ai-app/docs/RUST.md

Moving the app to Rust

Working document for the question Iris asked on 2026-09-04: what are the options for switching the phone app to Rust, ideally pure Rust with one UI framework shared with a future winit-based desktop application, at full feature parity and without giving up anything native, performance especially. Constraints she set: no Dioxus and nothing that draws through a WebView; no UI DSL (which rules out Makepad and Slint); the result should stay lightweight; platform-specific pieces are fine to maintain; reimplementing a framework piece from scratch where it does not fit is fine; effort and elapsed time do not matter, long-term robustness does; this clone is where things get tried before anything is committed to ai-app. Her own library, iris, is the in-house framework to be built up for this, with Masonry as the yardstick it is measured against.

Decisions get a date and a reason here, the way PLAN.md does. A section describes something that has been tried only where it says so, with a date.

Keep this file current as you work

This file is the handoff, and it is meant to let a session be cleared. Write each result into it as you get it, not at the end: the box ticked or the reason it could not be, the measurement with its number, the decision with its date and what it rejected, and anything that cost time to find out. Then a session that has filled its context can be cleared and the next one can pick up from this file alone, which is much cheaper than carrying a long conversation or re-deriving what was already measured.

Two things that follow. Write for somebody who was not here — name the command, the file and the number rather than "the fix" or "the earlier run". And write the failures and the dead ends too: "Venus is blocked by the emulator, not by Mesa" and "the present mode was not the cause" are worth as much as the successes, because they are what stops the next session spending an afternoon on them again.

Where things stand (2026-09-06, orchestrator plan)

Written by the design agent on picking the branch up after a /clear, so the next session can resume from here. P0 is delivered and Iris's phone report v2 is in (docs/bench/iris-phone-v2-2026-09-06.md); P1 stays gated on her verdict, so this pass works the P0 defects and the pure prerequisites in this order. Each item is ticked here by the agent that closes it.

The transcript's edges (2026-09-07, e922b73 + d507ae4)

Iris's two screenshots of the top edge -- rows drawn over the header in one, a blank band in the other -- were three faults, and the rule that fixes all three is the one the IRIS_TODO entry asked for: a row is drawn if any part of it overlaps the list's own box, and nothing outside that box reaches the screen (List::intersects_viewport). Neither suspected cause was right, which is worth reading before trusting the next suspicion in this file: there was no visible-range test comparing a row's top against the viewport's, and 03c6be8's header duplicate is untouched and still open.

What was actually wrong: nothing clipped the list at all (and it could not be .masked(), because Painter::set_mask aborted whenever an ancestor had a mask and the list's own rows use one -- so masks nest now, Mask::parent, walked in the fragment stage); the layout walk drew every row between the anchor and the viewport, which after panning is however far you have panned, measured at 64 rows placed for a 2012px viewport; and the list could rest past its own first row, which is the blank band, and is also the first item of Iris's later report the same day. Details and the six layer-1 test names are in docs/IRIS_TODO.md's 2026-09-07 entry.

Three things this says about the rig, since the rig is new:

• Layer 1 found all of it, in seconds. The tests open the real screen over the real fixture under a bench-app-shaped header (iris/transcript-fixture/tests/top_edge.rs, cargo test -p transcript-fixture, ~5s), and each was confirmed to fail on its own subject and no other by breaking that subject on purpose. The emulator was not used, and the phone will only be asked to confirm.
• Layer 2 is where the clip is visible. iris/run-headless.sh phone --phone --replay transcript-fixture/touch/flick-120hz.touch --shot /tmp/p.png -- -p transcript-fixture, run with .masked() removed, draws the bottom row's text over the composer bar; with it, the same flick clips cleanly at the bar. The window has no header, so the top edge is the window edge there -- the header case is layer 1's.
• An assertion that reads the wrong thing hides the bug it is for. fling_toward_the_start_stops_at_the_first_row asserted the first row's top was >= -0.5 while that row sat 1398px below a 600px viewport with the screen blank: extents then held rows that were not on screen, so the read was satisfied by the failure. extents now holds only what is on screen -- which is what key_at always claimed of it -- asserted at the end of every draw, and the test checks both directions.

Phone logging, 2026-09-07 (rebuilt on Dev Updater's own tab)

The problem: Iris tests these builds on a phone with no adb, and Android forbids one app reading another's logcat, so a log::info! in the iris app could not reach her at all. What she asked for was Dev Updater, which she already reads.

The route, in one line: the app keeps its own bounded log ring and exposes it on the device through a ContentProvider; Dev Updater's phone app -- on the same phone -- reads that while the component's Runtime tab is open and forwards what is new to its own build machine, into that APK component's runtime log. No tunnel, no token, no second enrolment.

It is Dev Updater's contract, not iris's feature. Written down in dev-updater's README.md under "An app's own log", so any app that server delivers gets the tab by implementing it. The shape:

authority	<applicationId>.devlog
read permission	dev.updater.permission.READ_DEVLOG (android:readPermission)
lines?since=<seq>	held lines with seq >= since, ascending: seq, t_ms, level, target, message
status	one row: held, dropped, newest_seq

newest_seq is -1 for an empty ring and is what makes a restart visible: the ring is in memory, so a new process starts again at zero and a reader holding a cursor would otherwise skip everything since, silently. insert/update/delete throw. notifyChange is not implemented -- the ring is filled by a log::Log backend on whatever thread logged, and routing that to a provider means a callback through client-core for every platform, so Dev Updater polls (about a second, only while the tab is open) and the contract says so.

What exists now:

• client_core::log_ring -- LogRing (2000 lines / 256 KiB, whichever bites first, with dropped reported rather than inferred), since(seq), newest_seq(), RingLogger (a log::Log backend that records and forwards to the platform's own logger), and install_process_logger. Reading does not consume, so the report and the provider are two readers of one ring.
• iris/android-app/src/devlog.rs and app/src/main/java/dev/iris/android/demo/DevLogProvider.java -- the platform glue only (the sharing rule): a String[] across JNI, a MatrixCursor on the Java side, and nativeReady telling Rust the authority the provider actually registered under.
• iris/android-app/src/app_log.rs -- android_logger as the logger to forward to, and the Diagnostics pane's two lines: how many lines are held, and devlog provider: content://<authority> (or "declared, not created yet", since Android creates a provider lazily). Named from what the provider registered rather than composed from the package here, so a screenshot of that pane is evidence the contract is live.
• On the Dev Updater side (its own repo, commit 013d711): the tab for every component rather than only a server's, DevLog.kt's reader and cursor, POST /apps/{key}/components/{name}/runtime-log, and $XDG_DATA_HOME/dev-updater/devlogs/<key>-<component>.log with one rotation at 4 MiB.

Deleted with it, so there is one mechanism: client-core's log_upload, POST /client-log on ai-server, the AI_APP_LOG_* baking in iris/android-app/build.rs (which left that file with nothing to do, so it is gone), and the uploader fields on both Android clients.

How to use it. Build and install the bench APK; open Dev Updater -> the ai-app project -> the app component's log button -> the Runtime tab. Nothing to configure: the tab finds the provider from the package the component installs. A component whose package exposes none says so in as many words, which is a different sentence from an empty log.

Verified 2026-09-07 end to end -- see "Verified" below.

Two rig traps this cost an hour to find, both in build-apk.sh, both still there. Written down rather than fixed because fixing them belongs with whoever next touches that script:

1. Gradle's merged-native-libs cache survives rm -rf jniLibs. The script removes app/src/main/jniLibs before each build (its own comment says why), but Gradle's mergeReleaseNativeLibs is up to date against its cached inputs, so a build that switches ABI packages the previous ABI. A --abi x86_64 release APK contained lib/arm64-v8a/libmain.so, installed fine, and aborted at startup with Could not get adapter!: NotFound { active_backends: VULKAN } under libndk_translation -- which reads exactly like the phone's own Vulkan problem and is nothing of the kind. rm -rf app/build/intermediates before the build is the workaround; check with python3 -c "import zipfile; print([i.filename for i in zipfile.ZipFile('...apk').infolist() if i.filename.endswith('.so')])".
2. The debug bench APK is 648 MB and will not install (INSTALL_PARSE_FAILED_NOT_APK): the debug libmain.so is 325 MB. Use release on the emulator for this app, notwithstanding the general rule that the emulator stays on debug -- there is nothing to measure here, and the debug build cannot be installed at all.

Also: the bench APK's package is dev.iris.android.demo.bench, not dev.iris.android.demo. An older non-bench build left installed answers to the second name, runs, looks right, and reports whatever it was built with -- which is how "log upload: this build has no server configured" came from a build that had one. The devlog authority carries the applicationId for exactly that reason: the two packages each get their own and neither can read the other's log.

iris::input/iris::frame diagnostics, 2026-09-07

Iris asked for a second button (or the same "Copy report" made to say more): "add another button to copy input event info so that I can do some stuff manually and then send the event log to you so you know what events the code is actually receiving. You may also want to instrument a lot of the code with timings so I can give you time reports too through the same button." This is that, built on the log ring rather than a second mechanism.

What it writes. iris::sense::log_input_event -- one line per platform pointer sample, from Android's on_touch_event (once per real MotionEvent, historical samples inline), the winit backend (once per pointer WindowEvent, no batching), and harness::Harness::touch (once per TouchScript line, also unbatched, which is what makes a harness replay round-trip exactly):

iris input: action=down x=540.0 y=1000.0 t=0ms history=0
iris input: action=move x=540.0 y=1196.0 t=16ms history=3 4:540.0,1040.0 8:540.0,1086.0 12:540.0,1138.0

iris::diagnostics::log_frame -- one line per frame, called from each backend's own frame function after the draw (or, on the harness, where a draw would be):

iris frame: n=42 now=701ms since_input=12ms layout=8.3µs draw=1.1ms redraw=Updates primitives=384 animating=true

n is UiRenderState::frame_number (counts every call to update, including one that redrew nothing); now is milliseconds since that render state's own construction (UiRenderState::epoch, set the same way Harness::base is, so it lines up with a harness's own t_ms); since_input is how stale the input driving this frame was, from the last sample SensorUi::run_sensors saw; layout/draw are Instant pairs around UiRenderState::update and the platform's own submit+present; redraw is RedrawKind::{None,All,Updates}; primitives is UiRenderState::active_primitive_count (everything currently on screen, not a per-frame delta -- take_counters' draw/rewrite/shape counts are that, and iris::frame does not duplicate them).

Reading a report: with tracing off (the default) neither target appears at all. With it on, the two interleave in arrival order, so a flick's shape reads as a run of iris input: lines followed by the frames they drove, and a release still shows the existing iris drag release:/ iris drag release samples: lines from sense.rs -- those were not duplicated, only (the samples: one) brought under the same gate.

Replaying one: iris/benches/report_to_touch.py < report.txt > replay.touch reads every iris input: action=... line (ignoring everything else in the report, prefix-agnostic -- it works on a bare message or a full ring line with its HH:MM:SS.mmm LEVEL target: header) and expands each event's inline historical samples into their own move lines first, oldest first, exactly as Android delivers and replays them. The output is an ordinary .touch file: Harness::replay(&TouchScript::parse(&text)?) plays it back at layer 1, or point run-headless.sh phone --phone --replay at it for layer 2. Verified round-trip, both directions: a harness replay of flick-120hz.touch with tracing on produces exactly six iris::input lines, and piping them through the script and re-parsing reproduces the same six t_ms action x y samples (iris/transcript-fixture/tests/ input_log_roundtrip.rs).

The toggle: iris::diagnostics::set_trace(bool), off by default. Not log::log_enabled!/log::set_max_level: the app already installs its logger at LevelFilter::Debug (iris/android-app/src/lib.rs's JNI_OnLoad), and client_core::log_ring::RingLogger::enabled is unconditionally true by design ("the ring wants everything"), so a log::Level::Debug line reaches the ring regardless of what this instrument would prefer -- the gate has to be a crate-level flag, checked before log::debug! is even reached, and that is what trace_enabled() is. The switch is the bench header's fourth control, beside Run benchmark / Copy report / Diagnostics: it reads Trace off or Trace on, because a toggle whose own appearance never changes is a button that looks like it did nothing. Its accessibility label stays the fixed Trace input and frames -- that is what run-bench.sh and ui-trace --do "tap '...'" find it by, and a control that renames itself when pressed is one no script can find twice. Pressing it rebuilds the header (the same bench_controls path on_insets_changed already uses) and shows the diagnostics pane, so the state is on screen at the moment of the press.

Why default off, and why the ring's size is the actual constraint: the ring is 2000 lines / 256 KiB (client_core::log_ring::DEFAULT_MAX_LINES/DEFAULT_MAX_BYTES); a 120Hz session logging a line per touch sample and a line per frame fills that in seconds, so a caller turns tracing on only for the length of whatever is being investigated, not for a whole session. This is also why the report should say at its top whether tracing was on -- a caller reading iris::diagnostics::trace_enabled() when building the report can print that. Both reports do: bench_client::trace_line is the one wording, and it takes the flag read at the start of the run as well as at the end, so a switch flipped half way through is reported as exactly that rather than as a confident "on" about a log covering half the run. Three states, because that third one happens -- the switch is on screen while a benchmark runs.

D1 from docs/REVIEW-2026-09-07.md: the review found that this gate existed (as iris/src/diagnostics.rs, uncommitted at the time) but four older per-frame debug! lines were not wired to it -- android/view.rs's two render(): lines, widget/list.rs's iris fling tick:, and widget/text/mod.rs's iris text render: -- each unconditional at Debug, and between them enough to fill the ring in under ten seconds at 120Hz before Copy report ever saw anything else. All four (and sense.rs's iris drag release samples:, which is lower-volume but the same shape) are now behind iris::diagnostics::trace_enabled(), moved onto the iris::frame/ iris::input targets where each belongs. iris::sense's iris drag release: (info level, one per gesture, low volume, and the line that already answered "why didn't that flick fling" from Iris's phone) is unchanged and ungated on purpose -- it is exactly the kind of always- useful summary line the ring is for.

Verification: iris/transcript-fixture/tests/ input_log_roundtrip.rs's one test replays flick-120hz.touch through a real capturing log::Log twice -- tracing off, then on -- and asserts (a) off leaves zero Debug-level lines from the whole replay (cargo test -p transcript-fixture catches a regression here immediately, not just this one), (b) on produces exactly one iris::input line per replayed sample and at least one iris::frame line with a non-zero layout=, and (c) the round trip through report_to_touch.py reproduces the exact script. A throwaway (not committed) 3000-frame timing loop through the harness with tracing off vs on, idle after the opening layout, measured 2.97µs/frame off against 3.40µs/frame on with no logger even installed -- the Instant::now() pairs and the trace_enabled() atomic loads that stay live either way. That is a layer-1 proxy, not the phone's own bench (no GPU work happens there at all), but it bounds the added cost at a few hundred nanoseconds against a 60Hz budget of 16,600 -- three orders of magnitude below where it could be seen.

APK size (2026-09-07)

Iris's question: the iris bench APK is about double the Compose bench APK (20.6 MB vs 10.1 MB, both release). Measured before this pass: 18,088 KiB of lib/arm64-v8a/libmain.so, stored uncompressed (extractNativeLibs=false), plus 2 MB of dex; the Compose APK's dex is 25 MB raw, compressed to 9 MB in the APK. iris/android-app/Cargo.toml's [profile.release] set only panic = "abort" -- no lto, no codegen-units, no strip, default opt-level. All numbers below are arm64-v8a release, built with ./build-apk.sh release (this checkout, nice -n 10, no CARGO_TARGET_DIR override, reusing the incremental target/), and are the raw file sizes (ls -la), not what du would round to.

profile.release	APK bytes	libmain.so bytes	delta vs previous
panic="abort" only (baseline)	20,678,956	18,546,488	--
+ strip = true	16,435,156	14,302,688	-4,243,800
+ lto = "fat"	15,751,212	13,618,744	-683,944
+ codegen-units = 1	15,185,204	13,052,736	-566,008
+ opt-level = "s"	13,326,076	11,193,608	-1,859,128
+ opt-level = "z" (not adopted, see below)	12,507,276	10,374,808	-818,800
+ platform fonts, no bundled Noto (2026-09-07)	9,577,940	7,445,472	-3,748,136

Adopted: strip = true, lto = "fat", codegen-units = 1, opt-level = "s". Baseline to final: libmain.so 18,546,488 -> 11,193,608 bytes (-39.7%), APK 20,678,956 -> 13,326,076 bytes (-35.5%).

opt-level = "z" was measured but not adopted. It is smaller still -- another 818,800 bytes off libmain.so (7.9 MiB total vs s's 8.7 MiB) -- but z trims more aggressively than s in ways that can cost frame time (fewer inlines, more size-motivated codegen choices, per rustc's own docs), and this pass did not have an iris-side frame-time benchmark run against it (the app's own run-bench.sh/render report was not exercised here, per this task's scope, and this checkout has no emulator currently up). Trading an unmeasured runtime cost for ~700 KB more off the download is not a call to make blind, so s is what shipped, and z is left as something to try only alongside a transcript-bench.sh/stream-bench.sh-equivalent run for iris to confirm it does not regress.

Not stripped (baseline) had a live .symtab (llvm-readelf -S): section 25, SYMTAB, 0x17ed40 bytes (~1.49 MiB) covering 65,223 raw symbols. strip = true removes it at build time -- notably, stripReleaseDebugSymbols (AGP's own strip task) had already logged "Unable to strip the following libraries, packaging them as they are: libmain.so" on the baseline, so Cargo's own strip is also the fix for that.

Baseline section sizes (llvm-readelf -S, before any profile change):

section	bytes
.text	6,463,032
.rodata	6,548,192
.eh_frame	721,872
.data.rel.ro	441,328
.gcc_except_table	13,652
.symtab	1,563,456

No bloaty on this machine (which bloaty empty); used llvm-nm -S --size-sort -C on the baseline (unstripped) .so, summed by the symbol's leading crate/module name. 7.02 MB of the 18.5 MB .so carries a name at all (the rest is .rodata blobs -- embedded data, padding, relocations -- that never get a symbol). Top 8 by that accounting:

crate	bytes (named symbols only)
naga	1,100,099
core (std)	698,862
wgpu_core	599,720
harfrust	372,694
alloc	347,324
read_fonts	326,824
wgpu_hal	299,648
skrifa	291,332

Also notable further down: hashbrown 244,712, jni 199,660, std 199,126, serde 168,256, zeno 157,320, pulldown_cmark 116,568, iris_core 95,112, parley 78,816, iris 76,616, swash 72,940, serde_json 72,312, fontique 45,464.

The other ~11.5 MB of .rodata/unnamed data is mostly the embedded fonts: iris/core/src/primitive/text.rs include_bytes!s six Noto Sans TTFs (iris/core/assets/fonts/) -- Regular/Bold/Italic/BoldItalic for Noto Sans plus Regular/Bold for Noto Sans Mono -- totalling 3.6 MB of raw font data (du -ch iris/core/assets/fonts/*.ttf). That is real render data, not something to strip: unlike the Compose app's Nerd Fonts icon subset (app/build-icon-font.sh, which subsets because the app only ever draws ~100 fixed glyphs), iris's Noto Sans embedding backs arbitrary text in a chat transcript, so a subset would have to be a Unicode-coverage subset (Latin/Latin-Extended/common punctuation, dropping CJK/Cyrillic/etc) rather than a fixed-codepoint one -- a real behaviour change (text in a language outside the subset would fall back to tofu or a missing glyph) and out of scope for a size-only pass. Left as a follow-up, flagged for Iris: subsetting would plausibly save 1-2 MB but changes what scripts render correctly, which is a product decision. Superseded 2026-09-07: Iris decided to remove the embedding outright rather than subset it -- see "Platform fonts (2026-09-07)" below.

Platform fonts (2026-09-07)

Iris's verdict on the open question above: "remove the font for now; just match what compose does." The Compose app takes its body text from FontFamily.Default (platform Roboto on Android) and its code/tool-output text from FontFamily.Monospace, and ships no text font of its own -- only its committed Nerd Fonts icon subset (app/build-icon-font.sh) for ~100 fixed glyphs, a different case (a small, known, closed set of codepoints, unlike arbitrary transcript text). iris had no equivalent icon font to keep; it draws no icons through a font today, so there was nothing parallel to preserve.

What changed: TextData::register_bundled_fonts and the six include_bytes! Noto Sans/Noto Sans Mono constants are gone from iris/core/src/primitive/text.rs, along with the .ttfs themselves and their OFL.txt (iris/core/assets/fonts/, now removed -- nothing else in the tree referenced the licence file). TextData::default now does nothing but FontContext::new(), which was already discovering the platform's fonts underneath the bundled ones -- fontique 0.11.1's CollectionOptions::system_fonts defaults to true, and both platforms this crate ships on have a real backend behind it: backend/fontconfig.rs on Linux (this VM's desktop has a full Noto install, confirmed with fc-match sans-serif/fc-match monospace), backend/android.rs on Android (parses /system/fonts and /system/etc/fonts.xml, mapping SansSerif/SystemUi to ["Roboto Flex", "Roboto", "Noto Sans"] and Monospace to ["monospace"] -- see the fallback finding below for why that last one does not actually resolve on this fontique version). So removing the bundled registration did not need a replacement call; the platform path was already live, just shadowed.

Fallback, and the unknown-glyph state (UI_RULES: design it, don't let it default to blank). parley's shaper sets both an explicit family list and a script/locale-keyed fallback chain per run (parley-0.11.1/src/shape/mod.rs's query.set_families/ query.set_fallbacks), so a codepoint the resolved family lacks still gets a query against fontique's fallback map before giving up. Checked with a throwaway example (iris/examples/font_check.rs, deleted after use -- not part of the crate) shaping "🎉🔥▸▾▲你好" headless on the desktop at run-headless.sh: the emoji and CJK characters drew as visible tofu boxes (the platform's own missing-glyph box, not blank space), and the chevron marks (U+25B8/25BE/25B4, the ones transcript-ui/src/tool.rs's CLOSED_MARK/OPEN_MARK/UP_MARK draw) shaped as real triangles. So the failure mode this crate now depends on is "the platform's own tofu," which is the correct unknown-glyph state per UI_RULES, not "nothing drawn." tool.rs's doc comment on those marks is updated to say this is now a bet on the platform's coverage rather than a checked fact about a bundled cmap.

One real gap, found on this checkout's emulator, not the desktop, closed 2026-09-07: fonts: 208 families found, default=Some("Roboto Flex") mono=None in the startup log (FontDiagnostics, read via adb logcat after installing the force-gles debug build -- the emulator's default Vulkan backend has no adapter here, a pre-existing, documented condition unrelated to this change, and aborts with Could not get adapter! without that feature). mono=None means fontique's Android backend never resolves the Monospace generic family at all on this system image, and it is two bugs stacked rather than one: reading fontique-0.11.1/src/backend/android.rs, DEFAULT_GENERIC_FAMILIES's ["monospace"] is looked up against name_map before the fonts.xml parse that adds the name runs, and even after parsing, this AVD's /system/etc/fonts.xml (and AOSP's/GrapheneOS's, same file format) names it with a <family name="monospace"><font ...>DroidSansMono.ttf</font></family> element rather than an <alias> -- whose <font> children that same parser's "family" match arm never reads (a TODO left in place), so the name gets registered with no font data behind it. family_by_name("monospace") therefore also comes up empty, on every Android device this fontique version runs on, not just this AVD. Checked against linebender/parley's main branch on GitHub the same day: neither bug is fixed there either, so there is no newer release to bump to. The visible effect was not blank text -- Family::Monospace's explicit-family list came up empty, but the script-based fallback chain (independent of the generic-family list) still resolved a real font, the same one SansSerif gets -- so code blocks and the tool-card chevrons rendered, just without a genuinely monospaced face, while Compose's FontFamily.Monospace (resolved through Android's own Typeface.MONOSPACE constant, a path fontique does not use) was unaffected.

Fixed in iris/core/src/primitive/text.rs's patch_android_monospace (#[cfg(target_os = "android")], called from TextData::default right after FontContext::new()): rather than pinning an OEM-specific name like "Droid Sans Mono" (the fragility this gap was originally left open over), it reads /system/etc/fonts.xml itself -- a plain substring search, not a new XML-parser dependency, for the one well-known AOSP file fontique already parses with a real one -- for the filename the "monospace" family declares, then searches fontique's own actually scanned families (the ones with real font data, from /system/fonts) for whichever one owns a font file with that name, and registers that family as the Monospace generic itself. This is the same authority Compose's Typeface.MONOSPACE resolves through, and it degrades safely to a no-op if fonts.xml is missing (a headless test) or nothing matches (a device naming it some other way) -- the pre-existing sans fallback, not a panic. Verified on this checkout's emulator: mono=Some("Droid Sans Mono") in the startup log, resolved ... mono=Some("Droid Sans Mono"), and a screenshot of the bench-fixture transcript showing the code block and tool-card value text in a visibly monospaced face next to sans body/heading text. The desktop's fontconfig backend was never affected (confirmed unchanged: ./run-headless.sh phone --phone --shot still shows monospaced code next to sans body text) -- the patch is Android-only and a no-op everywhere else.

Verified: cargo test -p transcript-fixture (6 tests, all headless layers) and cargo clippy -p iris-core --all-targets both clean; ./run-headless.sh phone --phone --shot (the bench-fixture transcript, bold/italic/monospace code fences all shaping correctly) and ./run-headless.sh tabs (the desktop example with several distinct text styles, per this task's "negative case" check) both look right by eye; the emulator's own bench run (run-bench.sh, force-gles debug build) completed a full fling/stream/type/keyboard cycle with text visibly drawn throughout and no crash. .so size: see the APK size table's final row, 11,193,608 -> 7,445,472 bytes, a 3,748,136-byte drop matching the 3.6 MB estimate almost exactly.

naga/wgpu backend features: investigated, not trimmed, because the trim would not change the binary. iris/core/Cargo.toml and iris/Cargo.toml depend on wgpu = "28.0.0" with default features, which via wgpu's own defaults (dx12, metal, gles, vulkan, wgsl, webgpu) forward naga/hlsl-out, naga/msl-out, naga/glsl-out, naga/spv-out, naga/wgsl-in, naga/wgsl-out -- Cargo feature unification is not per-target, so all of those are nominally "on" for the Android build too, not just the ones Android actually uses (glsl-out for GLES, spv-out for Vulkan). But wgpu-hal's own build.rs (cfg_aliases!) gates the modules themselves on the real target: dx12: target_os = "windows" AND feature = "dx12", metal: target_vendor = "apple" AND feature = "metal" (wgpu-hal-28.0.0/build.rs, wgpu-hal-28.0.0/src/lib.rs's #[cfg(dx12)] pub mod dx12; etc). So on aarch64-linux-android the dx12/metal modules never compile, nothing calls into naga::back::hlsl or naga::back::msl, and the linker's normal dead-code elimination already drops them: grep -c "naga::back::hlsl\|naga::back::msl\|naga::front::spv\|naga::front::glsl" /tmp/nm_size.txt on the baseline (no LTO yet) .so returns 0 -- none of that code reached the linked binary in the first place. regex (pulled in transitively by env_filter, which android_logger uses for RUST_LOG-style filtering) is in the same position: present in Cargo.lock but only a handful of small generic-drop symbols in nm, not a real contributor. Neither is worth a Cargo-level feature trim (which would also need a per-target dependency table to avoid stripping dx12/metal off the desktop build, adding real complexity for a change that measures as zero). No emulator use was needed for this finding since no feature flag changed; the earlier per-step size measurements (strip/LTO/cgu/opt-level) were likewise not re-verified on the emulator, since this task's brief scoped emulator use to the naga-trim step specifically, and that step's answer was "don't."

tabs-ui/tabs-screen: also investigated, also already dead. build-apk.sh's default features are "transcript-screen bench", passed without --no-default-features, so the crate's own default = ["tabs-screen"] (iris/android-app/Cargo.toml) is also on for every build this script produces, including the bench APK. src/lib.rs's doc comment already says the three screens are mutually exclusive at runtime (ActiveClient gives bench priority over transcript-screen, which takes priority over the default tabs-screen), and checking the actual #[cfg(...)] gates confirms it is mutually exclusive at compile time too: the Client struct and its one call to tabs_ui::build are behind #[cfg(not(feature = "transcript-screen"))], which is false whenever transcript-screen is on, so that code does not even get generated, let alone linked. llvm-nm -C on both the baseline and the final .so confirm it: grep -ci "tabs_ui\|sungals" is 0 in both. So there is nothing to trim here either -- tabs-ui and its sungals.png (8.9 KB) never reach the linked binary in a transcript-screen/bench build, regardless of the feature being nominally "on" in Cargo.toml.

Comparison Iris asked for, honestly: most of the remaining ~13.3 MB vs Compose's 10.1 MB is not a build-settings gap, it is what each app links. Compose's APK carries ~9 MB of compressed dex and links Android's own platform renderer, text shaper (HarfBuzz/Minikin) and font files from the system image at zero cost to the APK -- none of that is bytes Compose ships. Iris ships its own copy of all of that: wgpu+naga+wgpu_hal (a software/hardware-portable GPU backend and shader cross-compiler, roughly 2 MB of named symbols alone), harfrust+read_fonts+skrifa+swash+ parley+fontique (a full third-party font-loading/shaping/rasterizing pipeline, another ~1.2 MB of named symbols), and 3.6 MB of embedded font data because it cannot borrow the platform's fonts the way Compose does. Build settings (this pass) closed real ground -- 39.7% off libmain.so -- but did not remove any of those linked systems, because removing them would mean iris stops being a self-contained native renderer, which is the whole point of the port (AGENTS.md's "no Dioxus and nothing that draws through a WebView", no-dioxus-or-webview-ui-true-native-only memory). Install size (what dumpsys package/du on the installed lib/arm64-v8a/ directory would show) was not separately measured this pass: with extractNativeLibs=false the .so is mapped directly out of the APK rather than copied onto disk a second time, so install size tracks the APK size closely for the native library and is not a second, larger number the way it would be under the old extractNativeLibs=true default -- checking this precisely needs the emulator, which this task scoped to the naga-trim verification only.

Committed: iris/android-app/Cargo.toml's [profile.release] now reads panic = "abort", strip = true, lto = "fat", codegen-units = 1, opt-level = "s", with a comment naming the measured savings.

Queue, 2026-09-07 (orchestrator)

In order; two builders at a time. Each is ticked here by the agent that closes it.

• Test rig, layers 1 and 2 ("Three test layers" below), landed 2026-09-07.

• Fling parity with Compose, and the phone's keyboard push-up, with insets shown in the diagnostics overlay. Every part is ticked in IRIS_TODO's 2026-09-07 entries: the keyboard half in the night entry, the estimator in the "later" one (Lsq2, not Impulse -- see the box below), and the catch in b87f5a5. Not yet confirmed from the phone, which is what would close it for Iris rather than for us.

• Orchestrator note, 2026-09-07 late: the tree was found holding a non-compiling diff from two killed agents (catch-a-fling in sense.rs/selection.rs; shaped masks in the render files). One opus agent owns splitting and landing both (catch-a-fling first); one sonnet agent owns report hygiene and the bench header in bench_client.rs and client-core's log ring. If both boxes below are still open and nothing is running, that work was cut off again. 2026-09-08: it was cut off again -- the sonnet agent's two boxes had landed (7485d78, b8ea723), catch-a-fling had landed unticked (b87f5a5), and the shaped-mask diff was left uncommitted in the tree with one failing test. Both are closed below, by one session working inline rather than by agents. Nothing is running now.

• Rows at the transcript's top edge: culled too early in one state, drawn through the header in the other (docs/IRIS_TODO.md, 2026-09-07). Done 2026-09-07, e922b73 + d507ae4; the root causes and the test names are in that IRIS_TODO entry, and the short version is below.

• Phone logging through Dev Updater -- done 2026-09-07, see "Phone logging" above and docs/DECISIONS.md's entry of that date.

• APK size: release profile tuned (42af780), -35% APK, -40% .so; see "APK size (2026-09-07)". Iris's verdict, 2026-09-07: "remove the font for now; just match what compose does." Done same day -- the six bundled Noto Sans TTFs are gone, text now loads from fontique's platform collection (FontContext::new()'s default CollectionOptions::system_fonts), and .so dropped by 3,748,136 bytes (11,193,608 -> 7,445,472), matching the 3.6 MB estimate almost exactly. See "APK size" table's final row and "Platform fonts (2026-09-07)" below for the fallback behaviour and one real gap it surfaced: this fontique version's Android backend never resolves the Monospace generic family at all (mono=None in the startup diagnostic, measured on this checkout's emulator) -- code/tool-card text still rendered (the script fallback chain still landed on a real face, never blank), just not in a genuinely monospaced one. Compose did not have this gap; it resolves FontFamily.Monospace through Android's own Typeface constant rather than through fontique. Closed same day -- see "Platform fonts (2026-09-07)"'s "Fixed" paragraph: TextData::patch_android_monospace resolves the platform's own fonts.xml monospace declaration against fontique's actually-scanned families, Android-only, verified mono=Some("Droid Sans Mono") on this checkout's emulator.

• Scroll clamped at both ends (e922b73, List::clamp_to_content) and Compose's velocity estimator (docs/IRIS_TODO.md, 2026-09-07 later). Ticked 2026-09-08 against those entries, which were already [x] while this box was not. Two things this box's own wording had wrong, both corrected there by reading Compose's sources: the touch path is Lsq2 over absolute positions, not Strategy.Impulse (Impulse is the mouse-wheel/trackpad path), and there is no minimum fling velocity on it -- minimumFlingVelocity belongs to NestedScrollInteropConnection. So iris ports Lsq2, caps at 8000dp/s and floors at 1px/s, and has no 50dp/s threshold Compose does not have.

• APK runtime logs in Dev Updater (Iris, 2026-09-07: "please add android / apk runtime log support to dev updater"). Done 2026-09-07 -- dev-updater 013d711, and this repo's provider half; "Phone logging" above is the account, docs/DECISIONS.md the decision. What was designed and what was built agree except in one place: the provider does not call notifyChange (the reason is in both), and the tab is drawn for every component rather than only where a provider resolves, since its absence would be the one thing that could not say which of the several reasons there was nothing to read. The design as written: Supersedes the ai-server POST /client-log route, which becomes the second mechanism and is deleted once this works (log_upload.rs, app_log.rs's upload half, the route). Design: Android forbids reading another app's logcat, so the app carries its own log (client_core::log_ring, kept) and exposes it on-device through a ContentProvider that Dev Updater's phone app reads -- no tunnel, no token, no second enrolment, because the two apps are on the same phone. Authority <applicationId>.devlog, one table lines(seq, t_ms, level, target, message) plus a dropped count, queried with since=<seq> so a poll is incremental. Dev Updater's phone app: for an APK component whose installed package resolves that authority (PackageManager), the component gets a Runtime tab like a service's; it polls the provider while the tab is open and forwards new lines to its host server's existing per-component runtime-log store, so history survives the phone and the same tab code renders it. Read access guarded by a permission Dev Updater defines (dev.updater.permission.READ_DEVLOG, protection normal; signature level is not available because the two apps are signed with different locally generated keys -- state that trade-off in DECISIONS.md). The provider is Java in android-app reading the ring over JNI (platform glue, allowed by the sharing rule); the Compose app can implement the same contract later so both apps get the tab. Rejected: Dev Updater handing its server token to the app it installed (leaks the token into every managed app); the app posting to ai-server (needs its own enrolment first and puts the phone's logs in the wrong component).

