, and on API
+        // 36 it still takes; Android 16 deprecated it and Android 17 is
+        // where it appears not to. At 35+ edge-to-edge is not opt-in, so
+        // the app is handed the real overlap without relying on a
+        // deprecated call. If the phone still reports `ime_bottom=0` with
+        // a nonzero `dispatches` in the Diagnostics pane, this reading was
+        // wrong and the `WindowInsetsAnimation.Callback` in
+        // `MainActivity` is the other half to look at.
+        targetSdk = 37
         versionCode = 1
         versionName = "1.0"
     }
 
+    // A release build must be signed, and the key is per machine rather than per repo -- same
+    // reasoning and the same key as `app/build-apk.sh` (the Compose app): it is what a phone
+    // recognises the app by, and a secret never lives in a checkout (the mount is shared with an
+    // untrusted VM). `build-apk.sh` generates this key once and points at it through the
+    // environment; without it a release build here is unsigned, which is fine for everything
+    // except installing.
+    def keystore = System.getenv("AI_APP_KEYSTORE")
+    signingConfigs {
+        if (keystore != null) {
+            release {
+                storeFile = file(keystore)
+                storePassword = System.getenv("AI_APP_KEYSTORE_PASSWORD")
+                keyAlias = "ai-app"
+                keyPassword = storePassword
+            }
+        }
+    }
+
     buildTypes {
         debug {
         }
+        // P0's iris half (docs/RUST.md's P0 box): the build a phone actually runs. The `.so`
+        // itself is built separately with `cargo ndk --release --features "transcript-screen
+        // force-gles bench"` straight into src/main/jniLibs/ (this crate's own Cargo.toml) --
+        // Gradle here only packages and signs whatever is already there, the same division as the
+        // debug/tabs-screen build this project started with. `applicationIdSuffix` keeps it
+        // installable beside a debug build of the tabs demo rather than replacing it.
+        release {
+            applicationIdSuffix ".bench"
+            if (keystore != null) {
+                signingConfig = signingConfigs.release
+            }
+        }
     }
 
     compileOptions {

iris/android-app/app/src/main/AndroidManifest.xml

@@ -1,6 +1,15 @@
 <?xml version="1.0" encoding="utf-8"?>
 <manifest xmlns:android="http://schemas.android.com/apk/res/android">
 
+    <!-- Only needed by the transcript-screen feature (RUST.md's I5),
+         which talks to a real ai-server; the plain tabs demo (I2/I4) makes
+         no network call and never noticed this was missing. Absent,
+         UreqTransport::new's connect failed with EPERM (Operation not
+         permitted), not the ECONNREFUSED/ENETUNREACH a firewall or a dead
+         server would give: a seccomp-level socket denial reads nothing
+         like a network problem, which is what made it worth a comment. -->
+    <uses-permission android:name="android.permission.INTERNET" />
+
     <application
         android:allowBackup="true"
         android:label="iris android-view demo"

iris/android-app/app/src/main/java/dev/iris/android/demo/IrisView.java

@@ -1,6 +1,17 @@
 package dev.iris.android.demo;
 
+import android.app.Activity;
+import android.content.ClipData;
+import android.content.ClipboardManager;
 import android.content.Context;
+import android.view.Gravity;
+import android.view.View;
+import android.view.ViewGroup;
+import android.widget.Button;
+import android.widget.FrameLayout;
+import android.widget.LinearLayout;
+import android.widget.ScrollView;
+import android.widget.TextView;
 
 import org.linebender.android.rustview.RustView;
 
@@ -16,7 +27,7 @@ public final class IrisView extends RustView {
     protected native long newViewPeer(Context context);
 
     native void applyWindowInsetsNative(
-            long peer, int left, int top, int right, int bottom, int imeBottom);
+            long peer, int left, int top, int right, int bottom, int imeBottom, int imeVisible);
 
     native void unregisterInsetsNative(long peer);
 
@@ -24,8 +35,9 @@ public final class IrisView extends RustView {
         super(context);
     }
 
-    void applyWindowInsets(int left, int top, int right, int bottom, int imeBottom) {
-        applyWindowInsetsNative(mViewPeer, left, top, right, bottom, imeBottom);
+    void applyWindowInsets(
+            int left, int top, int right, int bottom, int imeBottom, int imeVisible) {
+        applyWindowInsetsNative(mViewPeer, left, top, right, bottom, imeBottom, imeVisible);
     }
 