• Iris app enrolment (done 2026-09-07): the iris Android app is told which ai-server to talk to by an aiapp://enroll link, exactly as the Compose app and desktop-app are, instead of having it compiled in. MainActivity registers the VIEW intent and hands the URI and the app's private files directory to Rust (src/enrollment.rs is the intent plumbing and nothing else); the parsing, the file and its 0600 mode are client_core::config's EnrolledServer/EnrollmentStore, shared with the desktop app. build.rs's AI_APP_TRANSCRIPT_HOST/_PORT/_TOKEN are gone, and with them a token in a built artifact. The CA travels with the link (&ca=, base64url of the DER) -- docs/DECISIONS.md, 2026-09-07, has the decision, the two rejected alternatives and what the longer link costs a QR code. That is what makes an APK cross-compiled here work against the server on the host. Diagnostics says which of three things is true -- enrolled: host:port, not enrolled -- open the enrol link from Dev Updater, or enrolment unreadable: ... -- because "could not find out" wants a different action from "nothing there yet". Dev Updater needed no change: its Enroll button already opens the minted link with ACTION_VIEW, and Android offers the chooser between this app and the Compose one. The log upload was deliberately left out of it: Dev Updater grew an on-device runtime-log reader instead (Iris, 2026-09-07), so log_upload, POST /client-log and the AI_APP_LOG_* baking went out whole rather than being rewired first -- done the same day, "Phone logging" above. build.rs had nothing left to do and is gone with them.

• [~] iris/android-app/build-apk.sh. First half done 2026-09-08 (4f6ec3a): the script now removes merged_native_libs, stripped_native_libs and merged_jni_libs alongside app/src/main/jniLibs, since mergeReleaseNativeLibs is up to date against its cached inputs and rm -rf jniLibs does not reach them. Scoped to those three rather than all of app/build, so an ABI change costs the native merge and not the whole Gradle build. Verified on the case that produced it -- an x86_64 lib left in the checkout from emulator work, then ./build-apk.sh release --abi arm64-v8a, whose APK's only .so is lib/arm64-v8a/libmain.so. Still open: the debug bench APK is 648 MB and will not install (INSTALL_PARSE_FAILED_NOT_APK), because the debug libmain.so is 325 MB; use release on the emulator until that is fixed.

• Input-event and timing instrumentation into the log ring, copied by the report button. After the logging route lands (same ring).

• Catch-a-fling (done 2026-09-07, b87f5a5; verified and ticked 2026-09-08). A down on a moving list ends the fling on that sample and enters Panning with no DRAG_SLOP wait, which is Compose's scrollable(startDragImmediately = isScrollInProgress); a catch released without moving is Released(None) rather than a Tapped, which is also what Compose delivers. DragArbiter::press_start takes a PressState -- what the target looked like when the press landed (already_selected, scrolling) -- rather than reading the list again later, because by then the fling it is asking about has already been cancelled. The defect layer 1 found on the way: one touch-down reaches every sensor under the finger, so a transcript row's block and the tool row containing it deliver the same PressStart twice, and re-reading PressState on the second delivery turned every catch back into an ordinary slop-waiting press. Tests: transcript-fixture/tests/catch_a_fling.rs (the recorded 120Hz flick, 150ms of fling, a down and three 2px moves -- the content tracks the finger sample for sample, failing at the parent commit with "the content 0.0px"), plus the_same_small_drag_on_a_settled_list_moves_ nothing, which is the half the change had no reason to touch: 6px is inside DRAG_SLOP, so pinning the content on every press would pass the first test and quietly take the slop away from every ordinary one.

• Report hygiene (done 2026-09-07). Ring takes Debug only from iris/client_core targets, Copy report always copies and trims the log. client_core::log_ring::ring_accepts is the one filter (Info+ from anywhere; Debug/Trace only from this app's own targets, and only while iris::diagnostics::trace_enabled() says tracing is on) -- RingLogger takes that as a plain fn() -> bool rather than depending on iris directly, since client-core sits below it; app_log.rs wires iris::diagnostics::trace_enabled through at install. Fixed the 1339-held/4050-dropped flood from naga::front/wgpu_core/jni logging at Debug unconditionally. bench_client.rs's copy_report no longer declines when nothing has run: with no benchmark yet it copies the diagnostics pane's own text instead, with a first line saying so, and always appends LogRing::tail_text(COPY_REPORT_TAIL_LINES = 150) (a first line naming how many older lines were left out) rather than the whole ring. client-core's tests cover the filter and the trim; the copy path was checked on the emulator (immediate Copy report tap with no prior button press now logs "copied to clipboard").

• Bench app header (done 2026-09-07). Four controls no longer fit at 1080px at HEADER_TEXT = 18, and a previous agent had shrunk it to 13 to make room -- UI_RULES: never shrink text to fit. Restored to 18 and split bench_controls into two rows (Dir::DOWN of two Dir::RIGHT pairs: run+copy, then diagnostics+trace), doubling the header's own height (HEADER_ROW_HEIGHT_DP) rather than the outer layout's reserved space, since top_bar sizes to its own content. Checked on the emulator: ui-trace show ... --field box confirms two clean rows with no overlap, and a screenshot shows the restored size reading clearly.

• Bench app crash-loops on this checkout's emulator (done 2026-09-07). Not the surface lifecycle at all, and not "once backgrounded" -- a build with the default features (no force-gles) never got a first frame. AndroidRenderer::new (iris/src/android/render.rs) asked wgpu for Backends::PRIMARY, which does not contain GL, and this emulator advertises a Vulkan ICD with no adapter behind it: RequestAdapterError::NotFound { active_backends: VULKAN, no_adapter_backends: VULKAN, supported_backends: VULKAN | GL }, .expected, so SIGABRT, so the launcher restarts it -- the loop. A force-gles build was never affected, which is why the crash looked like it belonged to whatever else was going on. Root cause: iris refused a device whose only usable adapter is a GLES one.

  Fix, two halves. (1) AndroidRenderer::new now probes for a PRIMARY adapter with an instance that never touches the window and rebuilds the instance on Backends::GL when there is none. The probe is surface-free deliberately: an Android window can be connected to one graphics API only, so a single instance carrying both backends fails differently and worse -- Vulkan's vkCreateAndroidSurfaceKHR claims the window in create_surface and the GLES surface built from the same window then reports In Surface::configure / Invalid surface, aborting one frame later in Surface::get_current_texture_view ("Surface is not configured for presentation"). That was measured here on the way to the fix, not reasoned about. Vulkan still wins wherever it has an adapter, so nothing changes on the phone. (2) The surface, adapter and device requests all report through the Result<Self, String> this function already returns, instead of two of the three panicking -- one rule for the set, and surface_changed already puts that string on screen and in the log ring.

  Evidence, this checkout's emulator (API 36, x86_64, debug): before, default features aborted on first launch with Abort message: 'Could not get adapter!: NotFound {...}'; after, iris renderer: no Backends(VULKAN | METAL | DX12 | BROWSER_WEBGPU) adapter on this device, falling back to GLES then new renderer built (Gl) and frames. Then the cases the fix had no reason to touch, clean on both the default build and a force-gles one: two background/return cycles, rotate to landscape and back (already_live=true, the reuse branch), a background/return after the rotation, and cold starts. Vulkan could not be exercised here -- the probe's own answer is that this emulator has no Vulkan adapter, which is the whole defect; Vulkan remains only testable on Iris's phone.

  Also landed with it, and worth more than the fix: a panic hook in iris/android-app/src/app_log.rs. Checked first, rather than assumed: under panic = "abort" a panic's message reaches the tombstone's Abort message and nothing else -- not log, so not the ring, so not Dev Updater's Runtime tab, which is the only surface Iris has on a phone with no adb. The hook writes the message and its location at error level, and -- because the ring is memory only and the process is about to die -- also to last-panic.txt in the app's private directory, which set_crash_dir (called from nativeSetFilesDir) replays into the ring at error level on the next start and deletes. So a crash loop now explains itself in the Runtime tab of the run that is still up. Verified on the emulator by building the unfixed renderer with the hook: iris panic at .../render.rs:140:14: Could not get adapter!: NotFound {...} in the ring on the run that died, and iris app log: the previous run died -- ... on the next one.

• Masks with a shape (done 2026-09-08). docs/LAYOUT.md's "Masks with a shape" carries the design and, at its end, the four places the code is deliberately narrower than it. Chaining landed early, 2026-09-07 (d507ae4). What landed now is the shape half and the hit-testing half:

  Mask is { primitive, parent } -- the slot of a primitive already written, and the mask this one nests inside. The fragment stage evaluates that primitive's own coverage at the masked pixel (mask_coverage in shader.wgsl, the same rounded_rect_coverage a drawn rect goes through) and multiplies it into the pixel's alpha, walking parent and multiplying every coverage on the chain. So the container's corner and its children's clipped corner are one piece of arithmetic, and two nested feathers dim a pixel twice -- the "alpha should be decreased / multiplied" Iris asked for.

  .masked() is unchanged for callers and writes an undrawn RectPrimitive at its own region (Drawn::No/NOT_DRAWN -- owned, moved, resized and freed like any other primitive, simply never rasterized), so "clip to my box" and "clip to that rounded background" are one mechanism rather than a square-cornered special case. New: .masked_by(shape) draws shape behind the content in its own layer and clips to the first primitive it drew, with no radius passed twice -- it replaces .masked().background(w), which drew the two and clipped to the box. transcript-ui's BlockFrame::Verbatim is its first caller, which is the code fence Iris raised this about.

  Hit-testing applies the shape: SensorUi::run_sensors asks UiRenderState::mask_admits (coverage above one half, which is where the drawn edge is) as well as the widget's own box -- the two ask different questions and both have to hold. primitive_corners is a floor-for-floor transliteration of the shader's corners_of, which is the whole reason it is not region.to_px(): the phone's 2.55 density puts nothing on a whole pixel, and skipping the rounding disagrees with the pixels by up to one along each edge.

  A mask's shape must be a rect, asserted by name in Painter::set_mask_to. A glyph would need a CPU-side alpha plane before the hit test could agree with the shader, and a standalone image a bind-group switch the fragment stage cannot make. So the design's texture-mask pass condition is not met and no texture mask exists -- the branch where one would go is in both mask_coverages.

  How it was checked, all four commands:

  cd iris && cargo test --workspace          # layer 1 + the GPU test
  cargo test -p iris --lib layout_tests::    # the four mask tests
  cargo test -p iris --test mask_sdf         # CPU/shader agreement
  ./run-headless.sh phone --phone --shot /tmp/mask.png --seconds 6 \
      -- -p transcript-fixture               # layer 2, for looking
  

  iris/tests/mask_sdf.rs is the only test in the workspace that needs a GPU: it lifts distance_from_rect and rounded_rect_coverage out of iris_core::SHAPE_SHADER by name and runs them in a compute pass over a grid of ~200k points at five radii, against the CPU iris_core::rounded_rect_coverage -- worst disagreement under 1e-5, and the negative control (a + 0.01 inside the shader's smoothstep) fails it at 0.03. It lifts rather than copies because a copy would be edited alongside the shader, which is exactly the drift it exists to catch. The layer-1 tests are in layout_tests.rs: the child's coverage swept across the container's corner arc equals the container's own exactly (the sweep goes from the arc's centre -- the straight chord between the arc's ends lies inside the circle everywhere, so the first version of this test proved nothing and said so); nested masks multiply rather than intersect, asserted where both feathers are partial, which is the only place the two differ; a press in a rounded-away corner misses while one inside the curve and one on a straight edge hit; and a_plain_mask_still_clips_to_a_square_box, the half the shape work had no reason to touch. The screenshot shows the fixture's horizontally scrolled code fence clipped on the curve at both top corners with no square pixels outside it.

  Also landed here, and not on this item's list: the winit backend had the same defect the Android one was fixed for in 85869d0 -- it asked for a Backends::PRIMARY adapter and .expected one. This VM's Venus device disappears whenever the host runs out of virgl contexts, which happened mid-task, and layer 2 aborted with Could not get adapter! while GL sat there working. default::render::UiRenderer:: new now probes and rebuilds the instance on Backends::GL exactly as Android does, and the adapter request names the backends it tried. The rule was written on one member of a set of two; this is the other.

• Compose app: the Reversed range crash in ToolInput.highlighted (docs/TODO.md). Main branch, not rustify.

Desktop and phone share the code (Iris, 2026-09-07)

Iris plans to develop a desktop app as well, and asked that most code be sharable between desktop and phone. The workspace already has that shape -- iris, client-core, transcript-ui and tabs-ui are platform-free, and android-app/desktop-app are the entry points -- so the rule is about keeping it: a platform crate holds only what the platform forces. Today that is JNI, the IME and insets bridge, the surface lifecycle and the bench JNI on Android; winit, argv and the config file on the desktop. What differs is the screen layout, since a phone screen with a finger and a desktop screen with a mouse want different arrangements -- a session list beside the transcript rather than a screen behind it, hover states, keyboard shortcuts. What does not differ is everything a layout is built from: the widgets (a tap button, a text field, a list, a card, a tool-call row), gestures, folding, paging, selection, and the styling -- colours, spacing, type, the surface ladder -- which is the exact same code on both, never a desktop palette beside a phone one. Those are written once in a shared crate, with a platform trait underneath when a behaviour genuinely differs (FocusHost, OpenUrl, and the insets/ime_visible feed are the existing examples). Two checks before finishing a change under iris/: does desktop-app still build and run with it, and is any UI logic newly in android-app that a desktop would also need? The bench client (android-app/src/bench_client.rs, ~1000 lines) is the first thing to look at moving, since a desktop bench on the same fixture is layer 2 of the test rig below.

Three test layers, cheapest first (decided 2026-09-07; layers 1 and 2 built the same day)

Iris's suggestion, adopted and layered: test at the cheapest layer that can answer the question, and go up only when it cannot. The emulator costs minutes a cycle; the desktop window seconds; the headless harness runs inside cargo test.

1. Headless, in-process, no compositor and no GPU -- the default. iris::harness (iris/src/harness.rs), plus the fixture crate it opens. Harness::new(size, density) builds an Rsc, a UiRenderState and a state whose FocusHost/OpenUrl record what the platform was asked for; frame(t_ms)/frames_until(..) run frames on a clock the test owns, and replay(&TouchScript) feeds a recorded gesture one sample at a time exactly as IrisViewPeer::on_touch_event replays Android's historical samples. The recordings are plain t_ms action x y files under iris/transcript-fixture/touch/, and flick-120hz.touch is the phone's own shape: DOWN, four samples 4ms apart, UP, 20ms in total.

   cd iris && cargo test -p transcript-fixture
   

   runs in about a second and asserts (a) the flick releases with a real velocity (List::fling_velocity, which only Released(Some(v)) fills), (b) the list travels and settles inside the AOSP spline's own FlingCalculator::duration, (c) a tap moves nothing and opens no link, (d) a long-press-then-drag leaves selected text and does not pan, and (e) the composer clears a simulated 1000px IME inset (Composer::set_bottom_inset). Each was confirmed to fail without its subject rather than assumed: dropping animate(id) from Selection::drag -- the phone's own "fling does nothing" defect -- and starting the fling curve at the wall clock each fail only the flick test; flinging on Tapped fails only the tap test; a 5s LONG_PRESS fails only the selection test; a set_bottom_inset that ignores its argument fails only the composer test.

   What still cannot be answered below layer 3: nothing renders here, so anything about pixels -- glyph rasterisation, the atlas, stale or duplicated primitives, colour, the surface lifecycle, the renderer rebuild -- is invisible to layer 1 and only looked at in layer 2. Frame times are not measurable at either: layer 1 does no GPU work at all and layer 2 runs a debug build on this VM's virtio GPU, so a number from either is not the phone's. Anything JNI (the IME, real insets, the clipboard, battery) is layer 3 by construction: layer 1 records that the platform was asked and layer 2 has no Android platform to ask.

   The one exception, added 2026-09-08: iris/tests/mask_sdf.rs needs a GPU but no compositor and no window -- it asks wgpu for an adapter, runs two functions lifted out of shader.wgsl itself in a compute pass, and compares the answers with the CPU transliteration in iris_core::render::sdf. It sits inside cargo test because what it checks is arithmetic rather than pixels: the fragment stage and the hit test have to agree about where a rounded edge is, and neither layer 1 (which cannot run the shader) nor layer 2 (where a half-pixel disagreement is invisible) can say whether they do. Reach for this shape only when the question is "do these two implementations of one function agree" -- anything about what is drawn is still layer 2.

2. A phone-shaped desktop window under headless sway -- for looking.

   cd iris && ./run-headless.sh phone --phone --shot /tmp/p.png -- -p transcript-fixture
   

   About 15 seconds warm. --phone sets the private sway output to 1080x2424@120Hz and exports IRIS_SCALE=2.55, which reaches iris the way DisplayMetrics.density does on Android (iris::default::content_scale) -- the desktop backend now lays out in physical pixels with a density instead of dividing into a separate logical space, so both platforms run one path. transcript-fixture's phone example opens the same screen from the same bytes as layer 1 and the Android bench.

   A gesture on screen uses the same recordings:

   ./run-headless.sh phone --phone --replay transcript-fixture/touch/flick-120hz.touch \
       --shot /tmp/p.png -- -p transcript-fixture
   

   writes /tmp/p-before.png and /tmp/p.png either side of the flick; looked at 2026-09-07, the list moved back about seven turns of the fixture and settled.

   swaymsg seat - cursor cannot drive it, and that cost an hour. This compositor runs the headless backend with no input devices (WLR_LIBINPUT_NO_DEVICES=1, LIBSEAT_BACKEND=noop): the cursor commands all report success and nothing whatever reaches the client, with swaymsg -t get_seats showing capabilities: 0 as the only sign. wlroots 0.19 dropped WLR_HEADLESS_INPUTS, and ydotool's uinput device would be ignored by a compositor that is not reading libinput. iris/rig-input's replay-touch uses the virtual-pointer protocol instead, which is a client protocol and needs neither devices nor root, and it parses iris::harness's own TouchScript. Two traps inside it, both found by printing winit's events: a button sent in the same frame as the motion that first puts the pointer over the window is dropped (the client sees the enter, the moves and the release, never the press), so the pointer is positioned and left to settle 200ms first; and a leftover window from an earlier manual run tiles beside the new one, halving the width and producing a screenshot that looks exactly like a duplicated- primitive rendering bug -- swaymsg -t get_tree and pgrep -af examples/phone are the check.

3. The Android emulator -- platform plumbing and the final pass. JNI, IME, insets, surface lifecycle, the renderer rebuild, and one verification run before a build goes to the phone. Not for iterating on layout.

The 2026-09-07 phone report on ed04d4c: the fling was linear, and the keyboard is a targetSdk

Iris's three lines on the ed04d4c build (Pixel 9 Pro XL, GrapheneOS Android 17, Mali-G715, content_scale 2.55, 120Hz): item 4 (the resume glyph corruption) is fixed, confirmed on the phone; "flinging now does technically do something, but it seems to just be linear velocity with an abrupt stop"; and "similarly, the keyboard raising up does not push things upwards."

The fling was exactly, arithmetically linear. Not approximately. android_fling_spline::distance_fraction(t) returned t for every t, which is a constant-speed slide for the full duration() and then a stop at full distance -- Iris's sentence, read straight off the code. Two transposed halves of one AOSP loop did it, and they compounded:

1. AOSP's SplineOverScroller static initialiser solves the bisection on the P1/P2 curve and samples SPLINE_POSITION[i] from the tension curve (coef * ((1-x) * START_TENSION + x) + x³); the second half of the loop does the reverse to build SPLINE_TIME. iris had both halves solving on the tension curve and sampling P1/P2 -- so its two loops were the same computation, and SPLINE_POSITION == SPLINE_TIME element for element.
2. The lookup then bracketed t between SPLINE_TIME entries and interpolated SPLINE_POSITION. AOSP brackets between the even time steps index / N and (index + 1) / N (SPLINE_TIME is used only by adjustDuration, which iris has no analogue of). With the two arrays identical, d_inf + (d_sup - d_inf)(t - t_inf)/(t_sup - t_inf) reduces to t_inf + (t - t_inf) = t.

Every test the calculator had compared it with itself -- monotonic, signed, integrates to the closed form, per-tick deltas non-increasing -- and all of them pass on a linear curve. That is the shape to distrust: <= is not deceleration.

Sources, read rather than remembered. frameworks/base core/java/android/widget/OverScroller.java from android.googlesource.com (?format=TEXT, base64), and androidx.compose.animation:animation:1.12.0's SplineBasedDecay.kt and FlingCalculator.kt out of the -sources.jar on dl.google.com/dl/android/maven2 (there is no androidx checkout here and cs.android.com is JS-only; androidx.tech is now a parked domain serving an unrelated site). The two agree line for line, which is why iris ports one curve rather than two. The formulas, for the record:

P1 = START_TENSION * INFLEXION            = 0.5 * 0.35
P2 = 1 - END_TENSION * (1 - INFLEXION)    = 1 - 1 * 0.65
SPLINE_POSITION[i]: solve coef*((1-x)P1 + xP2) + x³ = i/100 for x,
                    then take coef*((1-x)ST + x) + x³
physical_coeff = 9.80665 * 39.37 * density * 160 * 0.84
l              = ln(0.35 * |v| / (0.015 * physical_coeff))
distance       = 0.015 * physical_coeff * exp(rate/(rate-1) * l)
duration       = exp(l / (rate - 1)),  rate = ln(0.78)/ln(0.9)
at time t:  index = floor(100 * t/duration)
            vcoef = (POS[index+1] - POS[index]) * 100
            position = distance * (POS[index] + (t/duration - index/100) * vcoef)
            speed    = vcoef * distance / duration

What changed. iris/src/sense.rs's android_fling_spline builds one table, indexed by even time steps, and sample(t) answers AOSP's distanceCoef/velocityCoef pair; FlingCalculator gained velocity_at beside position_at. iris/benches/fling_spline_reference.py is an independent hand transcription of both sources and prints the numbers the tests assert on -- checked in because "numbers computed by the code under test" is exactly how the last three tests passed through this defect. Tests: the_spline_matches_aosps_own_table (the curve is not the identity: 27.4% of the distance at a tenth of the time, 85.8% at half), a_flick_decelerates_the_way_aosp_says_it_does (11064px/s at density 2.55: 6334px over 1.636s, speed 9202 -> 4733 -> 2650 -> 951px/s), and tick_fling_applies_shrinking_incremental_deltas strengthened from "non-increasing" to "the last delta is under 80% of the first". Negative control run: with sample forced back to returning t, exactly those three fail and the other eleven pass.

Emulator evidence (API 36 AVD, debug, force-gles, 2026-09-07). A ui-trace swipe of 900px in 120ms releases at v=3750 and the new iris fling tick: debug line reports, frame by frame, speed=-3746 -> -2624 -> -1834 -> -1144 -> -752 -> -449 -> -243 -> -83px/s over 32 frames ending at t=0.664s, with the per-frame dy falling 94 -> 34 -> 20 -> 13 -> 8 -> 4.4px. The end: the same flick in the other direction, from a list already at its newest end, produces exactly one tick and stops -- no overshoot. A finger during a fling: swipe, then a tap 200ms later, ends the fling at t=0.248s and 11 ticks instead of running its full 0.55s.

The keyboard: the bench app targeted SDK 34. app/build.gradle said targetSdk = 34 while compileSdk was 37 and the Compose app in app/ targets 37 -- and that Compose app's keyboard does push its transcript up on Iris's phone. Below target 35 a window keeps the legacy behaviour, where adjustResize shrinks the window for the IME so getInsets(ime()).bottom measures the overlap with an already-shrunk window and is zero; MainActivity's setDecorFitsSystemWindows(false) opts out of that and still takes on the API 36 emulator here, which is why every test run showed the push-up working. It is deprecated as of API 35 and Android 17 is where it appears no longer to. Fixed by targetSdk = 37, where edge-to-edge is not opt-in.

That is a reading, not a measurement -- no Android 17 device is reachable from here -- so the second half of the change is making the phone able to answer it. MainActivity now also registers a WindowInsetsAnimation.Callback (DISPATCH_MODE_CONTINUE_ON_SUBTREE, onProgress forwarding, onEnd re-reading getRootWindowInsets so an interrupted animation cannot leave a frozen value), which delivers the IME height on devices where only the animation path carries it and, on every device, makes the push-up animate with the keyboard: the emulator log now shows ime_bottom=509, 663, 833, 881, 883 instead of one jump to 883. And insets::Shared::updates counts every dispatch, which AndroidUiState::insets_report() puts in the Diagnostics pane:

insets: dispatches=27 left=0 top=142 right=0 bottom=63 ime_bottom=0 ime_visible=false

Screenshot-verified on the emulator. Iris has no logcat, and "the listener never fired" and "it fired with a zero height" look identical on screen; dispatches=0 prints its own sentence instead of the numbers, since those would be defaults rather than measurements.

What to look at on the next build. Open the keyboard, press Diagnostics, screenshot the insets: line. ime_bottom in the hundreds with the composer risen: fixed. dispatches climbing but ime_bottom=0: the targetSdk reading was wrong and the window is still being resized. dispatches=0: the listener is not being called at all, which is a different fault from either. For the fling, a flick should now visibly slow before it stops rather than running out at speed.

The fling started too slow: Compose fits a curve, iris averaged (2026-09-07)

Iris on the 4274b8b build: "flinging now actually works but is slower than Compose's immediately after releasing the flick (the slow down seems correct)." The curve was right -- that was the previous fix -- so the wrong number was the initial velocity.

What iris did. VelocityTracker held the last 100ms of per-frame pan deltas and answered their sum over the span between the oldest and newest: an average. An average cannot tell an accelerating flick from a steady drag, and a flick is by definition accelerating, so every fling started at roughly the speed of the middle of the gesture rather than the speed at the release. Every test it had asserted the average's own definition back at it, which is the same shape of self-grading the spline shipped a straight line through.

What Compose does, which is not what it is remembered as. Read out of the -sources.jar of androidx.compose.ui:ui-android:1.12.0 and androidx.compose.foundation:foundation-android:1.12.0 on dl.google.com -- the versions app/gradle/libs.versions.toml builds the Compose app against, which is the app being compared with:

• scrollable/draggable release through DragGestureNode. sendDragStopped, which calls the 2D VelocityTracker. On Android that delegates to Lsq2VelocityTracker -- two VelocityTracker1D(strategy = Lsq2) over absolute positions, fitting a degree-2 polynomial by least squares (polyFitLeastSquares, Gram-Schmidt QR) and taking its derivative at the newest sample.
• Strategy.Impulse is not on the touch path. Its only route in is DifferentialVelocityTracker, whose only caller is NonTouchScrollingLogic: mouse wheel and trackpad. AndroidComposeUiFlags.isFrameworkVelocityTrackerEnabled, which would swap in the platform's own (impulse) tracker, defaults to false. This was the surprise of the port, and the reason to read rather than remember -- the plan for this task named Impulse.
• Constants: HistorySize = 20, HorizonMilliseconds = 100, AssumePointerMoveStoppedMilliseconds = 40, minSampleSize = 3 for Lsq2. The walk back from the newest sample stops at the first sample older than the horizon or separated from its neighbour by more than the stopped gap.
• Which samples. sendDragStart adds the DOWN change; every subsequent MOVE, historical samples included, is added by sendDragEvent. The UP position is never added -- Lsq2VelocityTracker.addPointerInputChange wraps its addPosition calls in if (!event.changedToUpIgnoreConsumed()), and the UP branch only resets the tracker when 40ms have passed since the last MOVE (b/238654963). So a finger that comes to rest before lifting reads as a stop, not as a decelerating tail.

The clamps, both checked rather than assumed. Maximum: ViewConfiguration.getScaledMaximumFlingVelocity(), 8000 dp/s, passed into calculateVelocity(maximumVelocity) at the release -- 20400px/s at the phone's density 2.55, which this flick does not reach. Minimum: there is none on the fling path. ViewConfiguration. minimumFlingVelocity (AOSP's 50 dp/s) exists in Compose's ViewConfiguration interface, but its only use in either artifact is NestedScrollInteropConnection, for View interop. DefaultFlingBehavior.performFling guards with abs(initialVelocity) > 1f and says why in its own comment ("we need it since spline curve gives us NaNs"). So iris applies 1px/s, not 50dp/s: a 50dp/s floor would swallow slow deliberate releases that Compose flings. Both live in List::fling, which is the only place that knows the density the dp figure has to be multiplied by.

What changed. VelocityTracker holds positions, not deltas (Lsq2 refuses differential data in Compose too), capped at 20 samples; add_sample(delta, at) is now add_position(position, at), and DragGesture feeds the raw window coordinate along the drag axis at the press and at every Pan frame -- the same set Compose feeds, minus the one MOVE that crosses the touch slop, which Compose drops only because sendDragStart happens to add just the DOWN. poly_fit_least_squares is Compose's polyFitLeastSquares on fixed-size arrays.

iris/benches/velocity_reference.py is the independent transcription, same role as fling_spline_reference.py, and prints every number the tests assert. On the phone's own recording (transcript-fixture/touch/flick-120hz.touch, five samples in 16ms):

sample set	shipped (average)	Compose (Lsq2)
flick-120hz.touch	12250 px/s	15250 px/s
steady drag, 5px/10ms	500 px/s	500 px/s
accelerating flick, deltas doubling	1080 px/s	2445 px/s
old fast burst then 1px/10ms	9182 px/s	100 px/s
stopped 48ms, then released	2533 px/s	0 px/s
press + one move frame	12500 px/s	0 px/s

The recording understates the change (1.24x) because it is only 16ms long; the accelerating set (2.26x) is the shape of a real finger flick and is what Iris was feeling. The last three rows are the cases an average gets not just low but wrong: it flings from a standstill, and it flings from two points that describe no curve.

Corrected 2026-09-07 (docs/REVIEW-2026-09-07.md's D5): the first and last rows of the "shipped" column, and the 1.30x, read 11750 px/s -- one number in two rows for two different sample sets, which is the tell. The script prints 12250 for the recording (196 px over its own 16 ms) and 12500 for the press-plus-one-move set (100 px over 8 ms). Where 11750 came from is only half recoverable: it is 196 px over 16.68 ms, i.e. the recording's travel divided by a 60 Hz frame rather than by the span the file itself records, which explains the flick row and does not explain the other one -- that one was copied. The rule at the top of this section stands: these numbers come from velocity_reference.py, and a disagreement is fixed by running it, not by running the Rust.

Tests. Eight in sense::velocity_tracker_tests, two rewritten in sense::drag_gesture_tests (three samples is the fewest that can fling; one move frame answers 0, as Compose does), and phone_screen.rs now asserts -15250px/s rather than "more than 1000". Negative control run: with velocity reverted to total / span, exactly seven fail -- the flick recording, the accelerating flick, the horizon, the stopped finger, the minimum sample count, both drag_gesture flick tests -- and the steady drag, the tap, the selection release, all sixteen arbiter tests and the whole of phone_screen.rs bar the flick pass unchanged. That is the half the change had no reason to touch.

iris drag release: keeps its info line and gains a debug one, iris drag release samples:, printing every held sample as t_ms:position relative to the first. Iris has no logcat, so that is the only way a flick that felt wrong on her screen becomes something replayable: paste it into a touch/*.touch file for layer 1, or straight into velocity_reference.py.

The 22:16 phone report, worked 2026-09-06/07

Iris's four items are listed in docs/IRIS_TODO.md's "From the phone, 2026-09-06, 22:16"; this is what was found and what was run. Item 4 was committed on its own (ba2afba); items 1-3 and everything below landed together after the emulator evidence.

Item 4, text cooked after a resume -- root cause, fixed in ba2afba. Not "the cached text primitives are never redrawn": they are. IrisViewPeer::surface_changed (iris/src/android/view.rs) calls render.resize(...) on every surface event including the new-renderer branch, which sets UiRenderState::resized, which makes the next update take redraw_all rather than redraw_updates -- so after a resume every widget's draw runs again. The stale coordinates come from 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, so TextData::place is never reached, no glyph is re-rasterised into the fresh atlas, and the previous atlas's uv_min/uv_max/layer go straight back to the GPU. Text whose content changed after the resume -- the diagnostics pane Iris tapped -- re-shapes and is therefore perfect, which is exactly the split in her screenshot. The fix is one mechanism: a generation counter on GlyphAtlas, bumped by clear, recorded on each RenderedText, added to TextView::render's cache key, with a debug_assert_eq! in Painter::glyphs that a submitted quad's generation is the live one. Test clearing_the_atlas_re_renders_cached_text_instead_of_reusing_it, run and passing; still needs phone-side confirmation, since no emulator here has a Vulkan adapter and the GLES path may not destroy the surface at all.

Item 2, the keyboard would not reopen -- fixed and confirmed. attr.rs's on_press, already-focused branch, now calls focus_gained on a tap that stays inside DRAG_SLOP, which is what Android's own EditText does (showSoftInput is idempotent). Emulator, 2026-09-07: first tap mInputShown=true; back gesture; second tap mInputShown=true and the composer rises again. Negative control run: with that one call removed and nothing else changed, the second tap leaves mInputShown=false -- Iris's report exactly. The case the fix had no reason to touch, also run: a horizontal swipe across the focused composer and a vertical swipe out of it both leave mInputShown=false, so her earlier "if I swipe over the input bar it brings up the keyboard" has not come back.

Item 3, the IME height -- fixed and confirmed. MainActivity.java sends getInsets(ime()).bottom and isVisible(ime()) as two separate values (the height used to be sent as the boolean 1/0, which is why nothing could pad by it); Insets/WindowInsets carry both, and bench_client.rs reads the boolean for its state machine and the height for Composer::set_bottom_inset. The list follows for free -- it is .height(rest(1)) in the same Span as the composer bar, so the bar growing shrinks the list. Emulator, 2026-09-07: iris insets: ... bottom=883 ime_bottom=883 ime_visible=true, and the composer's box moves from 31,2277..1048,2329 to 31,1457..1048,1509 -- 820px, which is 883 less the 63px navigation bar it was already clearing. Screenshot checked: the transcript ends above the composer, which sits on the keyboard.

Item 1, the fling -- two more defects behind the first, all three fixed here; the phone is what settles it. The velocity half is what the report predicted: on_touch_event read only each MotionEvent's final position, so a batched 120Hz flick fed the tracker one sample and velocity() answered 0.0. It now replays every historical sample (getHistoricalAxisValue/getHistoricalEventTimeNanos) through the sensor pass, CursorState carries the sample's own time (so a replay loop's speed cannot become the measured velocity), and the press itself is a sample, as Android's own VelocityTracker does with ACTION_DOWN. iris drag release: samples=… span=…ms v=… outcome=… logs the decision. Then the emulator showed the two the report could not have known about:

1. Nothing ever advanced a fling. List::fling sets the state; tick_fling moves it; and tick_fling's only caller in the workspace was bench_client.rs's own fling phase, which drives it in a loop. So the benchmark flung and a finger never did -- and the earlier "verified flinging on the emulator with render() counts" was that benchmark measuring itself. Measured before the fix: frames stop on the same millisecond as iris drag release. iris now has one animation mechanism -- Widget::tick(now) -> bool, ids registered with UiData::animate, drained each frame by UiData::tick_animations, which both backends call before the draw and re-request a frame from while it answers true. List::tick is tick_fling; Selection::drag registers on Released(Some(v)). Test: a_registered_fling_is_driven_by_tick_animations_and_then_ unregisters, confirmed to fail without the registration.

2. The fling lasted 45 seconds. Visible only once flings animated at all. Two causes, both in FlingCalculator: List::fling hardcoded FlingCalculator::new(1.0) while the velocity it is fed is in physical pixels (List reads painter.density() now), and physical_coefficient multiplied by FLING_FRICTION (0.015) where AOSP multiplies by its own tuning constant 0.84 -- a coefficient 56x too small, put through exp(ln(…)/(rate-1)). Every existing test compared the calculator with itself (monotonic, signed, integrates to the closed form) and so passed throughout; a_flick_lasts_what_aosps_own_formula_says_it_does pins the absolute numbers against AOSP's formula worked by hand. Emulator after both: release at v=11064, frames for 1.62s, then none -- against AOSP's own 1.586s for that velocity at density 2.75.

   What Iris should look for on the phone: adb logcat | grep "iris drag release". samples=1 or span=0.0ms means the historical replay is not reaching the tracker on her device; a sensible samples/span with v= in the thousands and outcome=Released(Some (…)) means the gesture is measured correctly and anything still wrong is downstream of it. outcome=Tapped means the flick never crossed the slop.

Two things found on the way, both pre-existing at ba2afba.

• MOVE_CHAIN_LIMIT was 16 and the composer's chain is 17. Tapping the composer in any debug build aborted on resolve_move_chain's assert; in a release build (what Iris runs) the walk simply stops summing, on the CPU and in shader.wgsl, so a widget past the bound draws and hit-tests short by whatever the outer slots held, with nothing on screen to say so. Both constants are 64 now, and the assert prints the chain (64(0, 0) -> 63(0, 0) -> … -> 0(0, 0)) so a cycle and an honestly-deep tree can be told apart -- which is how this one was: 17 distinct slots.
• minSdk is 29, up from 26. getEventTimeNanos and getHistoricalEventTimeNanos are API 29, and a missing JNI method there is a hard crash on the first touch rather than a degraded fling. build-apk.sh's cargo ndk -P matches.

Still open and pre-existing: the composer bar's grey background is not drawn on the transcript-screen bench build, so the transcript shows through where the bar should be (Stack{StackSize::Child(1)} is the thing to look at). Unchanged by any of the above.

Task A, closed 2026-09-06: the composer scrolls on a finger

iris/transcript-ui/src/composer.rs is field.scrollable().masked() now. Verified on this checkout's emulator -- the evidence and the numbers are in docs/IRIS_TODO.md's ticked "composer has no touch-drag scroll" item.

The premise the task was given under was wrong, and that is worth recording: Scroll did not measure against the window. Its used.within_len(container).to_abs(output_size) came to exactly abs + rel * container_px -- the right number by a route that reads as if the window were the container, which is what cost a session. It is painter.px_size() and to_abs(container_len) now: same arithmetic, stated the way the invariant is. Scroll also still reports its content's size upward, deliberately -- reporting the container makes the answer a function of itself (the bar is sized from that report, so it collapses to nothing and never recovers; measured in the headless harness before the shape was settled).

What actually broke the composer was three separate defects, each now carrying a headless regression test in iris/src/layout_tests.rs that was confirmed to fail without its fix:

1. A MaxSize reported its cap as an unresolved dp. Span::draw places a child from the abs/rel of the length it reported, so dp(168) was worth zero and the bar got a slot of nothing the instant its content passed six lines; the Scroll inside then measured its container at -63px (the padding subtracted from nothing) and panned the whole message out of view. Emulator log, before the fix: container=-63 content=415.8 amt=478.8. Fixed by Len::fold_dp (new), used by MaxSize and Sized on the way out, and guarded for every widget by a debug_assert! in UiRenderState::draw_inner that a reported Size carries no dp. Test: a_dp_cap_is_reported_in_pixels_so_a_span_can_place_it.
2. A Masked allocated a fresh mask slot on every draw. draw_inner's unchanged-region fast path means its descendants are mostly not redrawn with it, so they kept clipping against the slot they were first drawn under -- measured on the composer's tree at four live mask entries, none of them the widget's current box, and the field drew nothing at all. The slot is allocated once and rewritten in place now (ActiveData::own_mask, Painter::set_mask), with its path out in remove's undraw branch. Test: a_masked_widget_keeps_one_mask_slot_that_is_always_its_own_region.
3. A panned widget's own hit box moved twice. mov updates active.region and accumulates the same delta on the widget's move slot, and resolved_region added both -- so after a finger pan the composer's field was untappable, while its descendants were fine (which is why hit_testing_follows_a_scrolled_widget, which checks a descendant, never saw it). ActiveData::move_applied records the part of the slot's delta region already accounts for. Test: a_panned_widgets_own_hit_box_moves_exactly_once (fails at exactly 2x the pan without it).

Still open, and pre-existing (present in the build before this change, so not the scroll area's doing): the composer bar's grey background is not drawn on the transcript-screen bench build, so the message reads as white text over the transcript. Stack{StackSize::Child(1)} is the thing to look at.

Rig fix on the way past: iris/android-app/run-bench.sh polled logcat for "iris bench report:", which copy_report also logs at startup ("nothing to copy -- run the benchmark first"), so it returned instantly and printed a report that had never been run. It polls for the report's own first line now.

Task B, closed 2026-09-06: a streamed delta costs one markdown block

A transcript row was one TextEdit holding the whole message, so every delta re-shaped every paragraph of a long reply through parley -- the one phase where iris trailed Compose on Iris's phone. A row is a column of one TextEdit per top-level markdown block now, and a delta that lands in the last block is one set_with_spans on that block.

• client-core/src/markdown_blocks.rs is the split: split_blocks (top-level blocks with their source, via the same pulldown-cmark the renderer parses with, so the two cannot disagree about where a block starts) and common_prefix. Seven tests, including the one that says the fast path must compare rather than assume: appending --- under a paragraph turns that paragraph into a heading, so an already laid-out block is not always still what it was.
• iris/transcript-ui/src/row.rs builds the column and owns RowBlocks::apply_delta; lib.rs keeps the tail row's blocks (TranscriptScreen::tail) since that is the only row a delta reaches.
• A block is the selection unit, not a row: Selection is keyed by SelKey = (RowKey, u32), which compares in reading order at both levels so every range query in that file is unchanged. The list-level (pointer-captured) half of a drag resolves the block under the finger from its drawn box (Selection::locate) instead of doing arithmetic from the row's extent.

Pass condition, met: a_delta_into_a_long_reply_redraws_the_same_widgets_as_a_short_one (transcript-ui/src/lib.rs) drives a real UiRenderState and asserts the Widget::draw count for one delta into a 100-paragraph (3,000+ character) reply equals the count for the same delta into a one-paragraph reply. 30 either way. It is a real test, not a tautology: it read 630 against 30 at three points on the way -- once because Span's measure pass redrew every child, and once because build_tree did not seed tail, so the first delta after opening a screen took the rebuild path with nothing on screen or in take_rebuilds() to say so.

Two things tried and dropped, so the next session does not redo them. Painter::measure (a container asking a clean child for its size instead of drawing it provisionally) fixed one of the 630s but the test passes without it once the tail seeding is right, so it was removed rather than kept on speculation. And the emulator's own numbers say the remaining cost is not in the block split.

Verified on the emulator beyond the counter: the transcript draws its blocks with their own spacing (heading, prose, fence), and ui-trace record --do "holddrag 300 700 700 1000 700 600" logs iris selection: begin at row (3187, 0) then extend to row (3187, 1) with the highlight crossing from the heading into the code block -- a selection that spans blocks, which is what the re-key had to keep.

Bench, stream phase, before and after Task B (emulator, 2026-09-06)

iris/android-app/build-apk.sh debug --abi x86_64 --features "transcript-screen bench force-gles" + run-bench.sh, this checkout's AVD. Emulator absolutes transfer nothing; the before/after ratio on the same emulator does.

Same AVD, same fixture, same build flags, 20 minutes apart. Emulator absolutes transfer nothing; the ratio does.

before                          after
stream: 202 frames over 21.0s   stream: 293 frames over 21.0s
  late: 197 (97.5%)               late: 285 (97.3%)
    p50 61.5ms                      p50 54.5ms   (-11%)
    p90 211.7ms                     p90 113.1ms  (-47%)
    p99 342.6ms                     p99 137.4ms  (-60%)
  worst 403.6ms                   worst 143.0ms  (-65%)

The tail is where the whole-message re-layout lived, and it is where the change shows: 91 more frames delivered in the same 21 seconds. The p50 moves least, which is consistent -- a delta into a short message never cost much. The phone number is Iris's to take; nothing here is a statement about her device.

Verification pass over Tasks A and B, 2026-09-06

Read of git diff fb6b459..HEAD -- iris/ client-core/ against LAYOUT.md, TEXTURES.md, IRIS.md/DECISIONS.md's 2026-09-06 entries and CODE_RULES.md, with the emulator. Verdict: deliverable to the phone. One real defect found and fixed, two missing guards added, one open item closed as stale.

1. The block model is correct. split_blocks was checked against the shapes a real transcript has -- a fence with blank lines, a --- inside a fence, a nested list, a fence directly under a heading, a table, a quote -- and against the property apply_delta rests on, at every character boundary of a message containing all of them: growing a message may rewrite its last block and never an earlier one, or common_prefix says so. No defect (client-core's every_prefix_of_a_streamed_message_keeps_all_but_its_last_block, commit a56a928). A delta closing a fence, a delta mid-word and a stream ending inside an unterminated fence are each their own test.

2. iris/core/src/ui/render_state.rs, draw_inner's size-independent fast path: fixed (commit e63e923). It rewrites the widget's own primitives in place and writes no move-slot delta, so unlike mov there is nothing for move_applied to count; 167862c counted one anyway, and resolved_region then subtracted a distance the chain never held. Every such widget's hit box sat short of its drawing by its last step, with the drawing correct -- nothing on screen to say so. Span reaches this on the first frame of any tree containing a Rect (the .background(rect(..)) idiom, list row tints), because it measures each child at the full region and then places it. Pinned by a_size_independent_widget_moved_by_its_parent_has_the_hit_box_it_is_drawn_at, the sibling of a_panned_widgets_own_hit_box_moves_exactly_once on the branch that fix had no reason to touch.

3. Selection across blocks is sound; its rebuild path had no test (commit 155d899). SelKey = (RowKey, u32) orders lexicographically, which is reading order at both levels, so begin/extend/locate and the range queries carry over unchanged; selected_text joining with a blank line is right for blocks as well as rows, since that is how markdown separates them. The gap was the tail rebuilt under the same key with fewer blocks -- the dropped blocks keep pointing at widgets replace_back's drop frees, and Selection::begin resolves every registered handle on an ordinary press, so the next tap anywhere panics. e1030d6's unconditional unregister is correct and now has a_tail_rebuilt_with_fewer_blocks_leaves_none_of_them_in_selection, confirmed to fail (3 blocks still registered, expected 1) without it. Selection::registered_blocks is the test-only accessor that lets it assert the contract rather than only that nothing panicked.

4. The three new debug_assert!s are whole-set, not one member. Len::fold_dp's is in draw_inner after every Widget::draw, so it governs the set by construction; Pad and Span were checked and already fold through apply_rest, and Sized/MaxSize are the two that reported a caller-written Len raw. own_mask's reuse lives inside Painter::set_mask itself, whose only caller is widget/mask.rs. move_applied has exactly two writers, mov and reposition (now one, after finding 2), and resolved_region is the only reader -- window_region goes through it.

5. The O(last block) claim now holds for parley, by counter (commit c3cfc67). take_counters gained a fourth number, text shapes, bumped in Painter::render_text -- which TextView::render only reaches on a cache miss, so it counts shapes and not requests. A draw counter cannot stand in for it either way. Measured: one delta into a 100-paragraph reply shapes exactly 1 text layout, the same as into a one-paragraph reply.

6. The composer bar's grey background is drawn -- IRIS_TODO.md's "still open, and pre-existing" note is stale and has been corrected. Measured by decoding the screencap rather than eyeballing it: the bar is rgb(41,40,49) (the declared 40,40,46 after sRGB rounding), full width, y2245..y2365 on the 1080x2424 AVD, with the field at 31,2277..1048,2329 and the 63px nav strip below. Whatever the note saw, Task A's MaxSize/own_mask fixes closed it.

Checks run: cargo fmt --all --check clean in both workspaces; cargo clippy --workspace --all-targets warning-free (only the pre-existing future-incompat note about wgpu/naga/winit); cargo test 81 (iris) + 13 (iris-core) + 20 (transcript-ui) + 123 (client-core), all passing. One bench run on this checkout's AVD with the assertions live, debug x86_64 force-gles, no abort and nothing in logcat: stream: 298 frames over 21.0s, late 287 (96.3%), p50 52.8ms, p90 108.1ms, p99 137.3ms, worst 148.9ms -- reproducing the "after" column above.

• Merge the DragGesture work -- done 2026-09-06 (merge commit f802de9, git merge --no-ff worktree-agent-a754368325fa06839, clean, no conflicts across the 8 files e12c708 touched). Targets two of the four bench-v2 defects: finger flings dropped by per-widget hit testing (pointer capture + CursorSense::Drop), and IME insets never redelivered (MainActivity.java edge-to-edge). Tap-vs-swipe/DragGesture overlap, reasoned through: attr.rs's on_press (composer focus) and sense.rs's DragArbiter/ DragGesture (list pan-vs-select) do not share a mechanism, but they don't need to -- on_press never calls capture_pointer, so it only ever sees an ordinary per-frame hit-tested Pressing/ PressEnd (run_sensors' region.contains(cursor.pos) check, unaffected by capture unless this widget requested it), the same as before DragGesture existed. The two only interact where a gesture starts on the composer and travels into the list's region; run_sensors already delivers Pressing to whichever widget's current position contains the pointer, so List starts getting frames the instant the finger crosses the boundary -- with no PressStart of its own, which is exactly what DragArbiter:: is_idle()'s 2026-09-05 recovery branch exists for. No consolidation needed; DRAG_SLOP is already the one shared constant (attr.rs imports it from sense.rs, not a second copy). Checks, 2026-09-06 merge pass: cargo fmt --all clean; cargo clippy -p iris -p iris-core -p transcript-ui --all-targets and the same for -p desktop-app -p tabs-ui, zero warnings beyond the pre-existing external-crate future-incompat notice (naga/wgpu/wgpu-core/wgpu-hal/winit); cargo test --lib -p iris -p iris-core -p transcript-ui and -p desktop-app -p tabs-ui, 97 passed/0 failed, including the review-fix tests below. cargo test --workspace/cargo clippy --workspace --all-targets (the full-workspace forms, which also build iris's winit examples) were abandoned after 40+ minutes each stuck compiling one example binary with uptime reading a load average of 66-78 on this 8-core VM (3-4 concurrent peer cargo/cargo check invocations the whole session) -- ps -o time on the stuck rustc showed 2 seconds of accumulated CPU time after 38 minutes of wall time, confirming scheduler starvation rather than a hang. The per-package --lib form above is what actually exercises the changed code and finished in under 4 minutes warm. android-app (iris-android-app) is excluded from the host workspace (iris/Cargo.toml, needs the NDK target) and is covered instead by the APK build below, which compiles it for x86_64-linux-android.

  **Emulator checks, 2026-09-06** (this checkout's `ai-app-2` AVD,
  `iris/android-app/build-apk.sh debug --abi x86_64 --features
  "transcript-screen bench force-gles"` -- plain Vulkan crashed on
  this AVD's boot this pass, `wgpu_core::instance: enabled backend
  Vulkan has no adapters`, unrelated to this merge and worked around
  with `force-gles` the way I5's own box already documents for this
  hardware):
  - **(a) tap-vs-swipe still holds.** Fresh app launch, `dumpsys
    input_method`'s `mInputShown=false` at rest. `ui-trace record
    --do "swipe 540 1510 540 700 200"` (a swipe starting on the
    composer's own box, read from `ui-trace show -m Message --field
    box` as `31,1488..1048,1540`) leaves `mInputShown=false` and the
    box unmoved (no keyboard-driven resize). `ui-trace record --do
    "tap 540 1510"` on the same field then reads `mInputShown=true`.
    Matches `20b1225`'s original result -- the `DragGesture` merge
    did not disturb it, confirming the reasoning above.
  - **(b) a real finger fling keeps the list moving after release.**
    Screenshot-hash sampling (`adb exec-out screencap`, `md5`, since
    transcript rows carry no per-row accessibility label yet -- I5's
    own leftover -- so `ui-trace show` cannot track them) at ~40-60ms
    intervals through and after a fast `swipe 540 1400 540 400 120`
    (with room to scroll confirmed by a preceding slow drag) caught
    two *distinct* post-release frames in one run (a settle-position
    beyond the raw drag's own last frame), and every run showed
    28-32 `iris::android::view: render()` log lines per gesture
    against an idle baseline of 0 in 1.5s and roughly 8 expected from
    a bare 120ms drag's own `Pressing` frames alone -- i.e. redraw
    kept being requested well past the finger lifting, which only
    happens while `List::tick_fling` is still returning `true`.
    Some runs' screenshots showed only the drag's own jump with nothing
    further *visibly different*, which is consistent with a real but
    small/fast-settling fling (a modest synthetic-touch velocity's
    spline tail moves little per frame) rather than absence of one --
    the render-count signal did not vary between those runs and the
    one with a visible second frame. Recorded as confirmed, with that
    caveat, rather than measured to a number; a phone verification
    (Iris's own report closes this properly) is still open per
    `IRIS_TODO.md`'s item.
  - **(c) `on_insets_changed` fires on an IME toggle, with confirmed
    cycles.** `run-bench.sh`'s report: `keyboard: shown 4/5, hidden
    5/5 (confirmed via on_insets_changed)` -- the "could not be
    shown" unknown-state line (`bench_client.rs::run_keyboard_phase`)
    did not fire, unlike the pre-`DragGesture` build this same report
    format existed for.
  Worktrees removed after the checks above: `agent-a754368325fa06839`
  (the source branch, its own emulator stopped first via `cd` into
  it + `emu down`), `agent-a27094a7db775552a`, `agent-a1ff0294b6c29127e`,
  `agent-a9002910a315fe719` -- each confirmed `git rev-list --count
  rustify..<branch>` = 0 and no uncommitted changes first; their
  branches deleted too. `agent-a16b22e34539b810e` and
  `agent-a6e37a2335f436d08` left alone -- both `git worktree list`
  `locked` to a live peer agent.
  

• Fix docs/REVIEW-2026-09-06.md, done 2026-09-06, after the merge (finding 1's shape and location in selection.rs/lib.rs were unchanged by the merge, which touched Selection but not apply's Rebuild arm). All ten findings fixed -- new Selection::clear() for finding 1 (the simplest option the review named: clear the same way List::clear() clears the list, let push_row re-register survivors), five debug_assert!s (2-5, plus 7's restructure), and three new tests (8, 9, 10), confirmed with the apply_tests::a_row_dropped_by_a_regroup_does_ not_outlive_itself_in_selection test passing (it exercises exactly finding 1's shape: build a real TranscriptScreen, force the same regroup diff_tests already covers, apply, then a surviving row's begin -- panics pre-fix, per the review's own test-8 ask). docs/IRIS.md's 2026-09-05 entry gained the line the review's "Docs" section asked for. See docs/REVIEW-2026-09-06.md's own "Fixed, 2026-09-06" section for the per-finding account. Committed together with the review file.

• [~] Iris's 11:39 phone report on the 02:07 build (four items, verbatim in IRIS_TODO.md's "From the phone, 2026-09-06, 11:39"): composer floating two thirds down the screen at launch with black below it; a swipe starting on the composer held until the finger leaves it; no fling (expected, DragGesture unmerged); text still lost on app-switch on the phone despite the emulator-verified atlas reset. The first and last are the same class as the next box and go to that agent; the middle two are the merge box's.

• [~] Stale primitives and invisible composer text, 2026-09-06: typed text is fixed and was never a renderer bug at all; the two duplicate-drawing halves are not reproducible on this checkout's emulator any more and are recorded below with what changed.

  **1. Typed text (P0 box item 2, `IRIS_TODO.md`'s own item) --
  fixed.** 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; an empty field lays out to a zero-width box, so
  tapping an empty composer granted focus and opened the keyboard with
  no caret, and `insert_str` returns early without one -- every
  keystroke was dropped in silence. **Gboard's suggestion strip is
  Gboard's own composing state, not a read of our buffer**, which is
  what made the earlier pass conclude the buffer held the text and
  send the search downstream into the renderer; the `accessibility`
  dump saying `text=""` for the `Message` node was the first
  contradicting evidence. Parley clamps a point outside the layout by
  itself, and a press reaching `select` has already been hit-tested to
  the widget, so the "outside" branch had nothing left to mean. Three
  new tests in `edit.rs`, one of which fails on the pre-fix code, plus
  a `debug_assert!` in `insert_str` so an insert with no caret fails at
  the mistake rather than dropping input -- it caught
  `layout_tests::composing_text_after_a_keyboard_resize_...` typing
  into an unfocused field the moment it was added. **Emulator
  evidence**: `ui-trace record --do "tap 'Message'"` then `adb shell
  input text` shows the text in the bar with a caret
  (`/tmp/final-typing.png`) and logs `iris text render: chars=5 ...
  glyphs=5`, against `glyphs=0` per keystroke before.
  
  **2. The header drawn twice after a keyboard resize (this box's own
  "(a)") no longer has a path to happen on this emulator, for a
  measured reason**: since `MainActivity.java` went edge-to-edge
  (`e12c708`), **opening the keyboard no longer resizes the surface at
  all**. Measured: `render()` reports `out_size=(1080, 2282)` unchanged
  across an IME open, while the new `iris insets:` line reports
  `bottom=63 ime_bottom=0` -> `bottom=883 ime_bottom=1`. So the IME is
  an inset now, not a `surface_changed`, and the two-phase `Span::draw`
  the duplicate was blamed on is not re-entered. Reproduction attempts
  this pass, all negative: `tap 'Message'` + `ui-trace elements`
  (exactly one "Run benchmark" in every frame of the trace), a
  screenshot with the keyboard open, and a real `adb shell wm size
  1080x2200` *while the keyboard was open* (a genuine
  `surface_changed`) -- one header row, no stray copy
  (`/tmp/resize-dup.png`).
  
  **3. The `Compacted:` row drawn twice on Iris's phone is still
  open**, and nothing here reproduces it. What was ruled out this
  pass: the widget arena (`list.rs`'s
  `replacing_the_last_row_many_times_does_not_leak_primitives`), the
  `top_bar` rebuild (`last_top_pad`, a previous pass), and now the
  keyboard-resize trigger above. One real defect *was* found by
  reading the path and is fixed, though it cannot be shown to be her
  bug: `UiRenderState::draw_started` -- the guard whose whole job is
  "do not redraw a widget an ancestor is drawing right now, or one of
  the two copies is orphaned" -- **tested its own set after removing
  the id from it**, so the test was constant `false` and the guard
  could never fire, while the set grew by one entry per widget ever
  drawn and was never emptied. It is now inserted around
  `Widget::draw` and removed when it returns, with a `debug_assert!`
  at the top of `update` that it is empty between frames. Her build
  has both.
  

• Stale primitives, the phone's half — root-caused and fixed 2026-09-06, commit 76b1f99. It was neither Span::draw nor List: UiRenderState::draw_inner read needs_redraw without consuming it, and used it to skip the whole if let Some(active) block — including the remove(id, false) that frees a redrawn widget's previous primitives. So a widget that was both already active and marked dirty, and was reached by an ancestor's draw rather than by redraw_updates picking it first (the order a HashSet makes arbitrary, which is why it was intermittent), wrote a second full set of primitives and then had active.insert overwrite the only handles that could ever have freed the first set. Those instances stay in the layer's buffer for the life of the process, with a leaked move slot and leaked mask refs, redrawn every frame at whatever region they last had — and List sets no mask, so a row measured at GENEROUS_PADDING leaves its ghost outside the list's own box, which is the copy below the composer. Painter::draw_twice (List::place's measurement pass) reaches draw_inner twice for one id in one frame and so hits the same fault with no ancestor involved. Fix: consume the mark at the top of draw_inner — this call is the redraw it asked for — and free the old primitives on the dirty path too. Why the earlier passes could not see it: replacing_the_last_row_ many_times_does_not_leak_primitives counts widgets, and the orphan's owner is very much alive; it is an earlier set of that same widget's primitives that is stranded. Guard: UiRenderState::orphaned_primitives() names every live instance no ActiveData owns, and update debug_assert!s it empty every frame in debug builds. The per-frame form is a count comparison (primitive_counts_agree, O(active widgets)); the O(primitives) walk only runs to build the failure message, because running it per frame made a debug build on the emulator too slow to finish a bench run at all (260s timeout, no report). Test: an_ancestor_redrawing_a_dirty_row_leaves_no_stale_copy (iris/src/widget/list.rs), which fails on the pre-fix code with 1 primitive(s) survived their own widget's redraw. Emulator evidence, 2026-09-06 (this checkout's ai-app-2 AVD, build-apk.sh debug --abi x86_64 --features "transcript-screen bench force-gles", a debug build so the guard is live): a complete run-bench.sh run — 3,142 frames over 147s across the fling, the 400-event stream (which is 400 apply calls including the fixture's compaction event), the typing and the keyboard phases — with the assert firing zero times and logcat showing no abort. That is the whole of the phone's reported scenario exercised with the invariant checked on every frame.

• [~] Composer touch-drag scroll for overflowed text — the mechanism is done, the composer is not. Scroll::drag (iris/src/widget/position/scroll.rs) takes its pan from the same sense::DragGesture List is driven by, and WidgetLike::scrollable() registers it beside the wheel handler, so every .scrollable() in the codebase pans on a finger with nothing added at the call site. No fling (Scroll has no per-frame tick and the areas it wraps are at most a screenful) — see IRIS.md and DECISIONS.md. A vertical drag inside a focused field no longer extends a selection either (attr.rs's on_press now applies the same DRAG_SLOP rule its unfocused branch already did), which is Android EditText's own behaviour and what lets the scroll area around a field win the gesture. Tests: four in scroll.rs (pan past the slop, a tap inside it, a horizontal drag, the end clamp) plus a_finger_drag_over_a_scroll_area_pans_it in sense_tests.rs, which drives the whole path — scrollable()'s registration, run_sensors' dispatch, Scroll::drag, arbitration and pointer capture — and fails with got 0 if the registration is removed. What is left, with the measurement: wrapping the composer's field in .scrollable().masked() was tried and reverted the same day. Scroll resolves content_len/container_len against Painter::output_size — the whole 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 (iris text render: ... size=(1016.7, 623.7) against a 441px cap) the bar collapsed to its padding with no text in it. Making Scroll measure against its own offered box is the next step, and it touches a widget the transcript and the bench shell both use. One real bug was found and fixed on the way (ActiveData::mask stored the mask a widget set rather than the one it was drawn under, and redraw feeds that field straight back in as the inherited mask — so a targeted redraw of any Masked handed it its own mask and aborted on set_mask's nested-mask assert; that is a real abort on the emulator, assertion failed: self.mask == MaskIdx::NONE, reproduced as redrawing_a_masked_widget_does_not_nest_its_own_mask in layout_tests.rs).

• Streaming re-layout (IRIS_TODO.md's last section) — after the above, since they make the stream phase unrepresentative today.

• client-core prerequisites for P1, in parallel (pure Rust, disjoint from iris/), closed 2026-09-06: TranscriptSource's cache-vs-server stitching (new client-core/src/transcript_source.rs) and joinPages/healSplitMessage/adoptRun page-boundary healing (new functions in transcript_fold.rs), per CLIENT_CORE.md. Ported against the Kotlin source and AGENTS.md's paging incidents as the spec (TranscriptSource.kt/TranscriptItems.kt had no JVM unit tests of their own to port test-for-test). client-core goes from 85 to 109 tests; cargo test/clippy --all-targets/fmt all clean. Both AGENTS.md regressions have a dedicated test: loadOlderPage's before == 0 guard moved into TranscriptSource::page itself (paging_before_the_first_event_makes_no_request_at_all asserts zero transport calls, not just an empty result), and a_clean_boundary_between_two_finished_runs_is_still_healed_into_one_run pins adopt_run running on every join rather than only the split-call path. One incidental fix needed to port TranscriptSource faithfully: api.rs gained fetch_transcript_lines (additive, the existing fetch_transcript_page untouched since iris/ depends on its signature), which pairs each event with the exact server bytes it came from via serde_json::value::RawValue rather than re-serializing a parsed Value -- needed so the cache and a live SSE frame agree byte-for-byte, the same class of bug as the float_roundtrip fix. Deliberately not ported: EventStream.kt's reconnect/backoff and cross-thread stream cancellation, which are runtime policy for whichever framework embeds this crate, not pure logic -- see CLIENT_CORE.md's new section for the full account.

• Then: redeliver the APK for Iris. Delivery is a push to the ~/repos/ai-app-bench repo (iris/build/outputs/apk/release/ iris-bench-arm64.apk plus a dated README section), which Dev Updater on the host pulls -- not ~/host/bench/, which nothing reads; two builds on 2026-09-06 went there and never reached her phone. Record any choice she should see in DECISIONS.md.

Where things stand (2026-09-05)