     @Override
@@ -33,4 +45,112 @@ public final class IrisView extends RustView {
         unregisterInsetsNative(mViewPeer);
         super.onDetachedFromWindow();
     }
+
+    /**
+     * Called from the Rust side (iris/src/android/view.rs's
+     * `show_renderer_error`) when `AndroidRenderer::new` fails instead of
+     * drawing -- an ordinary instance method rather than a `native` one,
+     * since this call is Rust reaching into Java rather than the other
+     * direction. Replaces the whole activity content with plain,
+     * selectable, scrollable text rather than leaving the last frame (or a
+     * blank surface) on screen with no way to report what happened:
+     * UI_RULES.md's "a failure is reported where it happened, and says
+     * what to do next." No dialog and no styling beyond what is needed to
+     * read and copy the text -- this path exists for exactly the crash it
+     * replaces, so it must not depend on anything that could itself fail
+     * to render.
+     */
+    void showRendererError(String report) {
+        Context context = getContext();
+        if (!(context instanceof Activity)) {
+            return;
+        }
+        Activity activity = (Activity) context;
+        TextView text = new TextView(activity);
+        text.setText(report);
+        text.setTextIsSelectable(true);
+        text.setGravity(Gravity.TOP | Gravity.START);
+        int pad = (int) (16 * activity.getResources().getDisplayMetrics().density);
+        text.setPadding(pad, pad, pad, pad);
+        ScrollView scroll = new ScrollView(activity);
+        scroll.addView(text);
+        activity.setContentView(scroll);
+    }
+
+    private static final String DIAGNOSTICS_OVERLAY_TAG = "iris-diagnostics-overlay";
+
+    /**
+     * The bench build's keyboard diagnostics capture
+     * (`bench_client.rs`'s `on_insets_changed` /
+     * `capture_keyboard_diagnostics`, via `bench_jni.rs`'s
+     * `PlatformHandle::show_diagnostics_overlay`): unlike
+     * `showRendererError` above, this adds a panel *over* this view
+     * (`MainActivity`'s `FrameLayout` still holds `IrisView` underneath,
+     * running) rather than replacing the activity's content, and gives it
+     * a Copy button and a Close that removes the panel -- so it draws
+     * (and can be read) whether or not iris itself is still putting
+     * anything on screen, without abandoning the session that produced
+     * it. Runs on the UI thread regardless of which thread calls it,
+     * since the call comes from a background task (a delayed capture
+     * after the keyboard opens), and touching the view tree off the UI
+     * thread is undefined.
+     */
+    void showDiagnosticsOverlay(String report) {
+        Context context = getContext();
+        if (!(context instanceof Activity)) {
+            return;
+        }
+        Activity activity = (Activity) context;
+        activity.runOnUiThread(() -> {
+            ViewGroup parent = (ViewGroup) getParent();
+            if (parent == null) {
+                return;
+            }
+            View existing = parent.findViewWithTag(DIAGNOSTICS_OVERLAY_TAG);
+            if (existing != null) {
+                parent.removeView(existing);
+            }
+
+            float density = activity.getResources().getDisplayMetrics().density;
+            int pad = (int) (16 * density);
+
+            LinearLayout overlay = new LinearLayout(activity);
+            overlay.setTag(DIAGNOSTICS_OVERLAY_TAG);
+            overlay.setOrientation(LinearLayout.VERTICAL);
+            overlay.setBackgroundColor(0xEE000000);
+            overlay.setPadding(pad, pad, pad, pad);
+
+            TextView text = new TextView(activity);
+            text.setText(report);
+            text.setTextIsSelectable(true);
+            text.setTextColor(0xFFFFFFFF);
+            ScrollView scroll = new ScrollView(activity);
+            scroll.addView(text);
+            overlay.addView(scroll, new LinearLayout.LayoutParams(
+                    LinearLayout.LayoutParams.MATCH_PARENT, 0, 1f));
+
+            LinearLayout buttonRow = new LinearLayout(activity);
+            buttonRow.setOrientation(LinearLayout.HORIZONTAL);
+            buttonRow.setPadding(0, pad, 0, 0);
+
+            Button copy = new Button(activity);
+            copy.setText("Copy");
+            copy.setOnClickListener(v -> {
+                ClipboardManager clipboard =
+                        (ClipboardManager) activity.getSystemService(Context.CLIPBOARD_SERVICE);
+                if (clipboard != null) {
+                    clipboard.setPrimaryClip(ClipData.newPlainText("iris diagnostics", report));
+                }
+            });
+            Button close = new Button(activity);
+            close.setText("Close");
+            close.setOnClickListener(v -> parent.removeView(overlay));
+            buttonRow.addView(copy);
+            buttonRow.addView(close);
+            overlay.addView(buttonRow);
+
+            parent.addView(overlay, new FrameLayout.LayoutParams(
+                    FrameLayout.LayoutParams.MATCH_PARENT, FrameLayout.LayoutParams.MATCH_PARENT));
+        });
+    }
 }