• Streaming no longer costs a full rebuild (P0's box, "Streaming no longer costs a full rebuild" subsection): iris::widget::List:: replace_back/clear plus transcript_ui::TranscriptScreen::apply replace the "refold + rebuild the whole ~3,200-row tree per event" path in all three clients. Worst/p99 frame time in the streaming phase dropped roughly 3x on this checkout's emulator (see the box for the exact numbers and their caveats). Two new scripts, iris/android-app/build-apk.sh and iris/android-app/run-bench.sh, now do the build/install/tap/read-report cycle that used to be typed out by hand each time.
• The three items the dropout-fix pass left open are all closed, 2026-09-05 (the ai-server build break -- event_model:: Event::LimitReached -- was already fixed on rustify by the time this pass started). Three clean -gpu host cold-boot iris-scroll.sh runs all scrolled 24/24 swipes (checked directly via clustered render(): timestamps, not inferred from frame count), and the host-GPU table's iris row is now a best-of-three. EMU_GPU=software + force-gles still cannot produce a GLES number on this hardware, now for a third, structural reason found this pass: SwiftShader's ES 3.0 GL path reports zero storage-buffer capacity, and shader.wgsl reads four var<storage> buffers unconditionally -- reaching that path needs a shader rewrite, not a limits fix, so the SwiftShader-Vulkan-vs-GLES question is closed as unanswerable on this hardware rather than answered with a number. A fresh cold-boot run-bench.sh reading for P0's bench build (frames=690 janky%=62.03 p50=19.6ms worst=62.5ms) is in line with or better than the P0 box's three warm-AVD readings, so that box's "needs a clean cold boot" caveat is resolved too. Full account in I5's box, "The three remaining I5 verifications, closed 2026-09-05."
• The intermittent touch-scroll dropout is root-caused and fixed, 2026-09-05. Not the previously-suspected coalesced first ACTION_MOVE (ruled out) -- a gesture's ACTION_DOWN can land on a row's own padding/gap or its header, which CursorSense has no sensor over, so the widget that ends up handling the gesture only ever sees Pressing frames and DragArbiter never gets press_start, leaving it stuck in Idle (answers Undecided forever) for the rest of that gesture. Fixed in Selection::drag (iris/transcript-ui/src/ selection.rs) via a new DragArbiter::is_idle() the caller checks to recover a missed press on the next Pressing frame. Four new unit tests (three in iris/src/sense.rs's drag_arbiter_tests, one in transcript-ui's selection::tests, the latter failing on the pre-fix code). See this box's own "Touch-scroll dropout root-caused, 2026-09-05" subsection for the trace. The aggregate verification an earlier pass could not complete (peer-emulator interference) is now done, 2026-09-05: three separate cold--gpu host-boot iris-scroll.sh runs each scrolled all 24/24 swipes, confirmed by clustered render(): timestamps rather than frame count alone -- see this same subsection's "Update, 2026-09-05" paragraph.
• iris no longer requests compute-shader limits it never uses, 2026-09-05. adapter.request_device's Limits::default() asks for desktop-tier compute limits unconditionally even though nothing in iris/iris-core uses a ComputePipeline -- confirmed by grep, not assumed -- which is what crashed request_device outright under EMU_GPU=software's force-gles path (SwiftShader's GL reports OpenGL ES 3.0, no compute at all). New shared iris_core::device_limits() zeros exactly the six compute fields; rigs/gpu-probe's own mirrored limits were updated and confirm IRIS DEVICE: ok on this VM's own Vulkan and GL adapters. Verified on-device 2026-09-05: a cold EMU_GPU=software boot no longer aborts on the compute-limit request this fix targeted -- adapter selection now succeeds -- but device creation still aborts, on a different, unfixed limit (max_storage_buffer_binding_size, SwiftShader ES 3.0 has no SSBOs either); see this box's "The three remaining I5 verifications, closed 2026-09-05" subsection, item 2. See this box's "Fixed, 2026-09-05, later the same day" subsection (under the software-mode crash it fixes) and DECISIONS.md.
• Decided 2026-09-05: iris over Masonry, by Iris, from the host-GPU numbers in I5's box and E1/E2's findings. See the Recommendation's item 3 and DECISIONS.md. Next: the remaining screens and the app on iris — a new ordered list is the next thing to write into this file.
• The port plan exists, 2026-09-05: "## The port, in order (decided 2026-09-05)", seven steps (P1–P7) below "Experiments, in order," ordered by risk to the daily-use path rather than by screen count. P1 — session screen parity — is next. One crate decision made there: screens grow out of iris/transcript-ui into iris/app-ui, with desktop-app/android-app as thin entry points over it.
• I5 is now [x]: a clean, single-session, like-for-like 24-swipe scroll comparison between Compose and iris exists, 2026-09-05. Same sandbox session content for both apps, same emulator, EMU_GPU=software (a second pair under -gpu host not yet taken). Headline: Compose (debug build) 1102 in-app-reported frames, 99.0% late, p50 33.8ms/p90 50.6ms/p99 79.5ms; iris (release build -- debug SIGSEGVs on this emulator, I4's finding) FrameReport 299 frames, 94.65% janky, p50 79.1ms/p90 98.6ms/p99 117.8ms/worst 212.6ms (repeat run: 233 frames, 94.42%, p50 109.3ms). Not a clean apples-to-apples number: different build profiles (forced, not chosen), different jank definitions/frame populations across the three measurement sources, and both are emulator numbers under software rasterisation -- all stated plainly in I5's own box, "Clean scroll comparison, 2026-09-05," which also has the sampler timeline (load rose during the gesture but did not correlate with a failure this pass) and the dropout finding (this pass's own script bug -- cding into /tmp changed which emulator ui-trace targeted -- not a reproduction of the previously-suspected touch- delivery starvation). DECISIONS.md's DEFERRED item has this table's numbers for Iris to decide from; the iris-vs-Masonry choice itself is still hers to make, not decided here. I5's own box, "Update, 2026-09-05, later the same day" has the full account.
• The -gpu host pair this box's own DEFERRED item flagged as missing is now taken, 2026-09-05, and it changes the picture. Under real GPU rendering (--features force-gles, since the default Vulkan backend has no adapter under plain host-GPU boot -- confirmed by the exact crash message), iris's median frame (15.0ms, FrameReport) is faster than Compose's (20.0ms, in-app report) on the same session content, the opposite shape from the software-mode table. The new CPU/GPU split (FrameReport::record_split, iris/core/src/render/frame_report.rs, commit e2a1fad) shows why: iris's own redraw-to-submit work is a median 0.2ms; almost the whole frame is time handing off to the driver. Software-mode force-gles crashes for a third, distinct reason (SwiftShader's GL path reports itself as ES 3.0, which has no compute shaders, and iris's device request assumes them unconditionally), so this pass could not isolate SwiftShader-Vulkan as the sole cause of the software-mode gap. A real intermittent touch-scroll dropout was also reproduced and left unexplained (not the same as the earlier pass's script-bug dropout). I5's box, "Where iris's frame time goes, 2026-09-05, the -gpu host pass" has the full account, all four findings, and what verification did and did not re-run. DECISIONS.md's DEFERRED item has the updated table.
• I5's Android integration is done and measured, 2026-09-05. The transcript screen runs on-device against a real ai-server, with real scrolling, real touch-drag panning and tap-by-name accessibility all confirmed by screenshot/log evidence on this checkout's emulator. Two real, previously-unknown bugs were found and fixed getting here (a missing INTERNET permission, and a background-thread redraw request that crashed the process via a Looper requirement neither this box nor Tasks::redraw_handle's design had anticipated) -- both in I5's own box, both in IRIS.md.
• Design choices for the two pieces before this are summarised in DECISIONS.md at the repo root, which is the file Iris reads for choices made without her.
• E4 done, 2026-09-05. iris/desktop-app: a winit window with a session list beside transcript-ui's screen (build_tree), against a real ai-server through client-core, enrolled from the same aiapp://enroll?... link a phone scans. Both pass conditions held on app/ui-sandbox.sh -- see E4's own box for the commands, the screenshot, and a real streaming-duplication bug the screenshot found and a regression test now covers.
• The I5 touch-drag pan-vs-select gap is closed, 2026-09-05, as a DragArbiter in iris/src/sense.rs wired into transcript-ui's selection -- see I5's own box below, "Gap closed, 2026-09-05".
• Done: E0 (toolchain), E1 (Masonry on android-view, which found the keyboard gap — now explained, see below), E2 (a transcript in Masonry, which found that Masonry has no touch-scroll on Android at all — see below), E3 (the Kotlin/Java shell over a JNI bridge into Rust, both pass conditions proved on the emulator — see its own box), E5 (the Gradle-free packaging xtask, both pass conditions proved — see its own box), I0a, I0b (iris builds on a pinned nightly and runs), I1 (parley + glyph atlas), I2 (iris on android-view), I3 (iris::widget::List), I4 (host half).
• E5 done, 2026-09-05. cargo xtask apk (new xtask/ crate at the repo root, zero dependencies) replaces Gradle for packaging app/shellApp: cargo ndk → javac/d8 → aapt2 → zipalign → apksigner, signed with the same key app/build-apk.sh uses. Both pass conditions held on this checkout's emulator: adb install -r over the Gradle-built shellApp succeeded (same key, so the signatures matched), and the notification service reached its follow-loop and posted a real notification while the app was backgrounded. E3's open kotlinc question resolved itself as a side effect of the one Gradle call still needed for AndroidX dependency resolution — see E5's own box for the full account, including the one disclosed place Gradle still runs and what was deliberately left undone (a real-device arm64-v8a install, dex shrinking).
• I5 — the transcript screen in iris: partial, 2026-09-05 (ticked [~] in its own box, not [x]). iris/transcript-ui/ builds a real transcript screen — markdown-folded rows in iris::widget::List, cross-row selection, a growing composer, tool-row expand-hold — on top of a new, genuinely useful iris capability this box added: SpanStyle, per-range text styling (core/src/primitive/text.rs), which is what lets one wrapped, selectable TextEdit carry a heading, bold, italic, inline code and a link all inside the same paragraph — exactly the inline-rich-text ceiling E2 found Masonry structurally unable to cross. Screenshotted via run-headless.sh (real inline styling visible, not just block-level). 9 new tests, all passing; cargo build/clippy/fmt/test --workspace and cargo ndk (both iris and transcript-ui) all clean. What did not happen this pass: any Android integration for this specific screen (no cdylib/Gradle shell exists for it yet, unlike tabs-ui's iris-android-app), and therefore the emulator-side pass condition (transcript-bench.sh against the Compose baseline, ui-trace tap-by-name on a row) — emu list showed the one emulator here held by another session, but the real blocker is that the integration work itself is unbuilt, not the emulator being busy. Full accounting, every citation, and the dated IRIS_TODO.md items are in I5's own box below. Update, 2026-09-05, same day: touch-drag panning over a row's own rendered text, which was not yet reachable for a specific, diagnosed reason (it competed with this box's own row-level drag-select for the same gesture, not an absent primitive), is now closed — a DragArbiter in iris/src/sense.rs, wired into transcript-ui's selection — see the box's "Gap closed" note. Android integration is the one item left before this box can tick [x].
• E3 done, 2026-09-05, and unlike E1/E2 it is committed to this repo (android-shell/ — a JNI-bridge crate on client-core — plus a new Gradle module app/shellApp/, left deliberately separate from app/androidApp so its ~13,000 lines of working Compose UI are untouched). Both pass conditions held: a notification arrived in Android's drawer while the app was closed, and a shared text share landed as a real userMessage in a sandbox session's transcript. Found and fixed three real bugs along the way — a generic JObject native parameter silently exporting the wrong JNI signature (UnsatisfiedLinkError), a class-by-name lookup failing from this crate's own background thread because a Rust-attached thread has no app ClassLoader (Error::NoClassDefFound, invisible without a logger installed), and onStartCommand opening two /notifications connections per enrollment — the last one a latent bug in Notifications.kt itself, found here rather than there. See E3's own box for the full account, the exact commands, and what was deliberately cut (attachment uploads, a session picker, the on-screen/banner suppression — all pending E4's screen).
• I4 — accessibility names via AccessKit: host half done and verified 2026-09-05, ticked in the box below. iris_core::ui::access::AccessTree builds one flat AccessKit tree from Widgets::named() (a side set only .label() populates, so an unnamed widget costs this nothing), pushed through accesskit_winit on the desktop and accesskit_android on Android, updated only when a name/role/bounds actually changes (a counter confirms it: 1 rebuild on first draw, 0 across an unchanged frame, 1 more after a real move). E1's detach-abort mitigation is carried (android/access.rs's raise_if_enabled). Every check that doesn't need the emulator is clean — see I4's own box for the exact numbers. What's left: the emulator itself is held by another session this pass, so ui-trace record --do "tap 'pad'" against iris-android-app's tabs screen (which now has five named buttons) has not been run for real yet — exact commands at the bottom of I4's box.
• E2 done, 2026-09-05, and its headline finding changes what "decide from the measurements" (recommendation item 3) can mean right now. Built a real transcript screen (~/src/android-view/e2-transcript, local, not committed — see E2's own box), fetching 854 real events from an app/ui-sandbox.sh session through client-core. Six of the seven "hard to get back" behaviours are answered with evidence either way; the seventh (measurable frames) is blocked before it can even start: neither of Masonry's scrolling widgets (VirtualScroll, Portal) reacts to a touch drag, only to a wheel-style PointerEvent::Scroll — confirmed by reading (virtual_scroll.rs:504-523, portal.rs:259-267) and empirically (a real swipe and a synthetic Android scroll event both moved nothing on screen). So transcript-bench.sh's own gesture cannot be performed against a Masonry transcript on Android today, which means the render-number half of E2's pass condition has no comparison to make yet — not a bad number, no number obtainable at all. Selection spanning rows and per-span rich text (bold/italic/inline code/links inside one paragraph) are also confirmed not possible on the pinned commit, each for a specific, cited reason. What did work: block-level rich text (heading size, monospace fences), real virtualisation of 854 rows, overwrite_anchor-based hold-top-edge on expand (screenshotted), and tap-by-name accessibility. Full writeup, every citation, and the exact repro commands are in E2's own box below.
• Done, 2026-09-04: the Widget::draw/layout redesign (LAYOUT.md). desired_width/desired_height/SizeCtx/Cache are gone; every widget in iris/src/widget/ implements one fn draw(&mut self, &mut Painter) -> Size. A moved widget (Scroll, Offset) now costs one move_offsets write resolved by a shared resolve_move WGSL function in both shader stages, independent of how many primitives are in its subtree — measured at 500 in iris/src/layout_tests.rs, which also covers the unchanged-frame, hit-test-after-move and mask-follows-move pass conditions as plain unit tests (no GPU or window needed, since UiRenderState touches neither). All four examples render pixel-identically to before the change. See LAYOUT.md's "Deviations found during implementation" for five real bugs the design's first draft did not anticipate — worth reading before touching Aligned, Sized, MaxSize, Scroll, or the move-slot lifecycle again. GpuTextures::grow_array (a second atlas layer opening) has now been exercised too, on tabs with PAGE temporarily lowered — see TEXTURES.md's "Exercised, 2026-09-04". Not done: a pixel-level screenshot check of a Masked-wrapped Scroll (no example builds one yet — the numeric check in layout_tests.rs stands in).
• E1's keyboard gap is Masonry's as_input_connection returning None (a TODO), not android-view or EditorInfo. android-view's own demo implements the InputConnection trait over a parley editor and gets real Gboard suggestions on this emulator — screenshotted 2026-09-04. android-view's accesskit_android adapter also has a reproducible abort (a client detaching, not attaching, is the trigger) — see E1 below for both, with the mitigation iris/I4 needs to carry.
• Resolved, 2026-09-04: iris's binding array does not survive real Android hardware. iris's texture pipeline used to ask every device, unconditionally, for VK_EXT_descriptor_indexing ("bindless" binding arrays), which a real share of Android hardware lacks. It has been rebuilt per TEXTURES.md's "Recommended shape": the glyph atlas is one texture_2d_array (a layer per page), a standalone image is its own ordinary Texture/BindGroup, and request_device now asks for no features and no binding-array limits at all. rigs/gpu-probe, rewritten to match, confirms request_device now succeeds on the emulator's software Vulkan (EMU_GPU=software, SwiftShader) — see TEXTURES.md's "Implemented, 2026-09-04" for the exact command and output, and for what was verified (rendering, via run-headless.sh) versus what was reasoned through but not separately stress-tested (a real second-atlas-page grow under load). Nothing here has been run on real Android hardware yet, only the emulator; the Android Vulkan Profile 2025 sourcing in "iris's binding array does not survive real Android hardware" below is what stands in for that until I2 gets a device.
• I2 — iris on android-view: done 2026-09-05. The android-view backend (iris/src/android/), the iris-android-app cdylib and Gradle shell, insets, the back gesture, and the full InputConnection bridge are all in and measured working — Gboard's suggestion strip reads real buffer content through it, the same bar E1 set. The render gap (nothing drew but the clear colour) is fixed: UiRenderNode::new seeded the GPU's window uniform from WindowUniform::default() (0, 0) rather than the surface's real size, so the vertex shader's / window.dim produced NaN/Inf clip positions on every primitive, on both Vulkan and GLES — winit's backend never hit this because winit fires an initial WindowEvent::Resized that corrects it before the first frame, and android-view has no equivalent event. Fixed by seeding the uniform from config.width/height at construction instead of depending on a later resize call. The tabs example now renders on the emulator on both backends (screenshotted); the GLES-only D2/D2Array warning was confirmed a red herring — still present post-fix, harmless. See I2's own entry below for the full writeup. E2 (a transcript in Masonry) is done — see its own box.
• I3 — iris::widget::List built and benchmarked 2026-09-05, ticked in the box below. Variable-height rows, virtualised, moved not relaid-out on scroll, insert-above-anchor and expand-hold both measured flat across N = 100/1,000/10,000. What is left is wiring it into an actual transcript screen and comparing against transcript-bench.sh's Compose baseline on the GPU emulator, which needs a session/scroll model around it (closer to I5's scope) — see I3's own box for the exact command once that screen exists. Read list.rs's module doc and IRIS.md's 2026-09-05 entry before touching it: a widget that fills whatever region it's offered (a Rect background) cannot be measured at a throwaway region and merely repositioned, a lesson that generalises beyond this one widget.
• client-core built (2026-09-04), item 1 of the recommendation: event-model/ (the event types, now shared with server/) and client-core/ (REST and SSE clients, transcript fold, cache, highlighter, ANSI parser, 85 ported tests). CLIENT_CORE.md maps Kotlin file to Rust module and lists what is not yet covered. ./run-tests.sh runs all three crates.
• The app itself is untouched. Everything so far is in iris/, in rigs/gpu-probe (a headless wgpu/Vulkan feature probe, pushable to a device with no APK — see the binding-array section), and in the other rigs; nothing under app/ or server/ has changed.
• Changed outside this repo, both in emulator-tools and both pushed: avd_serial now validates its cache by asking the device its AVD name rather than by checking the serial is still attached (a recycled port silently pointed this checkout at another session's emulator), and EMU_GPU=software was added as an opt-in that keeps a run off the host GPU and gives the guest a software Vulkan device. The default is unchanged, because -gpu host was measured and the Compose benchmarks depend on it.

What has to be reproduced

The app is ~19,000 lines of Kotlin. It splits three ways, and the split is what decides how much of a port is mechanical.

Pure logic with no Compose or Android in it, ~4,500 lines. Api.kt (1,142), Events.kt, EventStream.kt, Sse.kt, TranscriptCache.kt (589, touches java.io.File only), TranscriptSource.kt, MarkdownSyntax.kt, Languages.kt, Highlighter.kt, Ansi.kt, ResetCountdown.kt, Durations.kt, Sizes.kt, ModelName.kt, LoadState.kt, ImportableStream.kt. TranscriptUnits.kt and TranscriptItems.kt (the event fold into rows, ~940 lines) are logic with a handful of Compose annotations. This is also exactly the code that has JVM unit tests today. All of it ports directly, and most of it already has a Rust twin in server/: Events.kt is a hand-kept mirror of session/driver.rs's enum, the highlighter and the syntax scanner exist on the server for the explorer, and the cache compares the server's own JSON lines. Sharing these types between server and app is the single largest "keep things in sync" win available, and it does not depend on which UI framework wins.

Compose UI, ~13,000 lines. Screens, dialogs, the transcript list, the markdown renderer's customisations, tool cards, the file explorer viewer and editor. This is the part a UI framework choice is about.

Android platform code, ~1,500 lines, spread over 20 files. Every one of these is a Java-side object that no Rust framework can replace, because Android only offers them as Java classes:

• NotificationService — a foreground service holding the /notifications SSE stream while the app is closed, with its ongoing notification, specialUse type and the POST_NOTIFICATIONS request.
• MainActivity — edge-to-edge, the ACCESS_LOCAL_NETWORK runtime permission (Android 17), singleTop intent routing for aiapp://enroll, notification taps, and the share sheet (ACTION_SEND, any MIME type).
• ServerConfig — the bearer token sealed under an Android Keystore AES-GCM key, shared with Dev Updater through wg-app-link's :link.
• EnrollmentScanActivity — the in-app QR scanner (zxing, camera).
• Attachments — ContentResolver reads of shared URIs, BitmapFactory decode and downscale, EXIF orientation.
• SessionImage — bitmap decode for produced images.
• ScrollAnchor, Drafts — SharedPreferences; CrashLog — filesDir.
• TranscriptCache — cacheDir.
• DebugStats/FrameStats — Choreographer frame timing and the render report; runtime-tracing names composables in a system trace.

So "pure Rust" on Android means Rust owns every line of logic and drawing, behind a thin shell of Java stubs, and a packaging step that produces a signed APK. How thin, and whether Gradle is inevitable, are answered below.

How much Java is unavoidable, and why

Rust can call any Android API through JNI (jni crate, with ndk-context handing over the JavaVM and the Activity): posting a notification, startForegroundService, the Keystore, ContentResolver reads, permission requests, WindowInsets, the clipboard. None of that needs a line of Kotlin. What JNI cannot do is define a class that the system instantiates by name from the manifest — an Activity, a Service, an Application, a BroadcastReceiver. Those must exist as dex bytecode inside the APK before any Rust runs, because the framework constructs them and only then calls into native code. NativeActivity is the platform's own stub for the Activity case; there is no NativeService, and android-view ships its own View subclass for the same reason.

So the floor is roughly two Java classes of ten lines each: an Activity and a Service whose lifecycle methods are declared native and registered from JNI_OnLoad, plus whatever android-view already provides. Everything they would have done in Kotlin — insets, intent routing, the SSE follow loop, the notification builder — is Rust reached through those stubs. Writing the stubs in Java rather than Kotlin drops kotlinc from the toolchain; javac comes with the JDK Gradle already needs. Generating the dex from Rust is not worth it: there is no mature Rust dex writer, and the stubs never change.

Can the APK be built without Gradle?

Yes. An APK is a zip containing a binary-XML AndroidManifest.xml, resources.arsc, classes.dex, lib/<abi>/*.so and assets, aligned and signed with the v2 scheme. The tools are aapt2 (manifest and resources), d8 (Java bytecode to dex), zipalign and apksigner, all in the SDK's build-tools, none of them Gradle. Three ways to drive them:

• A cargo xtask (or build.rs-adjacent script) that runs cargo ndk for each ABI, javac + d8 for the stubs, aapt2 link, zipalign, apksigner. About 150 lines, every step visible, no AGP, no Gradle daemon holding 2.8 GB between builds. The pinned-CA constant becomes a build.rs reading the same certs/ca.pem path.
• cargo-apk2: the maintained successor to cargo-apk, and unlike it compiles java_sources / kotlin_sources into the dex and declares multiple activities and services with intent filters from [package.metadata.android], with per-profile keystores and optional aapt2. Exactly the shape needed; the question is whether a third-party tool with one maintainer beats 150 lines we own.
• cargo-apk / xbuild: unmaintained and NativeActivity-only. No.

What Gradle would take with it: Android Lint (which found two real bugs here, but in Kotlin that would no longer exist — with forty lines of Java stubs there is little left for it to find), manifest merging, R8, and the generated-source plumbing. What it gives back: one toolchain, cargo end to end, and Dev Updater keeps calling build-apk.sh exactly as now. Recommendation: the xtask, with cargo-apk2 read for the details it already got right (v2 signing, uses-feature, ABI splits).

The behaviours that are hard to get back

Reading the Compose code for what a replacement must be able to express, rather than what it happens to look like:

1. The transcript is one selectable body of text. One SelectionContainer around the whole lazy list, so a selection runs from a reply into the tool output beneath it. The framework needs selectable read-only rich text across many rows, with the platform's selection handles and clipboard on the phone.
2. Rich inline text: markdown with links (one tap detector per text, not a node per link), inline code chips drawn behind the text, tables with wrapping cells and a sideways scroll, syntax-highlighted fences, ANSI colour in tool output, Nerd Font icon glyphs. Needs a text layout engine with spans, not just styled labels.
3. A bottom-anchored virtualised list of variable-height rows, paged in both directions (800-event pages, HISTORY_SCREENS measured in viewports), with a saved scroll anchor per session, "hold the edge nearest the tap" when a row expands (holdTopEdge, done in the layout pass so the wrong frame is never drawn), and rows keyed so that a run of tool calls stays one row while it grows.
4. The soft keyboard: the composer resizes with the IME, the guard against a stuck inset animation, drafts per session, autocorrect and suggestions from the phone's own keyboard. This is where most Rust frameworks fail on Android today; see below.
5. Platform integration through the app model: foreground service, notifications, share sheet, deep link, Keystore, camera, back gesture, edge-to-edge insets, local-network permission.
6. Accessibility names on icon buttons, which the bench scripts depend on (ui-trace taps by label). A framework with no accessibility tree also breaks the measuring rig.
7. Measurable frames: the debug render report, and a way to attribute a frame's cost to a widget on the real phone.

The two constraints that decide it

1. Android text input. Every framework built on winit inherits winit's Android backend, and that backend cannot drive the soft keyboard properly: the IME tracking issues (#1823, #2766) are open, ReceivedCharacter is unimplemented on Android (#2305), and the android-activity groundwork for editor actions only merged in February 2026 (PR #214) with the winit half still to come. Composition, autocorrect and suggestions need an InputConnection implemented on the Java side, which winit's NativeActivity/GameActivity model does not offer. The frameworks that type on Android today each wrote their own Java glue (Slint, Makepad), and the one designed to do it the way Android intends is android-view: a Rust implementation of an Android View, with text input through InputConnection, accessibility, touch, callbacks on the UI thread, usable either as a whole app or embedded beside ordinary Android components. It is marked WIP. Both Linebender (its Masonry demo lives in that repo) and Robius/Makepad (Robrix's release notes say Android lacks a "full" keyboard and they are integrating android-view for it) are converging on it. That makes android-view the phone-side foundation whichever widget set sits on top, and the first thing to build and measure here.

2. Rich, selectable text and a virtualised list. Frameworks group by their text stack:

• Parley + Fontique + Vello (Linebender): rich spans, selection and editing utilities, IME support driven through ui-events, AccessKit text properties (Linebender 2026 Q1, parley). Used by Masonry/Xilem, and by Blitz. Vello proper needs compute shaders; vello_hybrid (CPU path processing, GPU compositing) is "roughly beta" and runs on GLES too, and Vello CPU exists as a no-GPU fallback.
• cosmic-text (iced, egui optionally): good layout, but the widgets on top decide selection. iced's markdown widget is not selectable (discourse).
• Slint's own: TextInput with read-only is the selectable-text trick; there is no inline rich text at all (issue #1325, markdown request #6684 both open). A markdown transcript with links and code chips cannot be drawn.
• Makepad's own: GPU/SDF text, a Markdown widget and a virtualised PortalList in makepad-widgets.

Options

A. Keep Compose, move the logic into a Rust core (uniffi)

A client-core crate (events shared with the server, API client, SSE, transcript fold, cache, markdown model, highlighter, ANSI) exposed to Kotlin through uniffi. Compose keeps drawing. Desktop would be a second UI (iced or Compose Desktop) over the same core.

• For: the logic and the wire types stop drifting from the server today, with tests in one language. Incremental and always shippable.
• Against: it is not what was asked for. The 13,000 lines of UI stay Kotlin, the desktop app shares no UI code, and the :link Kotlin module stays. uniffi's Kotlin Multiplatform bindings are a community fork; the Android-only bindings are Mozilla's and solid.
• Verdict: not the destination, but step one of every other option is building this crate, so it costs nothing to keep it as the fallback.

B. Slint

Rust on Android is officially supported (minSdk 26, android-activity backend, own Java IME glue, safe areas and keyboard insets since 1.15, Skia renderer needs clang). Royalty-free licence requires disclosing Slint use; GPLv3 otherwise. UI is a separate .slint DSL, not Rust.

• Against: no rich inline text (see above), so the transcript cannot be drawn as it is today; the UI language is not Rust, which forfeits the "compiler catches it" motivation for the half of the code that is UI.
• Verdict: rejected on rich text alone.

C. iced

Elm-style, Rust-only widgets, desktop-first, winit + wgpu. Has a markdown widget and rich_text with links. The maintainer states mobile is a non-goal (iced); a community Android example exists and its author could not get the soft keyboard working, patched widgets for touch, and notes no accessibility (HN thread). Markdown is not selectable; scrollable is not virtualised.

• Verdict: a fine desktop toolkit and the one Iris named, but every phone-side gap (IME, touch, accessibility, selection, virtualisation) would be ours to build and maintain against a project that does not want them. Not the shared framework.

D. egui

Immediate mode, winit-based on Android, AccessKit integration, selectable labels across a Ui. Repaints only on input by default, so battery is not the immediate-mode worry. Android IME is blocked on winit (discussion); the workaround is an in-app virtual keyboard, which is exactly the non-native keyboard to avoid. Variable-height virtualised lists are manual (show_rows assumes uniform heights). Looks like egui, not Material.

• Verdict: workable on desktop, wrong on the phone for the same reason as iced, plus a look that would need a full custom style.

E. Makepad

GPU-rendered, hybrid retained/immediate, live_design! DSL with hot reload, MIT, 1.0 in 2025 (makepad). Ships Android apps today with its own Java glue; Robrix (a Matrix chat client, the closest analogue to this app) is its reference application on Android, iOS and desktop. Has Markdown, PortalList (virtualised), TextInput. Robrix reports the Android keyboard is not "full" and is moving to android-view for it; the README says non-standard targets "may require minor fixes".

• For: the only option that already ships a chat-shaped app on Android and desktop from one codebase, with the widgets this app needs.
• Against: the DSL is its own language with its own shader-based styling, so a large part of the UI would not be checked by rustc; the rendering model (SDF everything) is a different world from Compose's, and selection across a Markdown widget is unverified.
• Verdict: rejected 2026-09-04 — Iris does not want a DSL. Kept here so its Android keyboard status stays a data point about android-view, not as an option.

F. Masonry / Xilem on android-view (Linebender)

Retained widget tree (Masonry) with a reactive view layer (Xilem) that reads like Compose; Rust all the way down; Vello, Parley, Fontique, AccessKit, ui-events. Widgets include Prose (selectable read-only rich text), TextArea, VirtualScroll, and this year Svg, Split, CollapsePanel, a new layout system, and IME through ui-events independent of winit. masonry_android_view exists in the android-view repo and is "not yet generally usable"; Xilem calls itself experimental. Desktop runs on winit. Vello needs a compute-capable GPU or falls back to vello_hybrid/CPU.

• For: the only stack where every hard behaviour above maps onto a component designed for it: selection and rich text (Parley/Prose), virtualised variable heights (VirtualScroll), native IME (android-view's InputConnection), accessibility (AccessKit, now with an Android crate), one Rust widget language on both platforms. The team is the one writing the Android integration everyone else is adopting.
• Against: pre-1.0 with API churn each release; a small team; no Material widget set, so every control's look is ours; some of the pieces (masonry_android_view, vello_hybrid) are explicitly unfinished. Being early means fixing things upstream ourselves, which Iris said is acceptable.
• Verdict: the option to try first, because it is the only one whose gaps are "not finished yet" rather than "not designed for this".

G. iris — the in-house library, and what "from scratch" means here

cat16/iris, read 2026-09-04 from the one public commit (2026-01-31, "portfolio copy"; ~8,700 lines in core, macro and the crate itself). Retained-mode widgets stored outside the render tree, wgpu 28 directly, winit 0.30, cosmic-text 0.16 (parley since I1), a relative-anchor-plus-offset layout with rest() and rel() lengths, a postfix builder API (rect(..).radius(30).on(CursorSense::click(), ..) .sized(..).align(..)), events handled where the widget is declared, and a single-threaded context passed explicitly — all of which reads like this codebase's own rules. There is text editing (widget/text/edit.rs), images, masks, spans and stacks; the TODO names text resizing as per-frame slow and scaling as unsolved. It requires nightly (fourteen #![feature] gates as vendored, among them const_trait_impl, unboxed_closures, portable_simd, associated_type_defaults; eleven after I0b and I1 — see those steps for the current list). Desktop only; no Android surface, no IME, no accessibility tree, no virtualised list, no rich-text selection.

That list is the work, and some of it is done. As of 2026-09-04 it builds on a pinned nightly, runs on this machine's GPU, has parley and a glyph atlas, and iris-core cross-compiles to Android. What it still lacks from the list above is the Android surface, the IME bridge, the accessibility tree and the virtualised list — I2, I3 and I4.

iris is not a candidate to be tested as it stands. It is the in-house library (Iris, 2026-09-04): "essentially a good start to a rewrite from scratch", to be maintained and extended by the sessions working here. So the list above of what it lacks is a work list, not a score. When the app needs something iris does not have, the answer is to build it into iris. The layer iris has is the widget and layout layer; the layers it needs are the same ones Masonry gets from android-view, Parley and AccessKit, and there is no reason iris cannot sit on those same foundations rather than reinvent them — the surface, the keyboard bridge and the accessibility tree are platform plumbing, not a framework's identity. The text stack was the first real design decision in that work, and it is settled: Parley, with a glyph atlas (I1, 2026-09-04).

Two things to carry into that work honestly. Nightly is the opposite of "holds up long term": a build that breaks on a toolchain update, on the machine Dev Updater builds on, unattended. Done in I0b — iris/rust-toolchain.toml pins nightly-2026-09-03 — and the gate list lives with I0b and I1, to be retired as they stabilise or are designed around; it is down from fourteen to eleven. And a one-person framework carries every gap itself, which is what Iris said she is willing to do.

"From scratch" therefore means iris, not a fourth thing. Masonry stays in the plan as the yardstick and the fallback: building its demo and its version of the transcript screen first says what a finished stack costs on this hardware, proves android-view before iris depends on it, and gives a comparison that is measured rather than remembered.

Not considered further: GPUI (Zed) mobile is a community fork that depends on unpublished crates; Dioxus/Blitz is excluded by Iris (its native renderer is Parley/Vello under HTML semantics, and the earlier tdep-survey/app-dioxus spike parked it on a vello_hybrid stroke bug and shipped the WebView); Compose Multiplatform Desktop would give a desktop app for nothing but in Kotlin, which is the opposite direction.

Weight and debug builds

Iris remembers the Linebender stack being slow in debug. What is behind that is the dependency graph — Vello, wgpu, Parley, Fontique, Skrifa — running unoptimised on the CPU side (path encoding, shaping), not the widget layer. Xilem's own advice is only split-debuginfo = "unpacked" to keep target/ small; the fix everyone with this shape of dependency tree uses is to optimise dependencies while leaving the app crate at opt-level = 0:

[profile.dev.package."*"]
opt-level = 2

Measured 2026-09-04, and it is not the widget layer. iris's own graph (wgpu + winit + cosmic-text at the time) built cold in 43s with a 2.1 GB target/, and 1m46s with a 1.5 GB target/ under the profile above — so the knob costs build time and saves disk here, and plain debug was never the problem. Masonry's graph is the one with Vello, Parley, Fontique and Skrifa in it, and E1 gives the number that matters for it: libmain.so is 181 MB in debug and 11 MB in release. That size is also a correctness issue rather than only a weight one — a debug build labels its Vulkan objects, and the emulator's driver segfaults in SetDebugUtilsObjectNameEXT when it does. Runtime cost of the profile knob is still unmeasured; resident memory and the frame cost of an 800-event page want E2. Vello proper needs compute shaders and carries a large shader set; vello_hybrid is lighter and Masonry can now render through either (or Vello CPU) via its imaging abstraction, so "keep it light" has a knob inside the same stack.

Recommendation

1. Build client-core now, whatever the framework (done 2026-09-04, see CLIENT_CORE.md). A Rust crate holding the event model (shared with server/ as one crate, ending the Events.kt mirror), the API and SSE clients, the transcript fold, the cache, the markdown block model, the highlighter and the ANSI parser, with the existing JVM tests ported. It is the part of the app that is already tested, already logic, and already duplicated on the server.

2. One foundation, two widget layers. The platform plumbing is shared whichever way the decision goes: android-view for the Android surface, keyboard and accessibility bridge; wgpu for the GPU; AccessKit for names; winit on the desktop. On top of it, Masonry as the yardstick (E1, E2) and iris as the thing being built (I0–I5), both aimed at the same transcript screen with the same pass conditions.

3. Decided, 2026-09-05: iris. Iris made the call from the host-GPU comparison in I5's box (iris p50 15.0 ms, Compose 20.0 ms, same content, same emulator) and from what E1/E2 found Masonry cannot do on Android today (touch scroll, per-span rich text, cross-row selection, the keyboard bridge). DECISIONS.md has the entry. The paragraph below is what the decision was to be made from, kept for the record. Decide when the transcript screen exists in both, from the measurements, and record the decision here with the numbers. If iris carries the screen within the Compose baseline, it is the app's framework and Masonry was the calibration. If it does not, the measurement says which parts of Masonry to adopt underneath it.

   Still not decidable by a render-time number, 2026-09-05 (updated) — what's missing, named rather than guessed at, and now for a different reason than before. E2 found Masonry's own scroll gesture path absent on Android entirely (its box, "measurable frames") — that has not changed. I5's Android integration is now built and confirmed working (real server, real scrolling, real touch-drag pan, tap-by-name — I5's own box, "Measurements taken"), so the earlier blocker ("no cdylib/Gradle shell exists for this screen") is gone. What replaced it: dumpsys gfxinfo, the tool transcript-bench.sh and this recommendation both assumed would give the comparison, cannot see a SurfaceView's own GPU-drawn frames at all — it instruments Android's ordinary View/Skia drawing pipeline, which a wgpu-rendered SurfaceView (iris's whole approach) bypasses entirely. Confirmed 0 frames reported across a 24-swipe gesture loop that visibly scrolled the screen (screenshots differ), and a dumpsys SurfaceFlinger --latency fallback returned no per-frame history either (just the display's refresh period) on this Android version's BLAST compositor. The Compose side of the same loop did produce a real number under identical conditions (EMU_GPU=software, same emulator, same session): 8.96% janky frames, 99th percentile 150ms. So this is now a one-sided number, not a missing one — the number needed to close item 3 is a render-time report from iris itself (the equivalent of the Compose app's in-app copy-button report transcript-bench.sh already reads), which does not exist yet and is real, scoped follow-on work (frame timing inside iris_core::render, exposed the way AccessTree or UiRenderState::take_counters already are) rather than a rerun of anything above. Until it exists, the decision still rests on the structural findings both sides did produce, now joined by a functional one: Masonry cannot do cross-row selection or per-span inline rich text at all today (E2's grep -rln, zero hits, cited in its own box); iris does both (I5's SpanStyle and selection.rs) and its touch-scroll now works end-to-end on a real device, not just programmatically (I3's benchmark plus I5's on-device screenshot evidence) — three structural points and one functional one in iris's favour, still with no opposing or supporting render-time measurement on either side.

   Update, 2026-09-05, later the same day: iris now has a render-time report of its own, and a real number from it, but not yet the clean comparison item 3 needs. iris_core::FrameReport (new, iris/core/src/render/frame_report.rs) is exactly the follow-on work named above — a per-frame wall-time ring exposed as two named on-screen controls, unit tested (6 tests over the ring/percentile math). Driven for real against a real touch-drag on this checkout's emulator, it read frames=34 janky%=61.76 p50=26.5ms p90=48.0ms p99=98.1ms worst=98.1ms — a genuine measurement through iris's own render path, not inferred. It is not yet the comparable number, for a newly found and separately named reason (I5's own box, "Update, 2026-09-05, later the same day"): gestures against this checkout's EMU_GPU=software emulator intermittently delivered zero touch input during this pass — reproducible, but not yet root-caused past one candidate (the emulator's own software rasterisation measured at ~78% of a CPU core continuously, a plausible source of input backlog, not yet confirmed with a sampler running during a failing gesture). So item 3 still cannot be closed by a clean number, now for a narrower and more tractable reason than before: the instrumentation exists and works, and what remains is making the emulator rig deliver touch input reliably enough to run the comparable loop. What Iris needs to weigh, updated: whether "iris works, Masonry's Android scroll path is absent entirely, and iris's own frame-timing report is real and working" is enough to decide without the final clean number, or whether to wait for the touch-delivery investigation above — still a product/tradeoff call, left to her (DECISIONS.md's DEFERRED item, updated with this session's numbers).

4. Then the shell (E3), the desktop window (E4) and the packaging (E5), which do not depend on the choice.

Experiments, in order

Each has a pass condition that is a measurement in this clone. The rig matters: this emulator runs -gpu host with host Vulkan switched off (GPU_HOST_FEATURES in emulator-tools, a gfxstream/Venus gap), so inside the guest a wgpu app gets GLES, not Vulkan; the earlier Dioxus spike also needed WGPU_GLES_MINOR_VERSION=1 for compute shaders and found wgpu's Android backend wants API 26 (a libc symbol). The real phone has Vulkan. Per the standing rule, a rig limit is something to fix before it is accepted.

• E0 — toolchain (done 2026-09-04). Installed under the user-owned SDK: NDK r29 (29.0.14206865), 2.4 GB at ~/Android/Sdk/ndk/29.0.14206865, the newest stable — r30 is still at rc.3. cargo-ndk 4.1.2. Verified by cross-compiling a scratch cdylib to both ABIs: file reports "for Android 26, built by NDK r29 (14206865)" for aarch64-linux-android and x86_64-linux-android. Two things to know at the call site. cargo-ndk 4's API-level flag is -P, not -p — -p is now passed through to cargo as --package, so the old cargo ndk -t arm64-v8a -p 26 panics with unknown package: 26 and dumps the whole environment to stdout as a bug report, which is worth not doing in a log somebody might paste. And the Android targets were installed for stable only; the pinned nightly needs its own, which iris/rust-toolchain.toml now declares.

• E1 — android-view's Masonry demo on this emulator (2026-09-04). It builds, renders on the GPU through Vulkan, exposes its accessibility tree, and the phone's own keyboard types into its editor — but with no autocorrect and no suggestions. Ticked because everything it was meant to establish is established, including the one gap; that gap is now E2's problem and I2's.

  *Build.* `~/src/android-view` at `bec6c62`, x86_64 rather than the
  README's arm64 because that is what this emulator is:
  `cargo ndk -t x86_64 -P 26 -o masonry-app/src/main/jniLibs/ build -p
  android-view-masonry-demo --release`, then
  `./gradlew :masonry-app:assembleDebug`. **`libmain.so` is 181 MB in
  debug and 11 MB in release** — the loudest single number about
  Vello's dependency graph, and the reason the release build matters
  for more than speed.
  
  *Renderer.* wgpu takes **Vulkan**, and the emulator log confirms it
  from the other side: `Created VkDevice ... for application:'wgpu'`.
  Two things were needed. The emulator must be given Vulkan at all —
  `-feature Vulkan` with `VK_DRIVER_FILES` pointing at the SDK's
  `vk_swiftshader_icd.json`, **plus `-no-snapshot-load`**, which is the
  piece this file had flagged as untested: without a cold boot the
  guest keeps the snapshot's old GPU config and `cmd gpu vkjson`
  reports zero devices however the host is configured. And the native
  library must be **release**: a debug build calls
  `SetDebugUtilsObjectNameEXT` to label its image views, and the
  emulator's own guest driver (`vulkan.ranchu.so`) segfaults inside it.
  On GLES, with no Vulkan available, it instead fails
  `Surface::configure` with "Invalid surface" — untriaged, since the
  Vulkan path works and Vello wants compute shaders anyway.
  
  *Accessibility works*, which E2's condition 6 and every bench script
  depend on. `ui-trace` reads Masonry's AccessKit tree: "Add task"
  arrives as a named `Button`, the editor as an `EditText` node. So
  tap-by-name works against a Masonry screen for any control carrying
  a name; the demo's editor carries none, which is the demo's omission
  rather than the framework's.
  
  *The keyboard: real input yes, suggestions no.* Tapping the editor
  opens the actual soft keyboard (`mInputShown=true`, Gboard), and
  tapping its keys types into Masonry — "teh" typed key by key, with a
  caret. What does **not** appear is Gboard's suggestion strip. The
  control is what makes that a finding rather than an impression: the
  **same three key taps in the Settings app's search field, on the same
  device in the same session, produce "teh | the | yeh"**. So the strip
  works here and android-view's editor is not asking for it — most
  likely the `EditorInfo` its `InputConnection` reports. That matches
  Robrix's report that the Android keyboard is not yet "full", and it
  is the single most important thing to fix or fund upstream, because
  composition, autocorrect and suggestions are exactly what the
  composer in this app needs and exactly what `winit` cannot do at all.
  
  **Cause found 2026-09-04, and it is Masonry's, not android-view's.**
  `~/src/android-view/masonry/src/lib.rs:531` is
  
      fn as_input_connection(&mut self) -> Option<&mut dyn InputConnection> {
          // TODO
          None
      }
  
  so the Masonry demo has **no `InputConnection` at all**; `RustView`
  returns null from `onCreateInputConnection` and the IME falls back to
  dispatching raw key events, which is exactly the behaviour observed —
  keys arrive, composition does not exist, so there is nothing for
  Gboard to suggest against. It is not a wrong `EditorInfo`, and the
  guess above that it was is withdrawn.
  
  android-view's **own** demo (`demo/src/lib.rs`, packaged by `app/`)
  implements the whole trait against a parley editor and asks for
  `INPUT_TYPE_CLASS_TEXT | CAP_SENTENCES | AUTO_CORRECT | MULTI_LINE`
  with `IME_FLAG_NO_FULLSCREEN | NO_EXTRACT_UI | NO_ENTER_ACTION`
  (`demo/src/lib.rs:588`). So the capability is present in the layer
  iris would sit on, and the 30-odd method `InputConnection` trait in
  `src/ime.rs` — `set_composing_text`, `set_composing_region`,
  `finish_composing_text`, `text_before_cursor`, `cursor_caps_mode`,
  `request_cursor_updates`, and `InputMethodManager::update_selection`
  to push the selection back — is the full surface an IME needs.
  **This changes what I2 costs**: the IME bridge is a trait to
  implement over iris's parley editor, not a gap to fund upstream. It
  also means E2 inherits Masonry's TODO, so a Masonry transcript will
  have the same dead composer until somebody fills that in.
  
  **Measured on the emulator, same session, same device.** Built
  android-view's own demo — `cargo ndk -t x86_64 -P 26 -o
  app/src/main/jniLibs/ build -p android-view-demo --release`, then
  `./gradlew :app:assembleDebug`, installed with `ANDROID_SERIAL=$(emu
  serial)` — and tapped into its editor. `dumpsys input_method` reports
  `mInputShown=true` with `mServedView=…viewdemo.DemoView`, and the
  screenshot shows **Gboard's suggestion strip populated with "dolor |
  Dolores | door"**: the caret had landed inside the word *dolor* in
  the demo's lorem ipsum, and Gboard read that word out of the Rust
  editor through `text_before_cursor`. So on this emulator, through
  android-view, a parley editor gets a real IME with real suggestions
  drawn from its own buffer. That is the bar E1 could not reach and the
  bar I2 is written against, and it is now known to be reachable.
  
  *One crash seen once — reproduced and diagnosed 2026-09-04.* With an
  accessibility client attached the app aborted, stack:
  `android_view::view::do_frame` → `CallbackCtx::finish` →
  `accesskit_android::event::QueuedEvents::raise` →
  `send_completed_event` → `unwrap()` on `Err(JavaException)`.
  android-view builds `panic = "abort"`, so a JNI call that throws
  takes the process. Two later `ui-trace record` runs left the app
  alive, so the trigger looked narrower than "a client is attached".
  
  **It is the opposite of "a client is attached": it is a client
  having *detached*.** `accesskit_android`'s `State` enum
  (`adapter.rs:161` in 0.4.0, `:192` in 0.8.0) is
  `Inactive | Placeholder | Active`, and **nothing ever moves it back
  to `Inactive`**. A client — `ui-trace`, which is uiautomator — calls
  into the node provider once, `get_or_init_tree` promotes the adapter
  to `Active`, and it stays there for the life of the process. Every
  later change then returns `Some(QueuedEvents)`, `raise` calls
  `ViewParent.requestSendAccessibilityEvent`, and that reaches
  `AccessibilityManager.sendAccessibilityEvent`, which on the main
  looper **throws `IllegalStateException("Accessibility off. Did you
  forget to check that?")` when accessibility is disabled**. jni-rs
  returns `Err(JavaException)`, `send_completed_event` unwraps it, and
  `panic = "abort"` ends the process.
  
  *The controlled run*, one process (pid 4085), `settings get secure
  accessibility_enabled` = 0 throughout:
  
  - tapped the editor and typed three keys with `adb shell input tap`,
    no client ever attached — **alive**;
  - one `ui-trace record -d 800` with no gesture at all, then two
    seconds' wait — **still alive** (the queue was raised while the
    client was still there);
  - the very next three keystrokes, same process — **aborted**, same
    stack.
  
  So the failure is not the recording; it is the **first thing that
  changes the accessibility tree after a recording ends**. That makes
  it a standing hazard for this project rather than an oddity:
  `transcript-bench.sh`, `stream-bench.sh` and `bench-lib.sh`'s
  tap-by-name all attach and detach uiautomator, so on a Rust app the
  typing or scrolling *after* a bench run is what dies, several
  seconds away from anything that looks like a cause.
  
  **Still present at head**: 0.8.0 is the newest `accesskit_android`
  (the demo resolves 0.4.0) and both the unconditional `unwrap` in
  `send_completed_event` and the one-way `State` are unchanged there,
  so upgrading is not the fix. **Our mitigation for I2/I4 is a gate we
  own**: ask `AccessibilityManager.isEnabled()` before calling
  `raise`, and drop the events when it says no. Worth reporting
  upstream as well — the honest fix is for `raise` to clear a pending
  exception rather than unwrap it, since a view can be detached or
  accessibility switched off between queueing and raising no matter
  who is calling.
  
  *A rig trap that cost a wrong conclusion.* Several bounded runs were
  given `sleep N; emu down` watchdogs, and one armed for an earlier
  experiment fired in the middle of a later one — the app vanished, adb
  hung, and it read exactly like the Vulkan path crashing. It was not.
  A watchdog must be scoped to the process it guards (`kill $pid`, with
  the pid captured at launch) rather than to whatever AVD is running
  when it wakes, and only one should be armed at a time.
  

• E2 — a transcript in Masonry (2026-09-05). Built and run on this emulator. It found the thing it was measuring for: a framework-wide gap that blocks the bench comparison itself, plus a full accounting of the seven behaviours. Ticked on E1's own precedent -- "everything it was meant to establish is established, including the one gap."

  *Where it lives.* `~/src/android-view/e2-transcript` (new workspace
  member, `crate-type = ["cdylib"]`, `lib.name = "main"`), packaged by
  a new Gradle module `~/src/android-view/e2-app` copied from
  `masonry-app` (`E2View`/`E2Activity`, package
  `org.linebender.android.e2transcript`). Neither is committed to
  `ai-app-2` or pushed anywhere -- same as E1, this is a local
  experiment against the `xilem` commit
  `e14ba3a5f9461b403cb30d95826187fba7f6924b` and the `android-view`
  commit `bec6c62a96cef8239b0fd7fedeef9b184d02e3a1`, reproducible from
  the commands below rather than from a remote.
  
  *Build.* Depends on `client-core`/`event-model` from this checkout by
  path (`../../../repos/ai-app-2/client-core`) -- real code, not a
  reimplementation: `ApiClient`/`UreqTransport` for the HTTP fetch,
  `fold_event`/`group_tool_runs` for the transcript fold, exactly what
  the app itself would use. The sandbox CA and a session's URL/token
  are baked in at build time via `env!()`/`include_bytes!()`, the same
  pattern the real APK uses to pin its CA (AGENTS.md), since this is a
  throwaway screen with no enrollment flow:
  
      cd app && ./ui-sandbox.sh start   # prints the port and token
      sid=$(./ui-sandbox.sh spawn e2test)
      ./ui-sandbox.sh send "$sid" @/tmp/big.md   # markdown content
      ./ui-sandbox.sh send "$sid" "/tools 3"     # a grouped tool run
      cd ~/src/android-view
      E2_SANDBOX_URL=https://10.0.2.2:<port> \
      E2_SANDBOX_TOKEN=<token> \
      E2_SANDBOX_SESSION=<sid> \
      E2_CA_PEM_PATH=$HOME/.config/ai-app/certs/ca.pem \
          cargo ndk -t x86_64 -P 26 -o e2-app/src/main/jniLibs/ \
              build -p e2-transcript --release
      ANDROID_HOME=~/Android/Sdk ./gradlew :e2-app:assembleDebug
  
  **`libmain.so` is 13.5 MB release** (E1's masonry-demo was 11 MB;
  the difference is `client-core`'s `ureq`/`rustls` stack, which E1's
  demo does not link). Release native lib, debug Gradle variant --
  the combination E1 found necessary (a debug build's
  `SetDebugUtilsObjectNameEXT` segfaults this emulator's Vulkan
  driver).
  
  *Emulator.* This checkout's own AVD (`ai-app-2`, not `ai-app`, which
  another session already had up), booted with Vulkan the way E1
  established: `GPU_HOST_FEATURES="-feature Vulkan"
  VK_DRIVER_FILES=$HOME/Android/Sdk/emulator/lib64/vulkan/vk_swiftshader_icd.json
  emu up`. `adb shell cmd gpu vkjson` confirmed a device before
  anything was installed. Torn down with `emu down` at the end of this
  session (see "Where things stand" below for the exact state left).
  
  *What it does.* `fetch_rows()` (`e2-transcript/src/lib.rs`) makes one
  blocking `fetch_transcript_page(session, None, 800, false)` call
  before the widget tree exists, folds every line through
  `client-core`, and groups tool runs -- 854 real events from a mixed
  sandbox session (markdown paragraphs/headings/fences plus a
  three-call tool run from the echo driver's `/tools 3`). Each
  `TranscriptRow` becomes one `VirtualScroll<dyn Widget>` child,
  built lazily from `VirtualScrollAction` the way
  `masonry_winit/examples/virtual_fizzbuzz.rs` does it. **This is a
  deliberate scope cut from "page 800 events" as live paging**: all
  854 rows' content is fetched once, and what `VirtualScroll` pages is
  *widget construction*, not a second round of network calls per
  scroll -- wiring a background-thread fetch woken across the JNI
  boundary (the way I2's `ssh.rs` attach-and-call works) is real work
  this experiment did not need to answer its question. `markdown.rs`
  is a `pulldown-cmark` event-stream walk into a small `Block` enum
  (`Text`/`Heading`/`Code`), with its own module doc explaining the
  one real ceiling it hit (below).
  
  *Verification.* `cargo fmt -p e2-transcript -- --check` clean.
  `cargo ndk -t x86_64 -P 26 clippy -p e2-transcript --all-targets`:
  **zero warnings in this crate** (the only clippy output at all is
  from `android-view` itself, a vendored dependency this experiment
  does not own). `cargo ndk -t x86_64 -P 26 test -p e2-transcript
  --lib` (run against the emulator, since the crate is
  `cfg`-unconditionally Android): 1 test, `markdown::parse`'s block
  split, passing. No larger test surface exists to port -- this is a
  throwaway screen, not a library, matching AGENTS.md's "match the
  codebase's testing posture."
  
  *Screenshots* (all `/tmp`, not committed -- see the standing rule
  against transcripts leaving this repo, which applies equally to a
  screenshot of one): `e2-screenshot2.png` first real content;
  `e2-expand.png` a tool row expanded with its top edge held;
  `e2-markdown.png` a heading/bold/italic/inline-code/link/fenced-code
  message (the "You said: ## A Heading" line is the sandbox's echo
  driver prefixing the literal input text before the `##`, which
  keeps `pulldown-cmark` from recognising it as a heading -- a fixture
  artifact, not a finding about Masonry).
  
  **The seven behaviours, each shown or given a sourced reason:**
  
  1. **One selectable body of text spanning rows -- not possible, and
     it is a real ceiling, not an oversight.** `Prose` wraps exactly
     one `TextArea<false>`, which wraps exactly one
     `parley::PlainEditor` (`masonry/src/widgets/prose.rs`: "Note that
     copying is not yet implemented"). Selection lives entirely inside
     that one editor: `TextArea::on_pointer_event`
     (`masonry/src/widgets/text_area.rs:414-459` in the pinned `xilem`
     commit) captures the pointer on `Down`
     (`ctx.capture_pointer()`) and drives `self.editor`'s own
     `extend_selection_to_point` on `Move` -- there is no code path,
     in `masonry_core` or `masonry`, that extends a selection into a
     second widget's editor. A drag that starts in one row's `Prose`
     and continues into the next is still that first row's own
     `PlainEditor` being asked for a point outside its bounds; it
     cannot reach the second row's text. Confirmed by reading, not
     guessed at: there is no `SelectionContainer`-shaped type
     anywhere in `masonry`, `masonry_core` or `xilem` (checked with
     `grep -rln "SelectionContainer\|cross.widget.*selection"`, zero
     hits).
  2. **Rich inline text -- block-level yes, inline no, and both for
     the same reason.** `TextArea::edit_styles()` returns one
     `&mut StyleSet<T>` for the whole editor
     (`masonry_core/src/core/text.rs:29-32` defines `StyleSet` as
     `parley::StyleSet<BrushIndex>`, applied editor-wide); the type's
     own comments say why nothing finer exists yet:
     `// TODO: RichTextInput 👀` and
     `// TODO: Support for links - https://github.com/linebender/xilem/issues/360`
     at `masonry/src/widgets/text_area.rs:43-44`. So bold, italic,
     inline code and a link *inside one paragraph* cannot each carry
     their own style without leaving `TextArea` for a hand-rolled
     `parley::Layout` (which loses selection, the caret and copy,
     since those live inside `PlainEditor` specifically). What **is**
     real: each markdown block is its own `Prose`, so a heading is a
     bigger font and a fenced code block is monospace, screenshotted
     in `e2-markdown.png` -- block-level style works because it is
     block-level *widgets*, not a rich-text API. Tables and per-token
     syntax colour inside a fence hit the identical ceiling (both are
     per-range styling) and were not attempted for the same reason.
     `markdown.rs`'s degraded rendering (backticks kept literally,
     `[text](url)` shown as `text (url)`) is the honest fallback,
     documented at the point it is produced.
  3. **Bottom-anchored virtualised list, paged, hold-top-edge on
     expand -- mostly shown, with one real gap in the anchor API.**
     `VirtualScroll<dyn Widget>` holds all 854 folded rows;
     `overwrite_anchor` before swapping a tool row's widget for its
     expanded/collapsed version is exactly the primitive
     `holdTopEdge` needs, and it worked: `e2-expand.png` shows the
     row growing downward from the same top edge it had collapsed,
     no jump. Virtualisation is real (`ui-trace elements` only ever
     lists the rows currently on screen, never all 854). **What did
     not come free: hugging the bottom of the screen.**
     `VirtualScroll::new`'s doc says "the item at `initial_anchor`
     will have its top aligned with the top of the scroll area" --
     so anchoring on the last row puts that row's top at the
     viewport's *top*, with empty space below it, not at the
     viewport's bottom the way a chat transcript wants (visible in
     `e2-screenshot2.png`). The complete public `WidgetMut` surface of
     `VirtualScroll` is `new`, `with_valid_range`,
     `will_handle_action`, `add_child`, `remove_child`, `child_mut`,
     `set_valid_range`, `overwrite_anchor`
     (`masonry/src/widgets/virtual_scroll.rs:257-428`) -- no
     scroll-offset setter and no reverse/bottom-up layout mode exist
     to ask for the other behaviour. Backward paging beyond the
     initial 800 was not exercised, per the scope cut above.
  4. **The soft keyboard -- inherited gap, not re-investigated.** E2's
     screen has no `TextInput`, only read-only `Prose`/`Button`, so it
     does not hit `masonry/src/lib.rs:531`'s `as_input_connection`
     returning `None` directly -- but it would the moment a composer
     is added, per E1's finding. Nothing new to add here.
  5. **Platform integration -- out of scope by design.** Foreground
     service, notifications, share sheet, deep link, Keystore,
     camera, back gesture, edge-to-edge, local-network permission are
     E3's list in RUST.md's own experiment order, not E2's.
  6. **Accessibility names -- shown, and the bench-script dependency
     actually exercised.** `ui-trace record --do "tap '> 3 tool
     calls'"` found the button by its label and pressed it (that tap
     is what produced `e2-expand.png`); `Prose` rows surface their
     text as their accessible name too (`ui-trace elements` lists
     "You said: One more short reply..." etc. as named nodes). Tap by
     name, the rule this whole project's bench scripts depend on,
     works against this screen.
  7. **Measurable frames -- blocked, and this is the finding E2 was
     really testing for.** Two separate problems, one of them fatal
     to the render-numbers half of this box's own pass condition.
     First, Masonry has no render-report/per-widget-cost
     instrumentation the way Compose's `DebugStats` gives this
     project -- building one was out of scope here. Second, and this
     is the one that matters: **neither of Masonry's two scrolling
     widgets responds to a touch drag at all.**
     `VirtualScroll::on_pointer_event`
     (`masonry/src/widgets/virtual_scroll.rs:504-523`) and
     `Portal::on_pointer_event`
     (`masonry/src/widgets/portal.rs:259-267`) both match only
     `PointerEvent::Scroll` (wheel/trackpad deltas) and do nothing
     with `PointerEvent::Down`/`Move`/`Up` -- there is no drag-to-scroll
     gesture logic anywhere in the widget set. `android-view`'s own
     Java bridge keeps the two paths separate at the source:
     `RustView.java`'s `onTouchEvent` forwards raw touch straight to
     Rust, and only `onGenericMotionEvent` (mouse/trackpad, not
     touch) reaches the `ACTION_SCROLL` branch that becomes
     `PointerEvent::Scroll`
     (`android-view/src/events.rs:530`). Confirmed empirically, not
     just by reading: a real swipe (`ui-trace`'s `swipe 540 1600 540
     400 300`, twice) moved nothing (`e2-scroll.png` is pixel-identical
     to the screen before it), and a synthetic Android wheel event
     (`adb shell input scroll 540 1200 --axis VSCROLL,-5`) also moved
     nothing. **This means `transcript-bench.sh`'s own gesture --
     a finger swipe -- cannot scroll a Masonry transcript on Android
     today, at all, on this framework commit.** So the "render
     numbers land within the Compose baseline" half of this box's
     pass condition cannot be attempted, let alone met: there is no
     way to perform the scroll the comparison asks for. This is not
     a performance shortfall to close by writing faster code: it is
     an absent input path upstream. The fix is a drag-to-scroll
     gesture in `on_pointer_event` (the same place `TextArea`'s own
     caret-drag logic already lives, so the pattern -- capture on
     `Down`, accumulate delta on `Move`, release on `Up` -- exists
     in this codebase already, just not wired into either scrolling
     widget), and it belongs upstream in `xilem` rather than in this
     project.
  
  **Net for RUST.md's recommendation.** Item 3 ("decide when the
  transcript screen exists in both, from the measurements") cannot be
  decided by a render-number comparison yet, because the comparison's
  own gesture does not work on Masonry on Android. What *can* be
  compared today is structural: iris already has a working scroll
  gesture and a working touch model (I2, 2026-09-05) that Masonry's
  upstream commit does not yet have for this exact case. That is a
  point in iris's favour that a frame-time number would not have
  shown any more clearly.
  

• E3 — the shell (2026-09-05). Both pass-condition proofs held on the emulator: a notification arrived while the app was closed, and a shared text share landed as a real message in a session's transcript. Committed to this repo (unlike E1/E2's external, uncommitted trees), since this is lightweight glue rather than a multi-gigabyte native build.

  *Where it lives.* `android-shell/` (new crate, `client-core` as its
  only real dependency) is the JNI bridge; `app/shellApp/` is a **new
  Gradle module**, not a rewrite of `app/androidApp` in place --
  that module is ~13,000 lines of working Compose UI this experiment
  does not touch or risk, and the two install side by side on one
  development device. `app/shellApp`'s manifest, channel names,
  notification wording and share intent-filter are copied from
  `androidApp`'s (`Notifications.kt`, `Share.kt`, the manifest) per
  AGENTS.md's "reuse rather than re-derive" -- see each file's own doc
  comment for exactly what was carried over. Two deliberate
  differences, both practical rather than behavioural: application id
  `com.example.aiapp.shell` and deep-link scheme `aiappshell` (not
  `aiapp`), so this experiment's install cannot collide with the real
  app's enrollment or Keystore alias on the same phone -- see
  `android-shell/src/settings.rs`'s `SCHEME` doc.
  
  *The Java floor, and one line more than planned.* Two classes,
  matching "How much Java is unavoidable" almost exactly:
  `MainActivity.java` (`onCreate`/`onNewIntent` forward to
  `nativeHandleIntent`) and `NotificationService.java`
  (`onStartCommand`/`onDestroy`/a `sync()` companion, three natives).
  Both ~30 lines including the license-free boilerplate Java itself
  demands (imports, `System.loadLibrary`). **One addition the analysis
  did not anticipate**: `MainActivity.toast(Context, String)`, a
  plain (non-native) static method Rust *calls* rather than
  implements, because posting a `Toast` from `share.rs`'s background
  thread needs a hop back to the main looper
  (`new Handler(Looper.getMainLooper()).post(...)`), and JNI can call
  an existing Java method on any thread but cannot construct a Java
  `Runnable` to hand to `Handler.post`/`runOnUiThread` without a
  reflection proxy uglier than three lines of Java. Recorded here
  because "the floor is two classes of ten lines" undersold this by
  exactly one small, call-only method -- the pattern (Rust calls
  Java, never Rust implements a Java interface) is worth keeping the
  next time this floor is estimated.
  
  *What client-core gained.* `notifications.rs`: `SessionNotification`,
  `NotificationKind` (mirroring `server/src/session/mod.rs`'s wire
  shape field-for-field) and `follow_notifications`, the SSE parse
  over `/notifications` built on the same `sse::SseReader` and
  `Transport` trait `event_stream.rs` already uses. `attention_line`
  is ported verbatim from `Notifications.kt`. 3 new tests (88 total in
  the crate); `android-shell` itself has none, since every function in
  it needs a live `Env` and there is no pure logic left to test in
  isolation once client-core owns the parsing -- matches E2's
  precedent ("a throwaway screen, not a library").
  
  *Scope cuts, each recorded at its own point in the code rather than
  only here:*
  - **Text-only share.** `Intent.EXTRA_TEXT` becomes a session message;
    a shared file/photo URI is not uploaded, because `client-core`'s
    `ApiClient` has no `/sessions/{id}/attachments` route yet either
    (`CLIENT_CORE.md`'s own "not covered" list) -- porting
    `Attachments.kt`'s `ContentResolver` reads and bitmap downscaling
    is real work belonging to whichever caller needs it next, not a
    detour inside this box.
  - **No session picker.** With no screen drawn yet (E4's job), a
    share attaches to whichever session has the latest
    `last_activity` -- documented as a placeholder in `share.rs`,
    not a designed behaviour.
  - **No banner/on-screen suppression.** `notify::show` skips
    `Notifications.kt`'s "nothing if this session is on screen" /
    "hand to the app as a banner" branches entirely: both read
    process-wide state that only means something once a screen
    exists to register against it, so every notification here takes
    the platform-drawer branch -- which is also exactly what the pass
    condition asks for. Revisit once E4 draws something.
  - **Keystore is not reimplemented in Rust.** `settings.rs` calls
    `wg-app-link`'s existing `ServerStore`/`ServerSettings` Kotlin
    classes over JNI rather than re-deriving the AES-GCM sealing:
    that code is shared with Dev Updater, already tested, and tied to
    a Keystore alias an existing enrolled phone depends on. This does
    mean `kotlinc` stays in the toolchain regardless of what E5 does
    with `javac`/`d8` for this module's own two classes -- a
    correction to "Can the APK be built without Gradle?"'s assumption
    that dropping Kotlin drops `kotlinc` outright; it drops it for
    *this app's own code*, not for a shared submodule pulled in as a
    dependency.
  
  *`jni` 0.22, not the older API most examples assume.* This is a
  real API split (`Env` for real work, `EnvUnowned` as the FFI-safe
  type a native fn receives, joined by `EnvUnowned::with_env`), and
  the `native_method!` macro (used for all four natives here, via
  `const _: NativeMethod = native_method! { ... }`) generates both the
  mangled `Java_...` export and the panic/error-handling wrapper from
  one Rust function signature -- chosen over hand-written
  `#[unsafe(no_mangle)] extern "system" fn Java_com_..._method` because
  a hand-typed export name and a hand-typed JNI signature string
  routinely drift from the Java they claim to match, silently (see
  the next two findings, both of which were exactly that drift).
  `error_policy = LogErrorAndDefault` reports a failure to logcat
  rather than throwing it back into Java as an exception that would
  crash the app over something recoverable -- matching
  `Notifications.kt`'s own "log, don't crash" posture, but it is a
  no-op without a logger backend (`android_logger`, Android-only
  dependency, `lib.rs`'s `ensure_logger`) installed; the class of bug
  this exists to report was found once with no logger and read as
  nothing having gone wrong at all.
  
  **Three real findings, each cost a failed run before being
  diagnosed, each written where the fix lives so a reader who touches
  that file again does not lose an afternoon to it:**
  
  1. **A generic `JObject` parameter type silently exports the wrong
     JNI signature.** `native_method!`'s shorthand
     `fn native_sync(context: JObject) -> ()` encodes the export as
     `(Ljava/lang/Object;)V`, because it has no way to know the
     intended Java type is `android.content.Context` from a bare
     `JObject`. The real Java method is declared
     `(Landroid/content/Context;)V`; the two mangled names never
     resolve to each other, and the failure is
     `UnsatisfiedLinkError: No implementation found`, thrown the
     moment Java calls it -- not a build error on either side. Fixed
     by spelling each parameter as its actual Java type in the macro
     invocation (`context: android.content.Context`, `activity:
     android.app.Activity`, ...), which the macro accepts directly
     per its "Java Object Types" syntax, while the Rust implementation
     function keeps the parameter as plain `JObject` (the "Built-in
     Types" fallback for a Java class with no dedicated Rust
     wrapper). `lib.rs`'s comment beside the first `native_method!`
     call is the citation.
  2. **A class looked up by name from this crate's own background
     thread fails, and only for app classes.** `android-shell`'s
     follow-loop and share threads are Rust-spawned and attached via
     `JavaVM::attach_current_thread`, which the platform never handed
     an app `ClassLoader` -- so `FindClass`'s default fallback (used
     internally by `find_class`/`new_object`/`call_static_method`/
     `get_static_field`, anything that resolves a class *by name*
     rather than from an object it already holds) only reaches the
     bootstrap loader's framework classes. `androidx.core.app.
     NotificationManagerCompat`, packaged inside this app's own APK,
     is invisible from there: `Error::NoClassDefFound`, logged by
     `notify::show`'s `LogErrorAndDefault` as "failed to resolve Java
     class ... (class not found or linkage error)" -- which on a real
     device is indistinguishable from "the notification silently
     never arrives," since the *ongoing* foreground notification
     (built on the main thread, before this thread exists) posts
     fine regardless, so nothing else looks wrong. Fixed in
     `jcall.rs`: `remember_class_loader` caches the app's own
     `ClassLoader` (`context.getClass().getClassLoader()`) the first
     time any entry point with a `Context` runs, and every
     class-by-name lookup goes through `LoaderContext::Loader`
     explicitly rather than the thread-dependent default -- correct
     on the main thread and this crate's background threads alike.
     `jcall.rs`'s module doc has the full account.
  3. **`onStartCommand` spawning a thread unconditionally opens a
     second connection, and `Notifications.kt` has the same bug.**
     Enrolling calls `sync()` twice in one launch (once
     unconditionally in `MainActivity.onCreate`, again inside
     `handle_enrollment` after saving the token), each of which starts
     the service, and Android runs `onStartCommand` once per start
     request -- so the follow-loop thread was spawned twice, caught on
     `adb logcat` as two `jni::vm::java_vm: Attached thread
     ai-app-notifications` lines for one enrollment. Kotlin's
     `onStartCommand` has the identical shape (`thread(isDaemon =
     true) { follow(settings) }`, no guard), so this is a latent bug
     in the reference implementation this port found by testing
     rather than something E3 introduced -- worth carrying the same
     guard back to `Notifications.kt` separately, not done here.
     Fixed in `notify.rs` with a `RUNNING` `AtomicBool`, `swap`ped
     true before spawning and reset in `on_destroy`; see its doc
     comment for the accepted race this shares with the pre-existing
     `STOPPING` gap below.
  
  **Known gap, not fixed, written where it will be found.**
  `notify.rs`'s `STOPPING` flag (checked between reconnects) cannot
  interrupt a `ureq` read already blocked inside one connection --
  unlike `HttpURLConnection.disconnect()`, `client_core::Transport`
  exposes no cancellation handle. `/notifications` is idle between
  events (a keep-alive), so in practice a stop is a bounded wait
  rather than a hang; closing this for real means adding a
  cancellation point to the `Transport` trait itself, a decision
  affecting every caller, not an `android-shell`-only fix.
  
  *Verification, exact commands.* `cargo fmt -- --check`,
  `cargo clippy --all-targets` (zero warnings) and `cargo build`
  clean for both `client-core` and `android-shell` on the host
  target; `cargo ndk -t x86_64 -P 26 clippy --all-targets` clean for
  `android-shell` on the Android target too (the `android_logger`
  dependency is Android-only, so this is the only way to compile-check
  it). `./run-tests.sh` from the repo root: 127 `server` tests, 88
  `client-core` tests (85 + the 3 new to `notifications.rs`), all
  passing -- the port added no regression to what already worked.
  `./gradlew :shellApp:lintDebug`: `No issues found` (the report at
  `app/shellApp/build/reports/lint-results-debug.txt`).
  
  *The two pass-condition proofs*, both on this checkout's own AVD
  (`ai-app-2`, GPU host per the default, torn down with `emu down`
  when this session finished) against `app/ui-sandbox.sh`:
  
  - **Notification with the app closed.** Enrolled via
    `adb shell "am start -a android.intent.action.VIEW -d
    'aiappshell://enroll?host=10.0.2.2&port=<sandbox port>&token=<token>'"`
    (per the sandbox's own banner, substituting the scheme), granted
    `POST_NOTIFICATIONS`, pressed home, then
    `./ui-sandbox.sh spawn e3notif2` and
    `./ui-sandbox.sh send <sid> "/question Should I proceed with the deploy?"`.
    `adb shell dumpsys notification --noredact` shows a
    `channel=sessions` record, `android.title=e3notif2`,
    `android.text=Waiting for you` (matching `attention_line` and the
    session's own title, exactly what `Notifications.kt` would have
    shown) -- posted while the app held no visible activity. Tapping
    it (`ui-trace record --do "tap 'e3notif2'"`, found in the
    expanded shade after `adb shell cmd statusbar
    expand-notifications`) launched
    `com.example.aiapp.shell/.MainActivity` with
    `dat=aiappshell://session/...`, confirmed in `adb logcat`'s
    `ActivityTaskManager: START` line -- the `PendingIntent` names
    the right session.
  - **A share lands in a session.** With the app enrolled and a
    session already active,
    `adb shell "am start -a android.intent.action.SEND -t text/plain
    --es android.intent.extra.TEXT 'Please check the deploy logs for
    errors.' -n com.example.aiapp.shell/.MainActivity"` (the classic
    `adb shell` quoting trap from `this-machine-android` applies here
    too: the whole `am start` invocation has to be one single-quoted
    string handed to the *remote* shell, or the extra's spaces get
    re-split away). `./ui-sandbox.sh api
    '/sessions/<sid>/transcript?limit=20'` shows
    `{"type":"userMessage","text":"Please check the deploy logs for
    errors."}` followed by the echo driver's reply -- the share
    reached the most-recently-active session as a real message, not
    a mock.
  

• E4 — the same screen on the desktop (2026-09-05). A new iris/desktop-app crate (added to the iris workspace's members, not excluded the way android-app is -- nothing here needs the NDK): a real winit window showing a session list (iris::widget::Span, rebuilt on selection) beside transcript-ui's screen (transcript_ui::build_tree, new this box -- see IRIS.md's 2026-09-05 entry), talking to a real ai-server through client-core's ApiClient/UreqTransport/follow_session_events. Enrolment is client_core::config::EnrolledServer::parse_link against the same aiapp://enroll?host=H&port=P&token=T link a phone scans, pasted via --link and persisted at $XDG_CONFIG_HOME/ai-app-desktop/enrollment.json (0600 -- iris/desktop-app/src/config.rs); the pinned CA is a --ca PATH argument, never baked in (DECISIONS.md, 2026-09-05).

  *Both pass-condition proofs held, against `app/ui-sandbox.sh`'s real
  server.* (1) The list showed the sandbox's spawned session
  ("Demo session", its live status); selecting it loaded the real
  transcript and the composer's `Submit` posted a message whose reply
  streamed in live over SSE, both proved by two `run-headless.sh`
  screenshots taken seconds apart around a real `./ui-sandbox.sh send`
  -- the second showed the new turn appended under the first with
  nothing duplicated or lost. (2) Screenshotted headless:
  `/tmp/iris_e4_desktop.png` (1920x1200, 15.9 KB, the real first-run
  state -- list populated, "Select a session." on the right, nothing
  selected yet). `run-headless.sh` gained a `--bin` flag for this
  (`cargo build --bin NAME` + `target/debug/NAME` instead of the
  `--example` path, since `desktop-app` is a real binary a person
  runs, not a demo) and `$RUN_HEADLESS_ARGS`, word-split into the
  launched binary's own argv (a real CLI's flags, which no example
  needed a way to pass before). Exact commands, from `iris/`:
  
      TOKEN=$(cat "${XDG_CONFIG_HOME:-$HOME/.config}/ai-app/sandbox-token")
      LINK="aiapp://enroll?host=127.0.0.1&port=<PORT>&token=$(python3 -c \
        'import sys,urllib.parse;print(urllib.parse.quote(sys.argv[1],safe=""))' "$TOKEN")"
      CA="${XDG_CONFIG_HOME:-$HOME/.config}/ai-app/certs/ca.pem"
      RUN_HEADLESS_ARGS="--ca $CA --link $LINK" \
        ./run-headless.sh desktop-app --bin --shot /tmp/iris_e4_desktop.png -- -p desktop-app
  
  **A real bug this screenshot found, not a synthetic one**: the first
  attempt resumed the live SSE stream from
  `items.iter().map(TranscriptItem::seq).max()` -- the *folded* item's
  seq, which for a still-open `AssistantMsg` is the seq of its
  *first* delta by design (`fold_event`'s own doc comment: "a row
  whose identity changed with every delta would be a new row every
  frame"). Resuming from there re-delivered every delta already
  folded into that message, and the screenshot showed the assistant's
  reply with its own tail duplicated ("You said: ... testsaid: ...
  test"). Fixed by computing the resume cursor from the raw wire
  `seq` of the last fetched line (`app.rs`'s `raw_seq`) instead of
  from any folded item -- regression test
  `the_resume_cursor_is_the_last_wire_seq_not_the_last_items_seq` in
  `iris/desktop-app/src/app.rs`. Exactly the class of bug CODE_RULES
  warns about under "a fix tried only on what it was meant to fix":
  the bare REST fetch (no live stream yet) looked perfect on its own,
  and only *resuming* a stream after it exposed the seam.
  
  **Deliberately left simple at the time, later fixed** (`app.rs`'s
  module doc has the full account): every incoming SSE event used to
  refold the session's whole item list and rebuild the entire
  right-hand widget tree from scratch on every event, rather than
  reaching for `TranscriptScreen::push_row`'s incremental append --
  `push_row` could only add a new row, and a streaming reply is
  exactly a row whose text keeps changing after it first appears.
  Fine at the size a desktop session's conversation is; wrong for a
  long, fast-streaming one -- fixed below (this same box's "Streaming
  no longer costs a full rebuild" entry) by giving `transcript-ui` a
  `TranscriptScreen::apply` that updates a row already on screen
  instead of rebuilding every row around it. `rebuild_transcript`
  still runs the whole tree once, for a freshly loaded/selected
  session and for `apply`'s own rare full-rebuild fallback. No history
  paging (I3's job, reused as-is if this becomes permanent) and no
  scroll-position preservation across a rebuild -- both still named
  rather than silently missing, and neither depends on the fix below.
  Background network I/O runs on plain `std::thread`s reporting back
  through winit's `EventLoopProxy<AppEvent>` rather than iris's own
  `Tasks`/`task_on`, because `Tasks` only requests a redraw once after
  its whole async closure finishes, which fits "one request, one
  update" and not a live stream that needs a redraw after *each*
  event it relays.
  
  Verification: `cargo fmt --all`, `cargo clippy --workspace
  --all-targets` (zero warnings), `cargo test --workspace` from
  `iris/` (7 new tests in `desktop-app` -- 4 for
  `config.rs`'s save/load/permissions/corruption, 3 for `app.rs`'s
  transcript folding and the resume-cursor regression above -- plus
  the existing 37 unchanged), and `./run-tests.sh` at the repo root
  (127 passing, `client-core` alone 93 -- the `EnrolledServer` parsing
  tests already existed before this box). Android is untouched by
  this step, as asked.
  

• E5 — the packaging xtask (2026-09-05). Both pass-condition proofs held on this checkout's own emulator: adb install -r of the xtask-built APK over the Gradle-built one succeeded, and the notification service reached its follow-loop and posted a real notification while the app was backgrounded. cargo xtask apk at the repo root (.cargo/config.toml's alias for cargo run --manifest-path xtask/Cargo.toml --) runs cargo ndk → javac/d8 → aapt2 → zipalign → apksigner with no Gradle driving the packaging itself -- one disclosed exception, below.

  *Where it lives.* `xtask/` (new, independent crate at the repo
  root -- **no Cargo workspace**, matching every other crate here;
  `run-tests.sh` already `cd`s into each rather than assuming one).
  **Zero dependencies**: every step is "run this SDK/JDK tool with
  these arguments and check its exit status," which needs nothing a
  crate would add (AGENTS.md's "new dependencies need a reason").
  `src/sdk.rs` finds the SDK root/build-tools/`android.jar` the same
  way `app/android-env.sh` does ($ANDROID_HOME, then
  $ANDROID_SDK_ROOT, then `~/Android/Sdk`); `src/keystore.rs`
  finds-or-generates the release key with the exact recipe
  `app/build-apk.sh` uses (same env vars, same path, same `keytool`
  invocation) so the two tools sign with the *same* key, plus a
  `--debug` path using the conventional `~/.android/debug.keystore`;
  `src/apk.rs` is the pipeline itself, `src/main.rs` the ~40-line CLI.
  About 420 lines total against RUST.md's earlier "about 150" guess --
  the difference is almost entirely dependency handling (below), which
  the earlier estimate didn't anticipate.
  
  *`:link`'s `kotlinc` question, resolved.* Checked first, since E3
  left it open: no standalone `kotlinc` exists on this machine (not on
  PATH, not under any SDK -- only `kotlin-compiler-embeddable` jars
  inside Gradle's own distributions). So the choice was never
  "invoke kotlinc" versus "port `ServerStore`/`ServerSettings` to
  Java" as originally framed -- a third route fell out of solving the
  *other* open dependency problem (androidx, next paragraph): the one
  Gradle call already needed for that also compiles `:link`'s Kotlin
  as a side effect, via Gradle's own embedded compiler, and hands back
  the resulting `classes.jar` in the same resolved-jars list. That is
  RUST.md's own "prebuild it once into a jar/aar E5 consumes as a
  binary input" option, arrived at for free rather than built
  specially -- no Java port of `ServerStore` was written, and
  `android-shell/src/settings.rs`'s JNI class-by-name lookup
  (`com/example/wgapplink/ServerStore`) needed no change.
  
  *One disclosed exception to "no Gradle in the loop": dependency
  resolution.* `app/shellApp` depends on `:link` (Kotlin, above) and
  on `androidx.core:core-ktx` -- not a compile-time dependency of the
  two Java stub classes (`MainActivity`/`NotificationService` import
  only `android.*`), but a **runtime** one: `android-shell/src/notify.rs`
  reaches `NotificationCompat`/`NotificationChannelCompat`/
  `NotificationManagerCompat`/`ServiceCompat`/`ContextCompat` by class
  name over JNI, so their bytecode has to be in the final dex even
  though nothing in this pipeline's own Java source mentions them.
  Reimplementing a Maven/AAR dependency resolver to avoid one Gradle
  call was not a good trade against "smallest honest route" (the
  standard this file already applied to `kotlinc`) -- so
  `app/shellApp/build.gradle.kts` gained one task,
  `printRuntimeClasspathJars`, which asks the `releaseRuntimeClasspath`
  configuration for its artifacts through an `ArtifactView` requesting
  the `android-classes-jar` attribute (the same post-AAR-transform
  view AGP's own dexing task consumes, so an AAR is already unpacked
  to a plain `.jar` by the time the xtask sees it) and writes their
  absolute paths, one per line, to
  `app/shellApp/build/xtask/runtime-classpath.txt`. `cargo xtask apk`
  runs `./gradlew :shellApp:printRuntimeClasspathJars` once (a few
  seconds, mostly UP-TO-DATE on a warm Gradle daemon), reads that file,
  and hands every jar in it to `d8` as an ordinary program input --
  `:link`'s `classes.jar` among them, per the paragraph above. Nothing
  past that one call touches Gradle. **What this trades away**: the
  pipeline is not Gradle-free end to end, only Gradle-free for the
  part that was actually expensive (assembling and dexing the app's
  own code, which the earlier options -- kotlinc, or a hand-rolled
  resolver -- were the two ways to avoid entirely). Recorded here
  rather than left implicit, matching how the `kotlinc` compromise
  above is recorded.
  
  *The rest of the pipeline, in order (`apk.rs`):* `cargo ndk -t
  arm64-v8a -t x86_64 -P 26 -o app/shellApp/src/main/jniLibs/ build
  --release -p android-shell` (both ABIs by default -- real phone and
  this machine's emulator -- `--abi` overrides; always `--release`
  for the native library regardless of the APK's signing variant, for
  the reason E1 already established: a debug build's Vulkan
  object-labelling segfaults this emulator's driver, and there is no
  reason for a signing choice to make this crate's `.so` bigger).
  `javac -cp android.jar` compiles `MainActivity.java`,
  `NotificationService.java` and a freshly generated `PinnedCa.java`
  (same template as the Gradle `generatePinnedCa` task, same
  opening-quotes-adjacent-to-`"""` rule from AGENTS.md's "Things that
  have bitten") into one `classes.jar` (`jar cf` -- `d8` rejects a
  bare directory of `.class` files outright, "Unsupported source file
  type", discovered by trying it). `d8 --release --min-api 24 --lib
  android.jar` dexes that jar plus every classpath jar from the
  paragraph above into one `classes.dex` (no multidex needed at this
  size). `aapt2 link` compiles `app/shellApp/src/main/AndroidManifest.xml`
  into the base APK's `resources.arsc` -- the checked-in manifest has
  no `package` attribute (Gradle injects one from `android.namespace`
  during a manifest merge this pipeline doesn't run), so `apk.rs`
  writes a copy with `package="com.example.aiapp.shell"` spliced in
  rather than editing the source manifest, and refuses to run at all
  if the source ever gains one of its own (a version-drift guard
  cheaper than a real merge). `--min-sdk-version`/`--target-sdk-version`/
  `--version-code`/`--version-name` are passed on the command line for
  the same reason -- the raw manifest carries none of them, Gradle's
  `defaultConfig` normally does. `jar uf` (not a hand-rolled zip
  writer -- `jar` ships with the JDK this pipeline already needs)
  merges `classes.dex` and a staged `lib/<abi>/libandroid_shell.so`
  tree into the base APK (cargo-ndk's `-o` writes
  `jniLibs/<abi>/*.so`, matching the Gradle source-set layout it was
  pointed at; Android's own zip convention wants `lib/<abi>/*.so` at
  the archive root, hence the staging copy rather than an in-place
  rename). `zipalign -f -p 4` then `apksigner sign` finish it, signed
  with `~/.config/ai-app/release.jks` by default or
  `~/.android/debug.keystore` under `--debug`. The signed APK is
  copied to `xtask/build/outputs/apk/<mode>/ai-app-shell-<mode>.apk`
  as a final step -- a Gradle-shaped path (`*/build/outputs/apk/*/*.apk`)
  chosen so Dev Updater's fixed-pattern APK discovery
  (`~/repos/dev-updater/server/src/discover.rs`'s `APK_PATTERNS`,
  which has no per-component path override) finds it without any
  change on that side; the working files above it stay under
  `target/xtask/apk/`, an ordinary build-cache location (gitignored,
  along with `xtask/target/`).
  
  *Wired into `.dev-updater.ron`*: a second `Apk` component, `shell`,
  `build: "cargo xtask apk"`, `modes: ["release", "debug"]`, no `cwd`
  (defaults to the checkout root, which both the `cargo xtask` alias
  and the publish path above need -- `.cargo/config.toml`'s alias
  resolves its `--manifest-path` relative to the *invoking* working
  directory, not to where the config file lives, which is what ruled
  out giving this component its own `cwd`). Dev Updater's `ByMode`
  appends the chosen mode word as the command's last argument
  (`build-apk.sh`'s own interface, per that component's comment), so
  `main.rs` accepts bare `release`/`debug` as well as `--release`/
  `--debug` for typing by hand. The existing `app` component
  (`build-apk.sh`, Gradle) is untouched.
  
  *Verification.* `cargo fmt -- --check` and `cargo clippy
  --all-targets` clean, zero warnings, for `xtask` (host target --
  nothing in it is Android-specific; it *runs* `cargo ndk`, it isn't
  cross-compiled itself). `./run-tests.sh`: 127 `server` + 88
  `client-core` tests, unaffected, still passing. `apksigner verify
  --print-certs` on the xtask's release output confirms a V3 signer
  with `CN=ai-app` -- the same key `build-apk.sh` generates.
  
  *The two pass-condition proofs*, both on this checkout's own AVD
  (`ai-app-2`, GPU host, brought up and torn down within this
  session):
  - **Installs over the Gradle-built one.** Built the Gradle release
    variant first (`AI_APP_KEYSTORE=~/.config/ai-app/release.jks
    AI_APP_KEYSTORE_PASSWORD=$(cat
    ~/.config/ai-app/release.jks.password) ./gradlew
    :shellApp:assembleRelease` -- needed its own signing block added
    to `app/shellApp/build.gradle.kts`, copied from `androidApp`'s,
    since `shellApp` had none before this), installed it fresh
    (`adb uninstall com.example.aiapp.shell` first -- an older debug
    install from E3 testing was signed with a different key and
    `install -r` over it fails loudly with
    `INSTALL_FAILED_UPDATE_INCOMPATIBLE`, which is the correct,
    expected failure for a mismatched key rather than a bug), then
    `adb install -r xtask/build/outputs/apk/release/ai-app-shell-release.apk`:
    **`Success`**.
  - **The notification service starts.** Enrolled via
    `adb shell "am start -a android.intent.action.VIEW -d
    'aiappshell://enroll?host=10.0.2.2&port=<sandbox port>&token=<token>'"`,
    force-stopped the app, then re-launched it once (enrollment calls
    `sync()` from `MainActivity.onCreate`). `adb logcat` shows
    `ActivityManager: Background started FGS: Allowed ... intent:
    ... cmp=com.example.aiapp.shell/.NotificationService`, immediately
    followed by `android-shell: jni::vm::java_vm: Attached thread
    ai-app-notifications`, a real TLS handshake to the sandbox's
    `10.0.2.2:<port>`, and `Response { status: 200 ... }` on
    `/notifications`. Pressed home, spawned a sandbox session and sent
    it `/question Should E5 proceed?`; `adb shell dumpsys notification
    --noredact` then shows a live `NotificationRecord` for
    `com.example.aiapp.shell`, `channel=sessions`, `tag=<session id>`
    -- posted while the app held no visible activity, the same bar
    E3's own proof cleared.
  
  *Left undone, honestly.* No attempt to shrink the dex (R8/minify is
  off, matching `shellApp`'s existing `isMinifyEnabled = false`, so
  the APK carries the full unshrunk `androidx`/Kotlin-stdlib/coroutines
  graph -- about 5.2 MB signed with both ABIs, most of it native
  libraries and that dependency graph rather than this project's own
  code). No `--abi arm64-v8a`-only real-device install was attempted
  this session (no physical phone reachable from here); the emulator
  proof above is `x86_64` plus a cross-compiled but unexercised
  `arm64-v8a` `.so` in the same APK. Multidex is unneeded at today's
  size but nothing in `dex()` checks for the 64k-method ceiling should
  the dependency graph grow.
  

The iris track

These build iris up to carry the app. Each is a feature added to iris with a pass condition, in dependency order. Work in iris/ in this repository on the rustify branch, and record in this file what each step measured.

• I0a — where iris lives (decided 2026-09-04). For now it is vendored at iris/ in this repository, history not carried, and consumed by path. Iris's decision: keep it close while it is being reshaped for this app, and give it back its own repository — iris/iris on the gitea remote, which already holds the full 244-commit history, on a branch of its own — once it has proved itself. The vendored tree is that repository's main at 7b54aaf ("readme", 2026-01-29), byte-identical to the public GitHub copy, so a later reconciliation has a known base. A crate that uses it says iris = { path = "../iris" }.

• I0b — make it build here (done 2026-09-04). iris now builds, clippy-clean and rustfmt-clean at the defaults, on a pinned dated nightly, and the tabs example draws on this VM's GPU.

  **The pin** is `nightly-2026-09-03` (rustc 1.100.0-nightly,
  `2e2b193f8`), declared in `iris/rust-toolchain.toml` along with the
  `clippy`/`rustfmt` components and the two Android targets, so a
  fresh clone provisions itself. It is dated rather than `nightly`
  because the whole failure below was a rolling channel moving under
  an unattended build. Installed with `--profile minimal`: 912 MB.
  
  **The 36 errors were one syntax change, and the earlier diagnosis in
  this file was wrong.** It is not that a trait must now be declared
  `const trait` — the vendored tree already declares them that way,
  which is how it was written in January. What changed is the *impl*
  keyword order: `impl const Trait for T` is now
  `const impl Trait for T`, and generics go on the `impl`
  (`const impl<T: [const] Foo> Bar for T`). Bounds are unaffected;
  `T: const Foo`, `T: [const] Foo` and `impl const Foo` in argument
  position all still compile. Everything else — the unresolved
  `UiVec2`/`Vec2`/`impl_op` imports, and a `Color<u8>` that resolved
  to `wgpu_types::Color` — cascaded from the seven files that failed
  to parse. The rewrite was mechanical across 20 sites and took the
  workspace from 36 errors to 0.
  
  **`#![feature]` gates, 12 after this step** (two were declared and
  unused, and were removed: `map_try_insert`, `const_cmp`).
  Load-bearing and worth watching: `const_trait_impl`, `const_ops`,
  `const_convert`, `const_destruct` are the const-traits family and
  the one that has already broken once — they move together, so
  advancing the pin means re-reading this section. `unboxed_closures`
  + `fn_traits` (postfix builder API) and `unsize` +
  `coerce_unsized` (widget handles) are pairs. The rest are
  individually small: `macro_metavar_expr_concat`, `portable_simd`,
  `associated_type_defaults`, `option_into_flat_iter`, and `gen_blocks`
  in the top crate.
  
  **Running it headless.** `iris/run-headless.sh EXAMPLE [--shot PNG]`
  with `iris/headless.conf`, the same trick `emu` uses: a headless
  sway, and `grim` for the picture. It deliberately starts its *own*
  compositor rather than joining `emu`'s — sway tiles, so adding a
  window to the one an emulator sits in resizes that emulator.
  Unlike `emu`'s it disables Xwayland, since winit speaks Wayland.
  **This VM has a real GPU for this**: Vulkan 1.4 through Venus onto
  the host's RX 7900 XT, and GL 4.6 through virgl — so desktop wgpu
  work here is not software-rasterised, unlike inside the emulator.
  
  **iris has no tests at all** (`cargo test --workspace`: 0 passed
  across 6 targets). Nothing to keep passing, and nothing to catch a
  regression — worth knowing before I1 changes the text stack.
  
  **`iris-core` no longer depends on winit, and now cross-compiles to
  Android.** It wanted exactly one thing from it — `PhysicalSize<u32>`
  in `UiRenderNode::resize`'s signature, for two numbers it immediately
  turned into floats — and that pulled a whole windowing backend into
  the layer below it, the wrong direction. `resize` takes
  `impl Into<Vec2>` now, like `UiRenderState::resize` beside it already
  did. The consequence is the point: with winit in the graph an Android
  build of the core failed in `android-activity` (which needs a backend
  feature nothing here selects), and without it
  `cargo ndk -t arm64-v8a -P 26 build -p iris-core` finishes in 30s and
  produces an rlib, wgpu's Android backend included. So **iris's
  widget, layout and render core already builds for the phone**, and
  what I2 has to supply is the surface, the input and the IME — not a
  port of the library.
  
  **Build weight, cold, on this VM's 8 cores** (`rm -rf target`, then
  `cargo build --example tabs`), since "the Linebender stack is slow in
  debug" was the worry behind this question: plain debug **43s** and a
  2.1 GB `target/`; with the `[profile.dev.package."*"] opt-level = 2`
  knob, **1m46s** and 1.5 GB. So iris's own wgpu + winit + cosmic-text
  graph is not the slow thing — which makes it a calibration for E1
  rather than an answer about Masonry, whose graph adds Vello, Parley,
  Fontique and Skrifa. Runtime cost of the knob was not measured here.
  
  **Fixed: iris never called `pre_present_notify`.** The symptom was
  that about one start in five kept the window's 800x600 startup layout
  on a 1920x1200 surface for good. What settled it was tracing iris's
  own decisions into memory and dumping them from another thread —
  `eprintln!` in the draw path makes the defect vanish, which is why
  earlier attempts kept losing it. The traces from a good and a bad run
  are **byte-identical**: both lay out and draw `redraw_all at
  (1920, 1200)` into a 1920x1200 texture with `suboptimal=false`. iris
  was drawing the right frame every time; the compositor was still
  showing the first one, and forcing a full repaint did not shift it.
  What was missing is winit's `Window::pre_present_notify`, called
  immediately before `present`, which on Wayland is what ties the
  commit to the surface's frame callback. Without it a frame drawn with
  nothing following it can sit unpresented with nothing left to flush
  it — which is exactly a window that has just settled after its
  opening resize. Measured: **0 bad in 40** with the fix, against 4 in
  20 before it, and — the stronger evidence — 0 in 20 in the
  instrumented configuration that had been 15 in 20. Runtime resizing
  still round-trips to a byte-identical layout.
  
  Two things ruled out on the way, both worth not re-trying: the
  present mode (the fault survived the move from `AutoNoVsync` to
  `AutoVsync` at the same rate) and the size cache (`redraw_all` clears
  it). A `desired_maximum_frame_latency` of 1 moved the rate without
  fixing it, and was reverted. Iris's own note that she had never seen
  the library fail to resize was the useful steer: it pointed away from
  the layout code, where two hours had already gone.
  
  One thing was fixed on the way, and it is not that bug: `update`
  redrew everything when `resized` was set, but `needs_redraw` — which
  is what decides whether to *ask* for a frame — did not know about
  `resized` at all. The two now share one `needs_redraw_all`, since a
  condition in one and not the other is a frame nobody requests. It is
  latent on Wayland only because winit asks for a redraw after a resize
  by itself; on Android, where the surface work of I2 will not have
  winit underneath it, nothing else here would have asked.
  

• I1 — parley, and a glyph atlas (done 2026-09-04). No bake-off: Iris decided for parley directly ("I wanted to switch it to parley anyways"), and then asked for the atlas as well ("just do the atlas, commit to it, we do want it"). Both are in.

  **What parley bought, beyond shaping.** Its editing model addresses
  text by byte offset into one string, where cosmic-text used
  `(line, index)` — so `select_content`, `delete_between`,
  `insert_inner` and `newline` collapse into ordinary string
  operations. Bigger: `Selection::geometry` and `Cursor::geometry`
  replace `iter_layout_lines`, `index_x` and `cursor_pos`, which walked
  runs by hand to place the caret and the selection boxes and were not
  bidi- or wrap-correct. `edit.rs` lost about 130 lines and gained
  Home/End. Its cursor motions map onto parley's `next_visual`,
  `previous_visual_word`, `next_line` and so on, in one function.
  
  **The atlas is what "text resizing (per frame) is really slow"
  was.** Every string used to be rasterised into its own `RgbaImage`
  and uploaded as a whole texture whenever anything about it changed,
  so a window resize re-rasterised and re-uploaded every visible
  string. Now a glyph is rasterised once per font, size and subpixel
  phase, shared by every string that contains it, and a resize
  re-emits quads without touching the GPU's copy. **The tabs example
  reports it: `views`, the number of texture views bound, went from 6
  to 1** — six per-string textures became one shared page. Supporting
  pieces: a `GLYPH` primitive that samples a sub-rectangle and tints
  it (the existing texture primitive samples a whole texture), a
  `Patch` texture update so a new glyph costs its own bytes rather
  than a 4 MB page, and `GpuTextures` keeping its `Texture`s, since a
  view cannot be written through.
  
  **Not yet measured**, and the honest gap in this step: the TODO's
  "really slow" was never given a number, so neither is the
  improvement. What is evidence rather than argument is the view count
  and the shape of the work — a resize no longer rasterises. A
  before/after timing wants the transcript screen of I5 to be worth
  taking.
  
  **Two bugs found on the way**, both pre-existing: `primitives!`'s
  `@count` rule recursed comma-separated while matching
  space-separated, so it terminated only for exactly two primitives
  and adding a third hit the recursion limit; and `Color` had no
  `Default`, which parley's `Brush` requires.
  
  **Fourteen tests**, iris's first. The editor is the one part that is
  pure logic rather than something needing a GPU and a window, and it
  was rewritten wholesale with no way to exercise it — synthetic input
  does not reach a client under the headless compositor, which has no
  seat devices. Two of the tests are aimed at what the rewrite could
  plausibly have broken: the IME preedit path, and editing multi-byte
  text now that offsets are bytes.
  
  Dropping cosmic-text and unicode-segmentation also retired two
  nightly gates — `portable_simd` (the old glyph compositing) and
  `gen_blocks` (the deleted line iterator). **Eleven left.**
  

iris's binding array does not survive real Android hardware (found 2026-09-04, resolved 2026-09-04)

Resolved the same day: see "Where things stand" above and TEXTURES.md's "Implemented, 2026-09-04". The measurement and sourcing below are unchanged and are why the fix looks the way it does; nothing here needs re-checking on its own account.

Iris asked, of the "unknown number of images" case — a transcript with an unbounded number of attached screenshots — whether iris's approach even works on a phone, since her recollection was that mobile does not support it. Checked rather than assumed, and the recollection is right, with sources rather than a guess.

What iris does today. Every texture — every Image widget (src/widget/image.rs) and every glyph atlas page — gets its own permanent slot in one array via Textures::add (core/src/primitive/texture.rs:65), and both the TEXTURE and GLYPH primitives sample it by view_idx into binding_array<texture_2d<f32>> at core/src/render/shader.wgsl:56, sized by UiLimits::default — 100,000 textures, 1,000 samplers (core/src/render/mod.rs:347). That needs three wgpu features: TEXTURE_BINDING_ARRAY, SAMPLED_TEXTURE_AND_STORAGE_BUFFER_ARRAY_NON_UNIFORM_INDEXING, PARTIALLY_BOUND_BINDING_ARRAY — Vulkan's VK_EXT_descriptor_indexing ("bindless"), promoted to core in 1.2. So a transcript with an unbounded number of images is exactly the case that grows this array without bound, one permanent slot per image.

Measured first on the emulator, and it fails outright. A rig (rigs/gpu-probe, a plain executable with no window, pushed with adb push and run from /data/local/tmp — no APK needed to ask a device what it supports) asks wgpu::Adapter::request_device for exactly iris's features and limits. Against the emulator's guest Vulkan — both SwiftShader (vk_swiftshader_icd.json) and lavapipe (lvp_icd.json, cold-booted) — request_device fails: Unsupported features were requested: TEXTURE_BINDING_ARRAY | SAMPLED_TEXTURE_AND_STORAGE_BUFFER_ARRAY_NON_UNIFORM_INDEXING | PARTIALLY_BOUND_BINDING_ARRAY. A second, raw query through ash (rigs/gpu-probe/src/vk.rs, bypassing wgpu) shows lavapipe's vkGetPhysicalDeviceFeatures2 actually reporting all seven descriptor- indexing sub-features as true at device api version 1.3 — so on this software renderer wgpu-hal's own feature detection is being more conservative than the driver, for a reason not chased further (a likely instance-version negotiation gap, since VK_EXT_descriptor_indexing was only promoted to core at 1.2 and wgpu-hal's own Instance::init may be requesting less). That part is an emulator/wgpu-hal question and not the finding that matters.

The finding that matters is about real phones, not the emulator, and it is sourced rather than recalled. The Android Vulkan Profile 2025 — Google and Khronos's current baseline, covering 80.1% of active Vulkan-capable Android devices as of October 2025 (developer.android.com/ndk/guides/graphics/android-vulkan-profile) — does not require VK_EXT_descriptor_indexing or any descriptor- indexing feature. It requires shaderSampledImageArrayDynamicIndexing (indexing an array of samplers by a value uniform across the invocation — Vulkan 1.0 baseline, unrelated to bindless) and stops there; the same is true of the 2021 and 2022 profiles. On the hardware side, Arm's own developer documentation states "VK_EXT_descriptor_indexing is supported on all Valhall and 5th Gen GPUs" (developer.arm.com/mobile-graphics-and-gaming/vulkan-api-best-practices-on-arm-gpus) — Mali generations from roughly 2019 (Mali-G77) onward, named affirmatively with no claim made for Bifrost, Midgard or Utgard, which are still common in budget and older Android phones still in use. So this is not a software-renderer artifact: a real, currently-shipping share of the Android fleet lacks the feature iris's texture pipeline asks for unconditionally, and the newest official baseline does not promise it either. (A crates.io/search-engine claim of "1% support on Android" for this extension was checked against its cited source, an Arm blog post, and was not actually there — that number does not appear anywhere primary and should not be repeated; the 80.1%-baseline-excludes-it finding above is the one with an attributable source.)

Recommendation, not yet implemented. iris already solved the identical problem for text in I1: the glyph atlas (core/src/render/atlas.rs) packs many small rasters into a handful of shared 1024×1024 pages and samples them by UV offset, so text needs none of the three features above — only ordinary single-texture sampling. The same technique generalizes to images: route an Image widget through a shared atlas when it is small enough to pack (thumbnails, downscaled attachment previews, avatars, icons), and fall back to one ordinary, non-array texture bind group — selected per batched draw call the way every immediate-mode 2D renderer already does — for anything too large to atlas well (a photo opened at full resolution). Either path is plain Vulkan 1.0 / GLES texture sampling, so it removes the descriptor- indexing requirement from iris's device request entirely, which is also what would make the emulator work regardless of the wgpu-hal question above: a device that never asks for the feature cannot be refused for lacking it. This is a change to iris's rendering core — the shader's binding group layout, Textures, the texture and glyph primitives, and ui/painter.rs — so it is written here as a recommendation rather than started, per the project's rule to confirm a load-bearing design change before making it. It should be resolved before I2 is called done, since I2's pass condition is the phone, not just the emulator, and this is exactly the kind of thing that passes on a desktop GPU and fails silently on real hardware.

• I2 — iris on android-view (2026-09-05). The android-view backend, the iris-android-app cdylib and Gradle shell, insets, the back gesture and the full InputConnection bridge are in and measured working; the tabs example now renders on the emulator (Vulkan/ SwiftShader and GLES/virgl both), and the composer's keyboard shows real Gboard suggestions through the IME bridge. See below for the render-gap root cause and fix.

  **Layout.** `iris/src/android/` mirrors `default/`'s module split
  (`view.rs` is `app.rs`+`state.rs` combined, since android-view has one
  harness type where winit splits `ApplicationHandler` from per-window
  state; `render.rs`, `input.rs`, `attr.rs` correspond directly;
  `ime.rs` and `insets.rs` have no winit counterpart). What used to live
  only in `default/` and had no winit dependency — `WidgetState`,
  `CursorState`/the sense machinery, `Tasks`, `Selector`/`Selectable`'s
  focus handling — moved to crate-root modules (`state.rs`, `sense.rs`,
  `task.rs`, `attr.rs`) so both backends use one copy; `Tasks`' redraw
  nudge is now behind a `RequestRedraw` trait (`Window` for winit, a
  `JavaVM`+`GlobalRef` attach-and-call for android-view) rather than a
  concrete `winit::window::Window`. `winit`/`arboard` and
  `android-view`/`send_wrapper` are now `[target.'cfg(...)']`
  dependencies, and `default`/`android` are target-gated modules,
  because winit's own Android support needs `android-activity` with a
  backend feature selected — exactly what `iris-core` was kept free of.
  Confirmed by trying it before the split (`cargo ndk -t x86_64 -P 26
  build -p iris` failed inside `android-activity` itself) and after
  (clean). `iris/tabs-ui` is the tabs example's widget tree factored out
  of `examples/tabs/main.rs` into a crate generic over `Rsc: HasEvents`
  + `Rsc::State: FocusHost`, so the winit example and
  `iris/android-app` (the new cdylib, excluded from the `iris` workspace
  because android-view needs the NDK sysroot to link — see that
  `Cargo.toml`'s comment) call the same `build()`.
  
  android-view pinned to `bec6c62a96cef8239b0fd7fedeef9b184d02e3a1`, the
  commit E1 measured against. `RustView.java`/`RustInputConnection.java`
  are vendored (no published AAR to depend on) into
  `iris/android-app/app/src/.../org/linebender/android/rustview/`, with
  one deliberate diff from upstream noted in a comment: `mViewPeer` is
  `protected` rather than package-private, so `IrisView` (a different
  package) can pass it to the window-insets native call android-view
  has no hook for.
  
  **Insets and the back gesture**, both without touching android-view.
  The back gesture takes no new plumbing at all: with no
  `OnBackPressedCallback` registered, Android still delivers it as an
  ordinary `KEYCODE_BACK` `KeyEvent` through the existing key path (the
  legacy behaviour every view-based app gets by default), handled in
  `view.rs`'s `on_key_down`. Insets have no such stand-in, so
  `android/insets.rs` registers one more native method
  (`applyWindowInsetsNative`) directly on `IrisView`, writing into an
  `Rc<RefCell<Shared>>` a second copy of which lives in
  `AndroidUiState` — the peer id android-view hands back from
  `register_view_peer` is opaque outside that crate, so this is a
  side table keyed on the same id rather than a way to reach the peer
  itself. `MainActivity` wires `setOnApplyWindowInsetsListener`,
  including the API 30+ `ime()` inset specifically (falls back to 0
  below that). Not yet consumed by any widget's layout — `insets()` is
  exposed on `AndroidUiState` but nothing reads it yet, since the tabs
  example has no chrome that needs to avoid the keyboard.
  
  **The IME bridge is implemented and its pass condition holds.**
  `android/ime.rs` implements the full `InputConnection` trait
  (`text_before_cursor`/`after_cursor`/`selected_text`,
  `cursor_caps_mode`, `delete_surrounding_text[_in_code_points]`,
  `set_composing_text`/`_region`, `finish_composing_text`,
  `set_selection`, `begin`/`end_batch_edit`, `send_key_event`,
  `request_cursor_updates`) directly against `TextEdit` — the same
  preedit-replace bookkeeping `default`'s `Ime::Preedit` handling uses
  (`compose_len`, in chars), with new byte<->UTF-16 conversion helpers
  since parley (since I1) is byte-indexed and Java strings are not.
  Two approximations, both commented in place rather than silently
  dropped: `set_composing_region` declines (no separate composing range
  exists to move) and `set_selection`/`delete_surrounding_text_in_code_points`
  collapse to an approximation rather than a real span/code-point
  count. `TextEdit` gained `text()`/`selection_range()`/`caret()`
  getters and `TextEditCtx::delete_byte_range`/`set_cursor_byte`, all
  unconditional (no winit dependency added); `apply_event`/
  `TextInputResult`, which do take a `winit::event::KeyEvent`, are now
  `#[cfg(not(target_os = "android"))]` instead of being ported, since
  android's own `input.rs` calls `TextEdit`'s primitives
  (`backspace`/`delete`/`motion`/`insert`) directly from
  `ndk::event::Keycode` and never needed a winit `KeyEvent` shape.
  
  **Measured on the emulator, 2026-09-05, x86_64 API 26,
  `-feature Vulkan` + SwiftShader per the Vulkan section below.**
  `adb shell dumpsys input_method` after tapping the composer field:
  `mInputShown=true`, `mServedInputConnection` is
  `org.linebender.android.rustview.RustInputConnection` attached to
  `IrisView`. `adb shell input text "hi"` followed by a screenshot
  shows **Gboard's suggestion strip populated with "hi | Hi | HI"** —
  capitalization variants read back out of the real buffer through
  `text_before_cursor`, the same kind of evidence E1 recorded (there:
  "dolor | Dolores | door"). That is the bar this box asks for, met.
  
  **Resolved 2026-09-05: the render gap was the window uniform, never
  the atlas.** `UiRenderNode::new` (`core/src/render/mod.rs`) seeded the
  GPU's `window_buffer` from `WindowUniform::default()` — width=0,
  height=0 — and the only thing that ever corrected it was a later call
  to `UiRenderNode::resize`, renamed `AndroidRenderer::resize` on the
  android side. winit's backend gets away with the same default because
  winit fires an initial `WindowEvent::Resized` before the first frame,
  which `default/mod.rs`'s event loop turns straight into that resize
  call — a real event this project never had to add on purpose, so
  nothing here noticed the node depended on it. android-view has no such
  automatic event: `surface_changed` (`src/android/view.rs:363-388`)
  only calls `self.render.resize(...)`, which is
  `UiRenderState::resize` — the CPU-side *layout* width the widget tree
  lays out against — not `AndroidRenderer::resize`, which is the one
  that writes the GPU uniform. `AndroidRenderer::new` builds a fresh
  `UiRenderNode` with the correct `SurfaceConfiguration` (so the surface
  itself was always the right size, and the clear colour reached it) but
  that node's window buffer was never subsequently written, so it sat at
  `(0, 0)` for the node's entire life. `shader.wgsl`'s `vs_main` divides
  by `window.dim` to reach clip space
  (`let pos = (top_left + uv * size) / window.dim * 2.0 - 1.0;`), so
  every primitive's clip position came out `NaN`/`Inf` and was dropped
  before rasterization on **both** backends — Vulkan and GLES alike,
  exactly the cross-backend symmetry that should have pointed away from
  a GL-specific cause sooner. The layout engine reporting the correct
  widget count and pixel region the whole time is consistent with this:
  that path never touches `window.dim` at all, since it is a separate
  copy of the window size (`UiRenderState`'s own, fed by
  `self.render.resize`) that the CPU-side layout and hit-testing use.
  
  **The GLES `D2`/`D2Array` warning was confirmed a red herring.**
  Reproduced again after the fix, unchanged, on a build forced to
  `Backends::GL` — it fires on every frame regardless, and primitives
  draw correctly on that backend anyway (screenshot below), so it is a
  cosmetic wgpu-hal heuristic notice, not a correctness bug in the atlas
  path. Left as-is; chasing it further is not warranted.
  
  **Fix** (`core/src/render/mod.rs`, `UiRenderNode::new`): seed
  `WindowUniform` from `config.width`/`config.height` — already the
  surface's real size at construction time on both backends — instead
  of `WindowUniform::default()`. This removes the dependency on an
  external resize call entirely (winit's initial `Resized` event still
  fires and still calls `resize()`, now idempotently) rather than
  papering over android-view's missing event with one more call in the
  android-specific path; a future third backend gets a correct window
  buffer from its first frame with no equivalent event of its own to
  remember.
  
  **Verified on the emulator, 2026-09-05, `ai-app-2`'s own AVD, x86_64
  API 26, `-feature Vulkan` + SwiftShader per the Vulkan section.**
  `logcat` after launch: `render(): after update active=39
  root_px=Some(PixelRegion { top_left: (0, 0), bot_right: (1080,
  2219) })`, no wgpu validation warnings on the Vulkan build. Screenshot
  (`/tmp/iris_i2_render.png`) shows the tabs example's coloured spans,
  the red rounded rect and the tab bar all drawn — the milestone this
  section asked for. Rebuilt with `Backends::GL` forced (reverted
  afterwards; the shipped code still requests `Backends::PRIMARY`) and
  reinstalled: same screenshot, same widgets, `AdapterInfo` logged as
  `Android Emulator OpenGL ES Translator (virgl (AMD Radeon RX 7900
  XT...` confirming the real GLES/virgl path, with the `D2`/`D2Array`
  warning present and harmless as above. Text glyphs render with visible
  artifacting on the GLES path specifically (not investigated further —
  out of scope for this box, which is about primitives appearing at
  all, and it does not affect the Vulkan path this app ships behind).
  
  **Not built yet**: anything consuming `insets()`, a real phone
  measurement (only the emulator so far — matches every other Android
  finding in this file), and AccessKit (I4's job, so `ui-trace`
  couldn't be used here; a raw `adb shell input tap`/`input text` stood
  in for driving the UI, which is why this section says "the same bar
  as E1" rather than citing a `ui-trace` transcript).
  
  **Verification.** Host: `cargo fmt --all -- --check`,
  `cargo build --workspace --all-targets`, `cargo clippy --all-targets`,
  `cargo test --workspace` (19 tests) all clean in `iris/`; `iris/run-headless.sh
  tabs --shot` still renders pixel-identically (27266 bytes, byte-for-byte
  unchanged). Android cross-compile: `cargo ndk -t x86_64 -P 26 build`
  and `... clippy` clean for both `iris` (with the android module) and
  `iris/android-app`. Emulator: `emu up` with
  `VK_DRIVER_FILES=.../vk_swiftshader_icd.json` and
  `GPU_HOST_FEATURES="-feature Vulkan -no-snapshot-load -no-snapshot-save"`
  per the Vulkan section; `cd android-app && cargo ndk -t x86_64 -P 26
  -o app/src/main/jniLibs/ build --release && gradle :app:assembleDebug`
  (release native lib per E1's segfault finding, debug Gradle variant --
  the jniLibs contents are what matters, not the Gradle build type);
  `adb install -r app/build/outputs/apk/debug/app-debug.apk`. Emulator
  torn down after verification (`emu down`) per the machine's memory
  rule.
  

• I3 — a virtualised, bottom-anchored list (2026-09-05). Variable-height rows, keyed, composed only while visible, paged in both directions with a "more" sentinel at each end, a scroll anchor that survives rows being inserted above, and "hold the edge nearest the tap" done in the layout pass. Built as iris::widget::List (iris/src/widget/list.rs, its module doc is the design writeup) -- see IRIS.md's 2026-09-05 entry for the public API and the one correctness lesson worth carrying elsewhere (a fill-shaped background cannot be measured at a throwaway oversized region and merely repositioned into place; it has to be placed at its cached real size, or measured-then-redrawn via draw_twice on first appearance).

  **Done**: the widget, 6 unit tests (`cargo test -p iris`, anchor and
  edge-hold logic, all pure -- no GPU/window needed, same harness as
  `layout_tests.rs`), `iris/benches/message_list.rs` rewritten to
  measure the real widget instead of a hand-built `Span`+`Scroll`, two
  new benchmark scenarios ((d) insert-above-anchor, (e)
  expand-a-row-holding-its-edge), and `iris/examples/message_list.rs`
  (800 rows, varied wrapped-text length, one in twelve with an image,
  mouse-wheel scrollable) rendered via `run-headless.sh` and visually
  verified (cropped with a throwaway PNG decoder, since this VM has no
  image tooling -- see the commit for the crop script's shape).
  
  **Numbers (2026-09-05, release, this VM), all flat across N =
  100/1,000/10,000 as required:**
  
      cd iris && ./run-bench.sh list
      (a) first frame:                 ~12.3-12.9ms  draws=80   rewrites=3   moves=0
      (b) scroll, 200 ticks:            4.8-6.5ms    draws=328  rewrites=12  moves=10131  (~0.025-0.033ms/tick)
      (c) input grows, 40 lines:        8.9ms        draws=1846 rewrites=102 moves=1195   (~0.22ms/line)
      (d) insert-above-anchor, 200 pushes: 0.4ms      draws=200  rewrites=0   moves=0      (~0.002ms/push)
      (e) expand-hold, 40 growths:      0.10-0.11ms  draws=119  rewrites=40  moves=15     (~0.003ms/growth)
  
  (d) is the cleanest confirmation: 200 rows prepended one at a time
  while scrolled to the loaded window's start cost 200 draws total (the
  list widget's own redraw each push) and **zero** row draws or moves
  -- none of the prepended rows ever entered the viewport, exactly as
  the anchor-by-slot-index design predicts. (e) similarly stays tiny
  and flat: growing one row 40 times, each preceded by `note_tap` at
  its own edge, costs a total of 15 moves (the rows on the far side of
  the held edge) regardless of how many thousand rows exist elsewhere
  in the list.
  
  **Verification.** `cargo fmt --all -- --check`,
  `cargo build --workspace --all-targets`,
  `cargo clippy --all-targets` (and `--benches --release` separately,
  since benches aren't always covered), `cargo test --workspace` (25
  passed) all clean in `iris/`.
  
  **What remains — the emulator half of the pass condition, blocked on
  the emulator being held by another session during this pass.** The
  condition as written ("800 rows of real transcript text from the
  sandbox scroll without a frame over the Compose baseline in
  `transcript-bench.sh`, measured on the GPU emulator") needs the
  transcript screen actually rebuilt on top of `List` (this box only
  built and measured the widget in isolation, per the task scope) and
  then driven through the real emulator rig. Once that screen exists,
  the exact command is:
  
      cd app && ./transcript-bench.sh -k    # or without -k for a fresh session
      # compare its render report against the iris build's equivalent
  
  This is a genuinely separate step (wiring `List` into an actual
  session screen, i.e. most of I5's work) rather than something this
  box's scope could finish alone -- recorded here rather than left
  silently undone.
  

• I4 — accessibility names via AccessKit, host half done and verified 2026-09-05; the emulator half done and verified 2026-09-05, same day as I5's Android integration (see bottom of this box for the exact run). Built iris_core::ui::access::AccessTree (iris/core/src/ui/access.rs) -- one flat AccessKit tree, a synthetic Role::Window root with every named widget as a direct child. Deliberately flat rather than mirroring iris's real widget nesting: nothing upstream of a named leaf needs a node, since a screen reader's traversal (and uiautomator's tap-by-name, this box's own pass condition) works from each node's on-screen bounds, not from tree structure -- and mirroring the real tree would rebuild intermediate nodes on every resize of any container above a named widget, which is most frames.

  **Modular the way input's sense registry is.** `Widgets` gained one
  `HashSet<WidgetId>` (`named`), populated only by `.label()`/
  `set_label` and drained by `free_next` (the same removal path a
  freed id already went through -- no second bookkeeping call added
  anywhere). `AccessTree::update` walks `widgets.named()` directly,
  never the full widget arena, so a widget nobody named costs this
  subsystem nothing -- not a visit, not a branch. Roles come from a
  new `Widget::access_role(&self) -> accesskit::Role` trait method,
  default `Unknown`; the one override so far is `TextEdit` ->
  `TextInput`/`MultilineTextInput` by `EditMode`. Bounds come from
  `UiRenderState::window_region`, which sits on `resolved_region`'s
  move-chain walk -- so a widget moved via `Offset`/`Scroll` (never
  redrawn from scratch) still reports where it actually ended up; see
  `bounds_follow_a_moved_widget_and_updates_stay_incremental` below.
  
  **Incremental, not per-frame.** `AccessTree` keeps the last
  `HashMap<WidgetId, Entry>` (name, role, bounds) it sent and only
  returns a new `TreeUpdate` -- and only then bumps its `rebuilds`
  counter, `take_rebuilds()`'s the AccessKit twin of
  `UiRenderState::take_counters` -- when that set actually differs.
  Confirmed by `bounds_follow_a_moved_widget_and_updates_stay_incremental`
  (`iris/src/access_tests.rs`): 1 rebuild on the first draw, 0 across an
  unchanged frame, 1 more after a real move, regardless of how many
  other widgets are on screen.
  
  **`SlotId::as_u64`** (`core/src/util/slot.rs`) encodes a `WidgetId`
  into accesskit's flat `NodeId(u64)`, offset by one so a real widget
  never collides with the reserved window node (`NodeId(0)`).
  
  **Pushed through two backends, each behind an inert action/activation
  handler** -- see below for why inert is correct, not incomplete.
  `default/access.rs` (winit): `accesskit_winit::Adapter`, built in
  `DefaultApp::new` with the window created hidden
  (`with_visible(false)`) and shown only after the adapter exists,
  which is what that constructor requires. `process_event` runs on
  every `WindowEvent`; `update_if_active` runs once per
  `RedrawRequested`, after `render.update()` so bounds reflect the
  frame just drawn. `android/access.rs` (android-view):
  `accesskit_android::Adapter` on `AndroidUiState`, `IrisViewPeer` now
  implements `AccessibilityNodeProvider`
  (`create_accessibility_node_info`/`find_focus`/`perform_action`), and
  `render()` (now taking `&mut CallbackCtx`, needed for the JNI handle
  any `raise` requires) pushes the same `AccessTree::update` after
  every draw.
  
  **Why the `ActionHandler`s are empty, not a placeholder for later
  work**: AGENTS.md's own "Driving the UI" section says it plainly --
  `ui-trace record --do "tap 'Save'"` resolves the label against the
  screen and performs a **real touch at that node's bounds**, the same
  as a person's finger. It does not call into AccessKit's action
  system at all. So once `AccessTree` reports correct bounds, the
  ordinary pointer path (already built, already tested) is what
  answers the tap -- there is nothing for `do_action` to do for this
  pass condition specifically. A future real screen reader's own
  double-tap-to-activate gesture works the same way, for the same
  reason. If iris ever needs to answer an AccessKit `Action::Click`
  injected without a matching touch (e.g. a switch-access scanner),
  that is new scope, not a gap in this box.
  
  **E1's abort mitigation, carried.** `android/access.rs`'s
  `raise_if_enabled` is the one place `QueuedEvents::raise` may be
  called: it asks `AccessibilityManager.isEnabled()` (a `getSystemService`
  JNI call, since android-view has no ready-made wrapper) immediately
  before every `raise` and drops the events instead when the answer is
  no. Every call site (`render`'s per-frame push, `perform_action`)
  goes through it, and each pushes it as a *deferred* callback exactly
  like android-view's own demo, so it runs after the current JNI
  callback has released whatever it's holding -- `raise`'s own
  documented requirement. Not independently re-triggered on this
  pass (that needs the emulator, see below); the mitigation is coded
  to the exact mechanism E1 diagnosed (`sendAccessibilityEvent`
  throwing when accessibility is off) rather than to the symptom, so
  there is no reason to expect it behaves differently here than it did
  there.
  
  **Verified, 2026-09-05, host only.**
  `cargo fmt --all -- --check`, `cargo build --workspace --all-targets`,
  `cargo clippy --all-targets` (both plain and `--all-targets`) clean;
  `cargo test --workspace` -- 28 tests in `iris/`, three of them new
  (`access_tests::a_named_widget_reaches_the_tree_with_its_role_and_bounds`,
  `::a_widget_with_no_label_never_reaches_the_tree`,
  `::bounds_follow_a_moved_widget_and_updates_stay_incremental`).
  `cd iris/android-app && cargo ndk -t x86_64 -P 26 build` and
  `... clippy` clean for both `iris` (with the android module) and
  `iris-android-app`, same shape as I2/I3's checks.
  `iris/run-headless.sh tabs --shot /tmp/iris_i4_tabs.png --seconds 4`
  still renders -- **27266 bytes, byte-for-byte identical to I2's own
  post-fix screenshot** -- confirming the hidden-window-then-adapter
  change to `DefaultApp::new` cost nothing visible. `tabs-ui`'s five
  switch buttons (`tabs-ui/src/lib.rs`) now carry `.label()`s matching
  their on-screen text ("pad", "span", "image span", "text layout",
  "text edit scroll") -- both so the desktop run above exercises a
  non-empty tree and so the emulator step below has real names to tap.
  Not independently checked on this pass: whether `accesskit_winit`'s
  Linux path (AT-SPI, via `accesskit_unix`) actually reaches a real
  assistive-technology client on this VM's headless sway -- there is
  no AT-SPI registry running here, so `default/access.rs`'s handlers
  are exercised as inert code paths (built, called, no panic) rather
  than confirmed end-to-end the way the emulator step below confirms
  the Android path.
  
  **Done, 2026-09-05, on this checkout's own emulator (`ai-app-2`,
  `EMU_GPU=software` -- see I5's box for why plain `-gpu host` and the
  documented Vulkan-feature recipe both could not be used here).**
  `cargo ndk -t x86_64 -P 26 -o app/src/main/jniLibs/ build --release &&
  gradle :app:assembleDebug`, installed, launched, then each of
  
      ui-trace record --do "tap 'pad'"
      ui-trace record --do "tap 'span'"
      ui-trace record --do "tap 'image span'"
      ui-trace record --do "tap 'text layout'"
      ui-trace record --do "tap 'text edit scroll'"
  
  resolved (uiautomator found the exact label every time -- `ui-trace`
  never failed a run). **Confirming the pane actually switched needed
  more than `ui-trace show`**: the tabs row is the *only* named
  structure on this screen, its five buttons never move, so
  `--field box` reports "nothing moved" on every run whether the pane
  behind it changed or not -- a screenshot before/after is what showed
  it, `adb exec-out screencap -p`, hashed to confirm difference; also
  confirmed live with a temporary `log::debug!` in `switch_button`'s
  click closure (reverted before committing) showing the exact index
  clicked matching the tapped label. The detach-abort check also
  passed: six consecutive `ui-trace record` calls against the same
  process (attach, detach, attach again, five more times) left it
  alive throughout -- `adb shell dumpsys window` still showed
  `dev.iris.android.demo/.MainActivity` focused and rendering
  afterward, no crash in `logcat`.
  
  **One real, unrelated bug found and fixed getting here, not part of
  I4's own design**: the release build was required -- a debug/dev
  profile build of this same APK reliably `SIGSEGV`s inside this
  emulator's Vulkan loader (`vulkan.ranchu.so`,
  `vk_common_SetDebugUtilsObjectNameEXT`) the moment `wgpu` creates its
  first bind group layout, because `wgpu`'s `InstanceFlags::
  from_build_config()` turns on debug object-labelling in a dev build,
  and labelling a `SwiftShader`-backed resource through this
  emulator's loader trampoline crashes. A release build's
  `InstanceFlags::empty()` never takes that path. Nothing in iris
  caused this and nothing here needed to change to avoid it --
  recorded because it looked exactly like a fresh regression the first
  time it was hit (mid-session, after adding an unrelated temporary
  log line forced a dev rebuild) and cost real time to separate from
  the actual touch-dispatch question being chased at the time.
  

• I5 — the transcript screen in iris (2026-09-05, updated later the same day, and again 2026-09-05 with the clean scroll comparison). The widget-tree half and the Android integration are both built and confirmed working on-device (real server, real scrolling, real touch-drag pan, tap-by-name), iris has its own frame-timing instrumentation (FrameReport), and long-press-then-drag-to-select is confirmed on-device (both by logcat and by a screenshot showing the highlighted selection). Ticked [x] now that a clean, single-session, like-for-like 24-swipe comparison against Compose exists -- see "Clean scroll comparison, 2026-09-05" near the end of this box for the numbers, what is and is not comparable between the two, and the dropout finding (this pass's own script bug, not a reproduction of the emulator touch-delivery candidate below).

  **Where it lives.** `iris/transcript-ui/` (new workspace member,
  `[lib]`), the same shape as `iris/tabs-ui`: generic over `Rsc:
  HasEvents` + `Rsc::State: FocusHost` so the same `build()` can run
  under winit (`transcript-ui/examples/transcript.rs`) or an
  android-view cdylib later. Depends on `client-core`/`event-model` by
  path (real code, matching E2's precedent) and `pulldown-cmark`
  (0.13.4, current stable). Four modules: `markdown.rs` (CommonMark ->
  plain text + `Vec<SpanStyle>`), `row.rs` (one `iris::widget::List`
  row per folded `TranscriptRow`), `selection.rs` (cross-row
  selection), `composer.rs` (the growing input field). `lib.rs`'s own
  module doc has the screen's shape and the one gap it documents up
  front (below).
  
  **New iris API, added in this box and recorded in `IRIS.md`:
  `SpanStyle`, per-range text styling.** This is the actual answer to
  RUST.md's E2 finding against Masonry ("rich inline text -- block-level
  yes, inline no, and both for the same reason":
  `masonry/src/widgets/text_area.rs:43-44`'s `TextArea::edit_styles()`
  returns one `StyleSet` for the whole editor, with `// TODO:
  RichTextInput` beside it). `core/src/primitive/text.rs`'s
  `TextBuffer` gained `spans: Vec<SpanStyle>` and `set_spans`;
  `SpanStyle{range, color, family, font_size, bold, italic,
  underline}` pushes into parley's `RangedBuilder` via `.push(property,
  range)` instead of only `.push_default(...)`, so one `TextEdit` can
  carry a heading's bigger bold font, an inline-code span's monospace
  colour, a link's colour+underline and an ordinary paragraph's base
  style all in the *same* wrapped, selectable buffer.
  `core/src/render/atlas.rs`'s `PlacedGlyph` gained a `color: UiColor`
  field (read from parley's own per-run `Style::brush`,
  `core/src/primitive/text.rs`'s `TextData::place`) and
  `core/src/ui/painter.rs`'s `glyphs()` now colours each glyph from
  that field instead of one colour for the whole `RenderedText` --
  the change that actually makes a span's colour reach the screen.
  **Real bug found and fixed while wiring this in**: `TextBuilder`'s
  `.spans(...)` was only threaded through `TextOutput::run` (the
  read-only `Text` widget), not the sibling `TextEditOutput::run` (the
  `TextEdit` every transcript row actually uses) -- a "rule that
  governs a set belongs to the set, not one member" miss, per
  CODE_RULES.md; found because `run-headless.sh`'s screenshot showed
  *no* styling at all despite `markdown.rs`'s own unit tests passing
  (they only check the string/range logic, not the render path -- see
  `iris/src/widget/text/build.rs`'s `TextEditOutput::run`, now fixed).
  
  **The seven behaviours, each shown or given a sourced reason, same
  structure as E2's own accounting:**
  
  1. **Selection spanning rows -- shown, with a scoped shortcut
     recorded rather than hidden.** `selection.rs`'s `Selection`
     coordinates each visible row's own `TextEditCtx::select`/
     `select_all`/`deselect` (already built for one field, I2) from a
     single drag that crosses row boundaries: rows between the anchor
     and the pointer get `select_all()`, the row under the pointer gets
     a true partial selection from whichever edge faces the anchor,
     and `selected_text()` concatenates the result in row order. The
     one shortcut: the *anchor* row is selected in full once the drag
     leaves it, rather than "from the click point to its far edge",
     because that needs the row's own laid-out size and
     `TextEditCtx`'s `layout()` helper is private
     (`iris/src/widget/text/edit.rs`) -- see `selection.rs`'s module
     doc. Pure range-membership logic (`in_range`, mirroring
     `begin`/`extend`'s row-selection arithmetic) is unit-tested
     without any render harness; the widget-level wiring is not
     independently screenshotted this pass (would need a synthetic
     drag injected into the winit example -- not attempted, time).
  2. **Rich inline text -- shown, genuinely inline this time.**
     `markdown::render_markdown` folds one row's whole markdown (not
     one block at a time) into one string plus spans, so a heading, a
     **bold** word, *italic* text, `inline code`, and a
     [link](url) inside the same paragraph render in one `TextEdit`
     that still wraps and selects as a single buffer --
     screenshotted, see below. Deliberately not attempted, each
     recorded at the point it would have gone in `markdown.rs`'s own
     doc: a background chip behind inline code (needs glyph-run
     geometry `TextEdit`-internal and not exposed, the same primitive
     `TextEdit::draw`'s selection highlight uses,
     `iris/src/widget/text/edit.rs:99`), a tappable link (same missing
     primitive), a real table layout, and per-token syntax colour
     inside a fence.
  3. **Bottom-anchored virtualised list, hold-the-edge on expand --
     shown**, reusing I3's `List` unmodified. A `TranscriptRow::Tools`
     row collapses to "N tool calls" and expands to every call's own
     tool/input/output on tap; `row.rs`'s click handler calls
     `List::extent(key)` to convert the tap's row-local position into
     the viewport-relative position `List::note_tap` wants, exactly
     the two-step contract `list.rs`'s module doc describes for
     `holdTopEdge`. Not independently screenshotted mid-expand this
     pass (no input-injection into the desktop example was built) --
     the mechanism is the same one I3 already benchmarked
     (`expand-hold`, flat at 0.10-0.11ms across N), applied to real
     content instead of a synthetic row.
  4. **The soft keyboard -- inherited from I2, not re-investigated.**
     The composer (`composer.rs`) is an ordinary `TextEdit` with the
     same `InputConnection` bridge I2 built and measured (Gboard
     suggestions over real buffer content); nothing new to add here,
     and no Android shell exists yet for this screen specifically to
     re-verify it against (see "What remains").
  5. **Platform integration -- out of scope by design**, same as E2:
     E3's list, not this box's.
  6. **Accessibility names -- shown for the composer, not yet for
     rows.** The composer field carries `.label("Message")` (I4). Rows
     do not yet carry per-row labels (a row's own text *is* its
     accessible content via `TextEdit`'s `access_role`, I4, but
     nothing calls `.label()` on it, so `Widgets::named()` does not
     include it) -- a small, real gap, recorded as an IRIS_TODO.md
     item rather than silently left, since AGENTS.md's bench scripts
     depend on exactly this for driving a screen by name.
  7. **Measurable frames / the render-number pass condition -- the
     gesture-conflict half is now fixed (2026-09-05); the emulator
     half is still not attempted, and unlike E2 that's not an absent
     gesture path.** `List` demonstrably scrolls (I3's flat
     draws/moves, programmatic `scroll()`) and mouse-wheel scrolling
     is wired here (`lib.rs`'s `CursorSense::Scroll` on `list`). What
     was *not* reachable at first was a **touch-drag pan starting on a
     row's own text**: `row.rs` registered `CursorSense::
     click_or_drag()` on each row's `TextEdit` for selection, and
     `TextEdit::draw` calls `painter.child_layer()`
     (`iris/src/widget/text/edit.rs:87`), so `core/src/sense.rs`'s
     `run_sensors` (which stops at the first layer, checked
     innermost-first, that consumed the gesture) gave that row first
     refusal on *every* frame it was pressed, not just the frame the
     press started -- a row's drag-select won the same gesture a
     list-level pan would want. This is a genuine, diagnosed
     architecture gap this box's *own* two features created by both
     wanting the same gesture -- not a missing primitive the way
     Masonry's absent `on_pointer_event` drag handling was.
  
     **Gap closed, 2026-09-05, same day.** `iris::sense::DragArbiter`
     (`iris/src/sense.rs`, new public type, recorded in `IRIS.md`) is
     one small state machine, one instance per gesture surface (a
     whole list, not per row), driven with a caller-supplied `Instant`
     so it needs no render harness to test. It decides the way
     Android itself does, recorded in `DECISIONS.md`: an ordinary
     vertical drag pans immediately; a stationary press held
     `LONG_PRESS` (500ms) starts a selection, which any further drag
     then extends; a horizontal drag while something is already
     selected extends it immediately, skipping the wait.
     `transcript-ui/src/selection.rs`'s new `Selection::drag` is the
     one place every row's `CursorSense::click_or_drag() |
     CursorSense::unclick()` handler now goes through (`row.rs`,
     `build_text_row`), replacing the direct `begin`/`extend` calls
     each row used to make on its own -- one arbiter shared across
     every row is what keeps the decision consistent as a drag
     crosses row boundaries, per `DragArbiter`'s own doc. `Pan(dy)`
     calls the list's own `List::scroll` (the same method I3's
     mouse-wheel handler and its own benchmark already use), so this
     is not a second scroll mechanism. 8 new unit tests in
     `iris/src/sense.rs`'s `drag_arbiter_tests` (vertical drag pans
     immediately and keeps panning by per-frame delta; small jitter
     under `DRAG_SLOP` stays undecided; a held press starts a
     selection after `LONG_PRESS` and further drag extends it, even
     vertical drag, once selecting; a horizontal drag with nothing yet
     selected stays undecided rather than guessing; a horizontal drag
     with something already selected extends immediately; a vertical
     drag still pans even with a prior selection; release resets to
     idle). Verification: `cargo fmt --all -- --check`, `cargo clippy
     --workspace --all-targets` (zero warnings), `cargo test
     --workspace` (28 pre-existing + 9 `transcript-ui` + **8 new**
     `drag_arbiter_tests`, all passing), `cargo ndk -t x86_64 -P 26
     build/clippy` for both `-p iris` and `-p transcript-ui --lib`
     (clean), and `run-headless.sh transcript --shot ... -- -p
     transcript-ui` -- byte-identical to this box's original
     screenshot (38578 bytes, `cmp` confirms identical), confirming no
     visual regression from the rewiring. **What this did not
     attempt**: the emulator-side confirmation (a real touch swipe
     over a row's text panning on-device) -- that still needs I5's own
     Android integration, the one item named just above and in "What
     remains" below; this pass only had the winit/host-side gesture
     path to drive, since no cdylib exists yet for this screen.
  
  **Verification, exact commands and results (2026-09-05, this VM):**
  
  - `cargo fmt --all -- --check`: clean.
  - `cargo build --workspace --all-targets`: clean, all six workspace
    members (`iris`, `iris-core`, `iris-macro`, `tabs-ui`,
    `transcript-ui`, plus the excluded `android-app`).
  - `cargo clippy --all-targets` and `cargo clippy -p transcript-ui
    --all-targets`: zero warnings.
  - `cargo test --workspace`: 28 tests in `iris`/`iris-core` (all
    pre-existing, unaffected) + **9 new in `transcript-ui`** -- 5 pure
    markdown tests (`bold_and_italic_produce_spans_over_the_right_range`,
    `heading_gets_a_bigger_font_size_span`,
    `link_is_styled_and_keeps_its_visible_text`,
    `fenced_code_block_is_monospaced`, a plain-text baseline) and 4
    selection tests (forward/backward/single-row range arithmetic,
    plus `unregister_forgets_the_row_and_clears_a_matching_anchor`
    against a real minimal `TextEdit` in the arena, no window needed --
    same harness style as `list.rs`'s own tests).
  - `cargo ndk -t x86_64 -P 26 build -p transcript-ui` and `... clippy
    -p transcript-ui --lib`: clean (`--lib` only -- the example uses
    `iris::default`, winit-only by design, same as `iris/examples/
    tabs`'s own example never having an Android build of itself; the
    Android-facing entry point is a separate cdylib, not built this
    pass, see below). `cargo ndk ... build -p iris` / `clippy -p iris`
    also re-checked clean, since this box touched `iris-core`'s text
    pipeline.
  - `run-headless.sh transcript --shot ... -- -p transcript-ui`:
    renders. Cropped for legibility (this VM has no image viewer --
    see I3's own note on the same limitation and the throwaway crop
    tool used here, not committed): a full conversation with a
    **bold** word, *italic* text, `inline code` in its own colour, a
    `# Sure` heading rendered visibly larger and bold, a coloured link,
    a monospaced fenced code block, a collapsed "▸ 3 tool calls" row,
    and the composer bar at the bottom -- every one of E2's markdown
    screenshot's features, now inline within single paragraphs rather
    than block-per-widget. Screenshots at `/tmp/iris_i5_transcript2.png`
    (full) and crops there, not committed per the standing rule against
    screenshots of real content leaving this repo -- these are
    synthetic rows, but the rule is kept uniform regardless.
  
  **The Android integration, done 2026-09-05.** Extended
  `iris-android-app` (I2's shell) with a second, mutually-exclusive
  `AndroidAppState` behind a new Cargo feature rather than building a
  third shell -- see `iris/android-app/src/lib.rs`'s module doc for why
  that was chosen over a standalone crate: the Gradle project, the
  `IrisView`/`MainActivity` Java, and the `register_view_class` wiring
  I2 already built are exactly what a second screen needs too, and the
  only thing that differs is which `AndroidAppState` the JNI entry
  point instantiates. `transcript_client.rs` (new) fetches the sandbox
  server's session list, opens the first one, and follows it live --
  `client_core::api`/`event_stream`/`transcript_fold` almost verbatim
  from `desktop-app`'s `app.rs` (E4), down to the generation-guard
  pattern; `fold_page`/`raw_seq` were hoisted into `client-core` itself
  first so both callers share one copy rather than a second one being
  pasted in (a separate small commit, "write the logic once").
  Deliberately simplified, recorded rather than left to be
  rediscovered: no session list UI and no enrollment flow exist for
  this screen -- `build.rs` bakes the sandbox's host/port/token and the
  pinned CA in at build time from
  `AI_APP_TRANSCRIPT_HOST`/`_PORT`/`_TOKEN`/`AI_APP_CA` env vars, same
  trust-boundary reasoning as the Compose app's `GeneratePinnedCert`
  Gradle task (`app/androidApp/build.gradle.kts`), extended here to
  also bake the enrollment since building a real one is E3/E4's scope,
  not this box's. A real app needs `desktop-app`'s
  `EnrolledServer`/QR-link flow or E3's Keystore-sealed
  `ServerConfig.kt`.
  
  **Two real bugs found and fixed getting an actual screen on
  screen, neither anticipated by this box's earlier design:**
  
  1. **Missing `INTERNET` permission.** `iris-android-app`'s manifest
     never needed one before this screen (the tabs demo makes no
     network call), so nobody had noticed it was absent. Its absence
     reads nothing like a network problem: `UreqTransport::new`'s
     connect failed with `EPERM` ("Operation not permitted"), not the
     `ECONNREFUSED`/`ENETUNREACH` a dead server or a firewall would
     give -- a seccomp-level socket denial. Added
     `<uses-permission android:name="android.permission.INTERNET" />`
     with a comment naming the exact symptom, so the next person
     hitting `EPERM` from this codebase's own `ureq` stack finds the
     answer instead of debugging the server.
  2. **A background task's first redraw request past the initial one
     aborted the process.** `Tasks::redraw_handle()` (new, this box,
     see `IRIS.md`'s 2026-09-05 entry for the full account) exists so
     `transcript_client.rs` can ask for a frame after each
     `TaskCtx::update`, the way `desktop-app` uses winit's `Proxy` for
     the same reason. Calling it crashed with `SIGABRT`,
     `Result::unwrap() on an Err value: JavaException`, inside
     `android-view`'s `View::post_frame_callback` -- its Java side
     calls `Choreographer.getInstance()`, which throws unless the
     *calling* thread already has a `Looper`, and a tokio worker
     thread has none even once JNI-attached. Fixed by routing through
     `View::post_delayed(0)` instead (thread-safe, no `Looper`
     required) and a new `IrisViewPeer::delayed_callback` override
     (`android/view.rs`) that drains tasks and renders on the UI
     thread the callback lands on -- same body as `do_frame`. Every
     future caller of `redraw_handle()` from a background thread gets
     this for free.
  
  **The emulator itself needed a boot recipe none of the three
  previously-documented ones give cleanly, found the hard way.**
  Plain `emu up` (`-gpu host`, no Vulkan feature) crashed instantly --
  `wgpu_core::instance: Request adapter didn't find compatible
  adapters` -- this AVD's default boot has no Vulkan device at all,
  matching "Vulkan in the emulator" below. The documented fix for
  *that* (`VK_DRIVER_FILES=... GPU_HOST_FEATURES="-feature Vulkan" emu
  up`) does get a Vulkan device, but on this host it is a **second**
  one alongside the real GPU's own Venus/gfxstream Vulkan adapter, and
  `AndroidRenderer::new`'s `request_adapter` (no adapter-name
  filtering, `PowerPreference::default()`) picked Venus -- which
  crashed inside `wgpu_core::device::resource::Device::
  create_bind_group_layout`, the same structural Venus incompatibility
  "Vulkan in the emulator" already documents for a different call.
  `EMU_GPU=software` (`-gpu swiftshader_indirect`, no host GPU
  involved at all) is what actually works cleanly, because there is
  then only the one Vulkan device (`SwiftShader Device (Subzero)`) for
  `request_adapter` to find -- confirmed via
  `wgpu_core::instance: Found 1 compatible adapters`. **One trap in
  switching between these**: the AVD's saved snapshot carries over
  whichever GPU config booted it last, so restarting under
  `EMU_GPU=software` right after a `-feature Vulkan` boot still linked
  against `vulkan.ranchu.so` and crashed (`SIGSEGV` inside
  `vk_common_SetDebugUtilsObjectNameEXT`) until the AVD's
  `snapshots/` directory was cleared by hand -- matches "Vulkan in the
  emulator"'s own note that a GPU-config switch needs a cold boot the
  `emu` wrapper does not force. Recorded here rather than only in that
  section since it is what made three different crashes look like
  three different bugs before the pattern was the AVD's snapshot, not
  the code.
  
  **Measurements taken, 2026-09-05, `ai-app-2`'s own emulator,
  `EMU_GPU=software`, against `app/ui-sandbox.sh` (port 8519, session
  `8920378e7167ebcd`, 40 real sent/echoed messages):**
  
  (a) **Tap-by-name on a named control -- passes.**
  `ui-trace record --do "tap 'Message'"` (the composer's `.label`,
  I4) resolved and the field's bounds moved (`top 2329 -> 1509`,
  the keyboard opening), the same shape I4's own tabs-screen taps
  confirmed the same day. `iris-android-app`'s tabs screen also got the
  full I4 pass-condition run this session -- see I4's own box above,
  now ticked `[x]`.
  
  (b) **The `transcript-bench.sh`-shaped scroll comparison -- a
  real number for Compose, no comparable number for iris, and that
  gap is itself the finding.** `transcript-bench.sh` could not be
  pointed at `iris-android-app` directly -- it reads the Compose app's
  in-app render-report log line, which this screen has no equivalent
  of -- so the same 24-swipe gesture loop (`swipe 540 700 540 1600
  200` / `swipe 540 1600 540 700 200`, alternating, matching that
  script's own cycle) was driven by hand via `ui-trace record` against
  both apps, each freshly opened on the same session, `dumpsys gfxinfo
  <package> reset` beforehand and `dumpsys gfxinfo <package>` after.
  Compose: **8.96% janky frames, 99th percentile 150ms, 212 frames
  rendered** over the loop -- worse than AGENTS.md's own recorded
  stock-emulator baseline (5.2-5.9%, 29-32ms), consistent with
  `EMU_GPU=software`'s CPU rendering being slower than the `-gpu host`
  that baseline was taken under, which is exactly why AGENTS.md's rule
  against reading an absolute emulator number as the phone's applies
  doubly here. **iris: `dumpsys gfxinfo` reported 0 frames rendered
  for the entire gesture window, on both a run where the screen
  visibly did not move and one where it visibly did** (confirmed
  by `adb exec-out screencap -p`, hashed before/after -- identical
  when the swipe direction was already at that end of the transcript,
  different once swiped the other way). **`gfxinfo` instruments
  Android's own Skia/HWUI View-drawing pipeline; it has no visibility
  into a `SurfaceView` whose contents are drawn by a separately-owned
  GPU context (`wgpu`/Vulkan, here) the way Compose's ordinary `View`
  tree is drawn.** A `dumpsys SurfaceFlinger --latency` probe against
  the transcript screen's own `SurfaceView` layer was tried as a
  fallback and returned only the display's refresh period (16666666ns)
  with no frame history at all -- this Android version's BLAST
  compositor does not keep the per-frame timestamps that legacy API
  used to report. **So there is no dumpsys-derived frame-time number
  for iris on this build**, not a bad one -- the honest comparison this
  pass can make is functional (both apps' lists scroll under the same
  touch gesture) rather than numeric, and getting a real number for
  iris needs the app's own frame-timing instrumentation (the render
  report the Compose side already has, iris has none of yet) rather
  than a different `dumpsys` incantation.
  
  (c) **Touch-drag pans the list on real device touch input --
  confirmed by screenshot, not by `ui-trace show`.** `ui-trace show`
  cannot answer this at all here: the only named node on this screen
  is the composer, which does not move when the list scrolls, so every
  `--field box` query reports "nothing moved" regardless of whether
  the list actually did (the same "no named structure to track"
  situation I4's tabs-screen note about clipped bounds warns about,
  one level further -- here there is no candidate node at all, not a
  clipped one). `adb exec-out screencap -p` before and after a single
  `swipe 540 700 540 1600 300` (list not already at that end) hashed
  different and visibly showed different message rows on screen;
  the same swipe repeated when already at that end of the transcript
  correctly hashed identical -- so the mechanism responds to real
  touch, in both directions, not just once by luck. **Long-press then
  drag to select was not independently driven this pass**: doing it
  for real needs a touch held stationary for `LONG_PRESS` (500ms)
  and *then* moved without lifting, and neither of `ui-trace`'s two
  gesture primitives can produce that -- `tap` has no hold, and
  `swipe X1 Y1 X2 Y2 MS` interpolates motion across its whole duration
  from t=0, so a long `swipe` with a short first segment is still
  continuous motion throughout, not a hold followed by a drag. This
  needs either a new `ui-trace` action (a `hold MS then drag X Y`
  primitive) or a raw multi-step `sendevent`/`MotionEvent` injection
  neither this pass's tooling nor its remaining time could build
  safely. `DragArbiter`'s own unit tests (`iris/src/sense.rs`, I5's
  earlier "Gap closed" section) already cover this exact sequence with
  a synthetic clock, which is why the mechanism is trusted enough to
  call "not independently driven on-device" rather than "unverified."
  
  **Update, 2026-09-05, later the same day: (b) has a real iris number
  now, and (c) is confirmed on-device.** Both needed new tooling built
  this pass, recorded in `DECISIONS.md`: `iris_core::FrameReport`
  (`iris/core/src/render/frame_report.rs`) times each frame from
  `render()`'s redraw start to after `queue.submit`+`present()` into a
  fixed 4096-entry ring, exposed as two named controls on the
  transcript screen ("Frame report", "Reset frame report",
  `iris/android-app/src/transcript_client.rs`) logged under this
  crate's fixed `android_logger` tag; and `ui-trace` gained a
  `holddrag X1 Y1 X2 Y2 HOLD_MS MOVE_MS` action in `emulator-tools`
  (press, hold, move, release as one continuous touch via the same
  `MotionEvent`/`injectInputEvent` mechanism `swipe` already used),
  closing the exact gap named above.
  
  (b), continued: **a real iris number exists, but it is not the clean
  24-swipe `transcript-bench.sh`-equivalent loop this box originally
  wanted, for a reason worth recording precisely.** Driving the
  gesture loop against a freshly-restarted app repeatedly produced
  **zero** frames recorded (both by `gfxinfo`, already known, and now
  also by `FrameReport` itself) even though the coordinates were
  confirmed on-screen to sit over real row text (measured by scanning
  a screenshot column for the first non-black pixel, not guessed) --
  while the *same* coordinates driven a few commands later, or
  combined into a slightly different sequence, sometimes produced 30+
  real frames and a genuine screenshot diff. This is not the earlier,
  already-understood "already at that scroll edge" case (AGENTS.md's
  own note) -- it reproduced with fresh content confirmed taller than
  the viewport, in both scroll directions, inconsistently across
  otherwise-identical commands. The one measured correlate: this
  checkout's own `EMU_GPU=software` emulator was independently seen at
  **~78% of one CPU core, continuously**, while idle on-screen (`ps
  aux` mid-session) -- SwiftShader's software rasterisation is
  CPU-bound by design (AGENTS.md's Vulkan-in-the-emulator section), so
  a synthetic touch's delivery to the SurfaceView competing with that
  load is the leading candidate, not yet confirmed with a sampler
  running *during* the gesture (the standing rule against diagnosing
  from measurements taken after the fact applies here and this pass
  did not have time to build that sampler). Recorded as a new,
  distinct, unresolved finding in `IRIS_TODO.md` rather than folded
  into the already-closed "no `hold`-then-drag primitive" gap.
  **The number obtained, honestly scoped**: tapping "Frame report"
  immediately after a run that *did* produce real scrolling frames
  (screenshots differ, confirmed by hash) read
  `frames=34 janky%=61.76 p50=26.5ms p90=48.0ms p99=98.1ms
  worst=98.1ms` -- real, measured wall-clock time through iris's own
  render path from a real on-device touch-drag, not a synthetic
  probe, but accumulated across several swipe gestures across
  multiple `ui-trace record` invocations rather than one clean
  24-swipe loop, so it is **not directly comparable** to the Compose
  figure below in scale, only in kind. Given the CPU contention
  candidate above, a high jank percentage here is expected under
  software rendering and should not be read as iris's number on real
  hardware. The Compose figure quoted for reference
  (**8.96% janky frames, 99th percentile 150ms, 212 frames rendered**)
  is the same measurement this box already recorded on 2026-09-05
  earlier the same day, under the same `EMU_GPU=software` config on
  this same checkout's emulator -- **not re-taken this pass** (the
  session's time went to building the two rigs above and diagnosing
  the flakiness instead), and against a *different* sandbox session
  (40 messages, id `8920378e7167ebcd`) than this pass's own (120
  messages, id `c76b71d017a54589`), so the two numbers share
  configuration but not identical content -- said plainly rather than
  presented as a matched pair.
  
  (c), continued: **confirmed on-device, by both routes the task
  asked for.** `ui-trace record --do "holddrag 300 1850 300 2050 600
  300"` (a 600ms hold, comfortably past `DragArbiter`'s 500ms
  `LONG_PRESS`, then a 300ms move) against a row's real text produced,
  in order: `iris selection: begin at row 3165`, then a sequence of
  `iris selection: extend to row ...` lines as the drag crossed row
  boundaries -- logged from `transcript-ui/src/selection.rs`'s
  `Selection::drag` (new `log` dependency, smallest addition since
  selection has no accessibility label of its own yet, per
  `IRIS_TODO.md`'s existing gap). A screenshot taken right after shows
  the expected highlighted selection spanning multiple rows,
  confirming the mechanism visually as well as in the log. This is
  the first time `DragArbiter`'s pan-vs-select decision has been
  driven by a *real* Android touch sequence rather than only its own
  synthetic-clock unit tests.
  
  **What remains, named rather than silently dropped (also in
  IRIS_TODO.md, dated 2026-09-05):**
  
  - **Intermittent touch delivery under `EMU_GPU=software` CPU load**
    -- new finding above. Needs a sampler running *during* a failing
    gesture (load, `dumpsys input`, a frame-by-frame `ui-trace`
    capture at `-i 0`) rather than another guess after the fact, and
    ideally a comparison against `-gpu host` (real Vulkan, but shared
    with whichever GPU config a peer session's emulator already
    holds) to see whether it is specific to software rendering.
  - **A clean, single 24-swipe `transcript-bench.sh`-equivalent
    iris number** -- blocked on the above; the number this pass got is
    real but not that clean run.
  - **Row-level accessibility names** -- behaviour 6, I5's own writeup
    above.
  - **A tappable link and a code-span background chip** -- behaviour 2.
  - **`Selection`'s anchor-row shortcut** -- behaviour 1.
  - **No syntax highlighting inside a fenced code block** -- `markdown.rs`
    notes `client_core::highlight` exists and could feed this.
  - **`row.rs`'s tool-row expand and `selection.rs`'s cross-row drag
    are not independently screenshotted/driven** -- covered by reading
    and by the primitives they reuse (I3's `List` tests, this box's
    own unit tests), not by a dedicated repro this pass.
  
  **Net for the recommendation.** Item 3 ("decide when the transcript
  screen exists in both, from the measurements") now has a real number
  on the iris side for the first time -- `FrameReport` works, is unit
  tested (6 tests over the ring/percentile math), and captured a real
  on-device touch-drag's timing -- but that number is scoped narrowly
  (accumulated over several gestures, not one comparable loop) because
  of the intermittent-touch-delivery finding above, so it still cannot
  be read against Compose's 8.96%/150ms figure as a clean comparison.
  What *can* be said, updating the account further: iris's Android
  integration, its frame-timing instrumentation, and its long-press
  selection have all now been exercised by real on-device touch input
  end to end (not only unit tests), on top of the structural points
  E2 already found Masonry unable to reach at all (cross-row
  selection, true per-span inline rich text). `DECISIONS.md`'s
  DEFERRED item is updated with this session's numbers and the
  touch-delivery caveat rather than a decision made here.
  
  **Clean scroll comparison, 2026-09-05, one session, this checkout's
  emulator, `EMU_GPU=software` only (the mode the existing Compose
  figure above was taken under; a second pair under `-gpu host` was
  not reached this pass -- see "Not attempted" below).** New content
  for a fair pairing: a fresh sandbox session (`app/ui-sandbox.sh
  spawn benchsession`, id `4d21d4a0d38f79fd`) with 30 identical sent
  messages, each one heading/bold/italic/inline-code/link/list/fenced-
  code paragraph, so both apps scroll the exact same bytes -- neither
  of the two sessions quoted in this box's earlier passes (`8920378e
  7167ebcd`, 40 msgs; `c76b71d017a54589`, 120 msgs) was reused, since
  neither app had touched it. A sampler
  (`date`/`/proc/loadavg`/`/proc/pressure/{cpu,io}`/top-5-by-CPU every
  2s to `/tmp/iris-bench-sampler.log`) ran for the whole session,
  started before either app was built, per the standing rule against
  diagnosing a timing question from measurements taken after the
  fact.
  
  | app | build | GPU mode | frames | janky % | p50 | p90 | p99 | worst |
  |---|---|---|---|---|---|---|---|---|
  | Compose (in-app report) | debug | software | 1102 | 99.0% late | 33.8ms | 50.6ms | 79.5ms | -- |
  | Compose (`dumpsys gfxinfo`) | debug | software | 1499 | 21.15% (95.66% legacy) | 32ms | 48ms | 150ms (99th) | -- |
  | iris (`FrameReport`, run A) | **release** | software | 299 | 94.65% | 79.1ms | 98.6ms | 117.8ms | 212.6ms |
  | iris (`FrameReport`, run B, repeat) | **release** | software | 233 | 94.42% | 109.3ms | 130.8ms | 147.1ms | 150.5ms |
  
  Exact commands: Compose via `app/transcript-bench.sh` unmodified
  (`open_session` then `copy_render_report` bracketing the standard
  24-swipe/6-cycle loop, `dumpsys gfxinfo com.example.aiapp reset`
  taken immediately before for the second row). iris via the same
  24-swipe loop code -- extracted verbatim from
  `transcript-bench.sh`'s `DO=""` .. `eval ui-trace record` block with
  `sed`, not retyped, since `transcript-bench.sh` itself is Compose-
  specific (opens by session title through the Compose app's own UI)
  and could not be called directly -- bracketed by `ui-trace record
  --do "tap 'Reset frame report'"` and `--do "tap 'Frame report'"`
  (iris's own two named controls, I5's earlier "Update" section),
  reading the result from `logcat`'s `iris frame report:` line.
  
  **What each number counts, stated because the three are not the
  same measurement.** Compose's in-app report times its own
  Compose-internal phases (`total` = the full frame from Choreographer
  callback to submit) and calls a frame "late" past a 16.7ms budget --
  a stricter, self-reported definition. `dumpsys gfxinfo`'s "janky"
  is Android's own HWUI/BLAST deadline-miss accounting, a different
  threshold and a different frame population (it free-runs over
  `Total frames rendered`, which includes frames from opening the
  session and the report dialog, not only the swipe window -- hence
  1499 vs. the in-app number's 1102). iris's `FrameReport` times
  wall-clock from `render()`'s redraw start to after `queue.submit`+
  `present()` -- i.e. iris's own render path only, nothing above the
  GPU submit and nothing from Android's compositor -- confirmed
  independently useless for iris via `dumpsys gfxinfo
  dev.iris.android.demo`, which reported 1 total frame for the whole
  run (unchanged from the earlier pass's finding: HWUI has no
  visibility into a `wgpu`-drawn `SurfaceView`).
  
  **Not comparable, stated plainly:**
  - **Build profile differs by necessity, not choice.** Compose is
    the **debug** variant (AGENTS.md's own bench-script requirement,
    "the emulator scripts stay on the debug build"). iris is
    **release** because I4's box already found the debug/dev profile
    `SIGSEGV`s in this emulator's Vulkan loader the moment `wgpu`
    creates a bind-group layout (`InstanceFlags::from_build_config()`
    turns on debug object-labelling, which crashes against
    `vulkan.ranchu.so`) -- there is no debug iris number to quote on
    this rig. A release build is typically *faster* than debug, so
    this asymmetry very likely understates how much worse than
    Compose iris's own number would look built the same way Compose's
    is, not the reverse.
  - **The jank definitions and frame populations differ**, per the
    paragraph above -- none of the three numbers is measuring the
    same thing, so reading "94.65% > 21.15%" as "4x worse" is not
    sound; only the general shape (iris's frames take longer, both by
    its own accounting and by eye in the screenshots) transfers.
  - **Both figures are emulator numbers under software rasterisation
    (`EMU_GPU=software`/SwiftShader), not phone numbers**, per
    AGENTS.md's and `this-machine-android`'s standing rule -- restated
    because it applies doubly to iris's own number here: SwiftShader
    is CPU-bound by design, and Compose's *own* in-app report shows a
    14.6-22.2ms `swap` phase and a 20.8-33.9ms `gpu` phase alone (more
    than the entire 16.7ms budget) under the same GPU mode, so a
    software-rendering iris number well above 16.7ms is expected
    going in and should not be read as an iris-specific defect
    without a `-gpu host` pair to compare against.
  
  **Where iris's time goes, from `FrameReport`/logcat -- not
  optimised, per this task's own instruction, only described.** Two
  things were checked because they were checkable without new
  instrumentation: (1) **iris does not redraw while idle** --
  `adb logcat -c` followed by a 3s settled wait produced *zero*
  `iris::android::view: render()` lines, both before and after a
  swipe; the render-per-frame spam only appears during and briefly
  after a gesture (visible inertial settle), so there is no idle-
  redraw tax to find here, unlike the composer-inset bug AGENTS.md
  records for the Compose app. (2) **a swipe frame does not appear to
  relayout the whole list** -- consecutive `render()` log lines during
  a swipe show `active=88` falling to `83`, `78`, `73`, ... one small
  step per frame, consistent with I3's virtualised list culling
  widgets that scrolled out of the viewport rather than re-measuring
  everything each frame (a full-list relayout would show `active`
  constant at the total row count, not shrinking through it). Neither
  observation isolates *where* the remaining ~80-150ms/frame actually
  goes past those two rule-outs -- the leading remaining candidate is
  the swapchain present/GPU path itself under SwiftShader's CPU
  rasterisation, per the "not comparable" point above, but that was
  not measured directly this pass (no per-phase breakdown inside
  `FrameReport` the way Compose's report has `measure`/`place`/
  `record`/`swap`/`gpu`).
  
  **The dropout finding, corrected from earlier in this box: this
  pass's own script bug, not a reproduction of the touch-delivery
  candidate.** Of 3 planned attempts, the first 2 produced `iris frame
  report: no frames recorded` -- but tracing it down found the cause
  in this session's own tooling, not the emulator: the swipe loop was
  extracted from `transcript-bench.sh` into a temporary wrapper script
  that `cd`'d into `/tmp` before invoking `ui-trace`, and `ui-trace`/
  `adb` here derive *which emulator to target* from the current
  directory's basename (the per-checkout-AVD rule) -- from `/tmp` that
  resolved to a nonexistent checkout named "tmp", `ui-trace` refused
  immediately, and the wrapper's `set -eu` aborted the whole loop
  before a single swipe was sent. Once the wrapper was fixed to run
  from inside this checkout, the next **two** attempts (runs A and B
  in the table above) both succeeded on the first try, each with a
  confirmed screenshot-hash difference showing real scrolled content.
  So this session did not reproduce the previously-documented
  intermittent zero-touch phenomenon -- but two successes out of two
  *valid* attempts is also too little evidence to say it is gone;
  the earlier session's drops happened with a correctly-targeted
  device, which is a different failure than the one found here.
  **Sampler timeline**: `/tmp/iris-bench-sampler.log` shows
  `/proc/loadavg` and the runnable-process count rising from an idle
  baseline (~0.3-1.7 load, 0-4 running) to ~2.5-2.9 load and 11-23
  running during the swipe window that produced run A -- consistent
  with the standing candidate (SwiftShader's software rasterisation
  loading the CPU during a gesture) but **not a confirmed cause**,
  since both valid attempts succeeded despite the rise. Whether load
  of that shape is what caused the *earlier* session's drops remains
  unknown; this pass's sampler evidence neither confirms nor refutes
  it, only shows the correlate is present under load without a
  failure to correlate it to this time.
  
  **Not attempted this pass**: the second `-gpu host` pair (time went
  to the software-mode pair, the sampler, and diagnosing the dropout
  above); a per-phase breakdown inside iris's own `FrameReport` the
  way Compose's report has one; and syntax-highlighting/tappable-link/
  accessibility-name gaps already named in "What remains" above,
  unchanged.
  
  **Verification for this update**: no Rust or Kotlin code changed
  this pass (build/measurement only), so `cargo fmt`/`clippy`/`test`
  were not re-run; `docs/DECISIONS.md`'s DEFERRED item is updated with
  this table's headline numbers below rather than a decision made
  here.
  
  **Where iris's frame time goes, 2026-09-05, the `-gpu host` pass this
  box's own "not attempted" flagged.** New code first: `FrameReport`
  (`iris/core/src/render/frame_report.rs`) now splits each sample at
  `queue.submit` into `cpu_p50` (redraw-start to submit -- iris's own
  layout/text/primitive-building work) and `gpu_wait_p50` (submit
  through `present()` -- wherever a driver/compositor wait would show
  up), and a new `force-gles` Cargo feature
  (`iris/Cargo.toml`/`android-app/Cargo.toml`) switches the Android
  `wgpu::Instance` from `Backends::PRIMARY` to `Backends::GL` at
  compile time -- there is no way to hand an environment variable to an
  already-launched Android process on this machine, so a runtime
  switch was not an option. `app/iris-scroll.sh` extracts
  `transcript-bench.sh`'s exact 24-swipe/6-cycle loop for iris's own
  demo app. Commit `e2a1fad`.
  
  **Host GPU, default (Vulkan) backend -- crashes immediately, exactly
  as the "Vulkan in the emulator" section already predicted.** Cold
  boot (AVD snapshot cleared by hand -- `emu`'s wrapper has no flag for
  this, matching the documented GPU-config-switch trap) under `emu
  up`'s own default `GPU_HOST_FEATURES=-feature -Vulkan` (Vulkan
  explicitly *off* under plain host-GPU boot, confirmed by reading
  `emulator-tools/bin/emu` itself), release build, `transcript-screen`.
  `dev.iris.android.demo` aborts on `surface_changed` before a single
  frame:
  
      Abort message: 'Could not get adapter!: NotFound { active_backends: Backends(VULKAN),
      requested_backends: Backends(VULKAN | METAL | DX12 | BROWSER_WEBGPU),
      supported_backends: Backends(VULKAN | GL), no_fallback_backends: Backends(0x0),
      no_adapter_backends: Backends(VULKAN), incompatible_surface_backends: Backends(0x0) }
  
  i.e. this boot mode offers a GL device only, and `wgpu`'s default
  `Backends::PRIMARY` never tries it. Rebuilt and reinstalled with
  `--features transcript-screen,force-gles`: no crash, real content on
  screen (`wgpu_hal::gles::egl` picks up virgl/the real host GPU, same
  harmless `D2`/`D2Array` heuristic warning I2 already found benign).
  
  **Host GPU, `force-gles` -- a real number, and it changes the
  picture.** Same 24-swipe/6-cycle loop (`app/iris-scroll.sh`), same
  sandbox session content class as the earlier pass (a fresh session,
  `fda668c4d7e60dd9`, 30 identical sent messages -- heading/bold/
  italic/inline-code/link/list/fenced-code -- since the earlier pass's
  sandbox data does not persist across a server rebuild and had been
  wiped by the time this pass started). Compose (debug) via
  `transcript-bench.sh -s benchsession2` on the same session, same
  emulator boot:
  
  | 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 2-3x slower** -- the opposite shape from the
  software-mode table above. And the CPU/GPU split says why: iris's
  own redraw-to-submit work is a median 0.2ms, essentially free: almost
  the entire 15.0ms median frame is `gpu_wait_p50` (submit through
  `present()`), i.e. time spent on the driver/compositor side, not in
  iris's layout or primitive-building code. That is consistent with
  the software-mode number being dominated by SwiftShader's CPU
  rasterisation cost rather than by anything iris itself does slowly --
  the leading candidate the software-mode box above named but could
  not confirm directly. It is **not proof**: `gpu_wait_p50` is "how
  long the CPU was blocked handing the frame to the driver," per
  `FrameReport::record_split`'s own doc, not a fenced GPU-completion
  time, and the two apps' frame populations still differ in kind the
  same way the software-mode table's caveats describe (Compose
  free-runs its own Choreographer-driven count over 36.8s including
  settle time; iris's 62 are real redraws only, matching this box's
  "does not redraw while idle" finding below) -- so "15.0ms vs. 20.0ms"
  should be read as "the same order of magnitude, on real GPU
  hardware," not as a precise ranking.
  
  **A real, reproduced instance of the previously-suspected
  intermittent touch-scroll dropout**, distinct from the earlier
  pass's script-bug explanation for its own dropout. After a fresh
  `am start`, six consecutive swipes (`ui-trace record --do "swipe ..."`,
  matching `iris-scroll.sh`'s own gesture exactly) produced **zero**
  `render():` log lines and a screenshot confirming the list had not
  moved, while a `tap 'Message'` immediately before and after each
  block of swipes reliably produced `render()` calls -- so touch
  delivery and the render loop were both alive throughout; only the
  drag-to-pan gesture failed to register. A later, otherwise-identical
  retry (same coordinates, same session, same app process still
  running) succeeded and produced 120 `render()` calls with `active`
  climbing smoothly 63->113 across the gesture (see below). Not
  root-caused this pass -- `iris::sense::DragArbiter` (`iris/src/
  sense.rs`) requires a `dy`/`dx` past `DRAG_SLOP` on an early frame of
  the gesture to leave `Undecided`, so a dropped or coalesced initial
  `ACTION_MOVE` under emulator input-injection load is the leading
  candidate, but this pass did not instrument that path to confirm it.
  Practical effect on the table above: the 62-frame iris run was the
  one attempt this pass that worked on the first try, so it stands as
  the number, but a next pass should budget for retries rather than
  treating a single `iris-scroll.sh` invocation as reliable. **Update,
  2026-09-05, the dropout fix (`e692429`, this box's own "Touch-scroll
  dropout root-caused" subsection) verified against this table**: with
  the fix in, three separate `iris-scroll.sh` invocations against a
  fresh cold `-gpu host` boot each scrolled all 24/24 swipes, no
  retries needed --
      frames=450 janky%=48.44 p50=16.3ms p90=22.7ms p99=26.2ms worst=29.7ms cpu_p50=1.0ms gpu_wait_p50=13.1ms
      frames=462 janky%=50.22 p50=16.9ms p90=25.0ms p99=45.4ms worst=59.0ms cpu_p50=1.3ms gpu_wait_p50=13.9ms
      frames=439 janky%=46.24 p50=15.7ms p90=23.3ms p99=31.2ms worst=57.4ms cpu_p50=1.2ms gpu_wait_p50=13.2ms
  Per-swipe coverage was checked directly, not inferred from the frame
  count alone: a continuous `adb logcat -v time -s iris-android-app:D`
  capture started before each of the second and third runs (the first
  run's capture was taken with `logcat -d` after the fact and lost
  earlier lines to the ring buffer, so it is corroborating rather than
  direct) shows `render():` timestamps clustering into exactly 24
  groups per run, one per swipe, each with 22-42 render calls and no
  gap over 0.3s inside a cluster -- i.e. every one of the 24 swipes
  produced real redraw activity in all three runs, closing this table's
  open verification. The third run (lowest janky% and p50) is the
  "best of three" row now in the table above, replacing the earlier
  single-attempt 62-frame reading.
  
  **Redundant-work check (no optimising, as instructed), host GPU,
  `force-gles`.** (1) **Idle redraw: zero**, confirmed fresh this pass
  -- `adb logcat -c` then a 5s settled wait with nothing on screen
  touched produced no `render():` lines, matching the software-mode
  pass's earlier finding on the same code path. (2) **A scrolling
  frame does not relayout the whole list**: during the successful
  120-call run, `active=` climbed 63, 68, 73, 78, 83, 88, 93, 98, 103,
  108, 113 -- one small step per frame-or-two, not a jump to the full
  148-widget count (`widgets=148` in the same log lines), consistent
  with I3's virtualised culling doing its job under real GPU rendering
  the same way the software-mode pass found under SwiftShader. Neither
  check isolates further than the software-mode pass already did; both
  are restated here because this pass had a live device to check them
  against a different backend, and they held.
  
  **Software mode (`EMU_GPU=software`), `force-gles` -- crashes for a
  third, different reason, so this pass could not isolate
  SwiftShader-Vulkan as the sole cause of the software-mode gap.** Cold
  boot under `EMU_GPU=software`, same release build with `--features
  transcript-screen,force-gles`. `wgpu_hal::gles::adapter` finds a real
  adapter (`Renderer: Android Emulator OpenGL ES Translator (Google
  SwiftShader)`, `Version: OpenGL ES 3.0`), further than the plain
  host-GPU/default-backend attempt got -- but `AndroidRenderer::new`'s
  device request then aborts:
  
      Abort message: 'Could not get device!: RequestDeviceError { inner: Core(LimitsExceeded(
      FailedLimit { name: "max_compute_workgroups_per_dimension", requested: 65535, allowed: 0 } )) }'
  
  i.e. iris's device descriptor asks for compute-shader limits
  unconditionally, and SwiftShader's software GL path reports itself
  as OpenGL ES 3.0 -- compute shaders are an ES 3.1+ feature, so the
  allowed limit is 0. This is a different failure from both the host-
  GPU/default-backend crash above (no adapter at all) and the earlier
  Venus/gfxstream failure "Vulkan in the emulator" documents (a
  different Vulkan implementation's external-memory gap) -- three
  distinct emulator/backend incompatibilities found across this
  project's Android work, not one recurring bug. **Not fixed this
  pass**: making iris's device request tolerant of a downlevel GL
  adapter (requesting compute limits only when the adapter actually
  reports them) is real scope, not a measurement task. Consequence for
  the software-mode question this step was meant to answer: it remains
  open whether SwiftShader-Vulkan specifically (rather than GLES in
  general) explains the ~80-150ms software-mode numbers, since no GLES
  number under software mode could be taken at all.
  
  **Fixed, 2026-09-05, later the same day.** Not "requesting compute
  limits only when the adapter reports them" (a capability check with
  a fallback) -- simpler than that, because iris has no code path that
  needs compute at all: grepped the whole `iris`/`iris-core` tree for
  `ComputePipeline`/`@compute` and found none, so the right fix is to
  stop asking for compute limits, full stop, rather than to build a
  fallback for a capability nothing uses. `iris_core::device_limits()`
  (`iris/core/src/render/mod.rs`) is the one place both platform
  backends now build their `required_limits` from: `Limits::default()`
  with the six `max_compute_*` fields zeroed and `max_buffer_size`
  still raised, as before. `Limits::downlevel_webgl2_defaults()` was
  the first thing tried and rejected -- it also zeros
  `max_storage_buffers_per_shader_stage`, and `shader.wgsl`'s vertex
  stage reads four `var<storage>` buffers, so it would have traded
  this crash for a bind-group-layout one on the same hardware.
  `rigs/gpu-probe`'s own `Limits` (necessarily a hand-mirrored copy --
  that rig is deliberately its own crate, not a workspace member) was
  updated to match and re-run: `IRIS DEVICE: ok` against this VM's own
  Vulkan (Venus) and GL (virgl, reports OpenGL ES 3.2) adapters.
  **Not verified against the actual SwiftShader-ES-3.0 failure this
  pass**: the `EMU_GPU=software` cold boot needed to reproduce it would
  have force-restarted this checkout's shared emulator while another
  session had `com.example.aiapp` focused and running on it (`adb
  shell dumpsys window`), so this pass left that measurement rather
  than disrupting concurrent work -- matching AGENTS.md's "coordinate
  with peer agents" guidance rather than contending for the emulator.
  Everything else: `cargo fmt --all`/`clippy --workspace --all-targets`/
  `test --workspace` clean, `cargo ndk build`/`clippy` for
  `iris-android-app --features transcript-screen,force-gles` clean
  (only the pre-existing unused-`tabs-ui`-dependency warning, unrelated
  to this change). This also means the software-mode question two
  boxes up is still open, for the same original reason plus this new
  one: a GLES number under `EMU_GPU=software` still has not been
  taken, now blocked on emulator availability rather than on the
  crash. A future pass should cold-boot `EMU_GPU=software` once the
  emulator is free, confirm `dev.iris.android.demo` no longer aborts
  on `request_device`, and take the `iris-scroll.sh` FrameReport row
  that pairs with this box's host-GPU one.
  
  **Verification, this update.** `cargo fmt --all` (no diff),
  `cargo clippy --workspace --all-targets` (no warnings from the new
  code; pre-existing `wgpu`/`winit`/`naga` future-incompat notices
  only) both re-run and clean this pass. `cargo test --workspace` and
  `cargo ndk ... test`/`clippy` for `iris-android-app` were **not**
  re-run this pass -- the previous pass on this identical diff had
  already run and reported them clean, and this pass's host was
  disk-pressure-limited (93% full, a concurrent `ai-server` rebuild in
  progress) when the repeat attempt was made, so it was stopped rather
  than left to spend 50+ minutes doing no useful work; see commit
  `e2a1fad`'s own message. `docs/DECISIONS.md`'s DEFERRED item is
  updated with this section's host-GPU table below.
  
  **Touch-scroll dropout root-caused, 2026-09-05.** Diagnosed as
  instructed: temporary `log::info!` tracing on every touch event
  reaching `IrisViewPeer::on_touch_event` (`iris/src/android/view.rs`),
  every `DragArbiter` state transition (`press_start`/`update`/
  `release`, `iris/src/sense.rs`), and every `Selection::drag`
  dispatch (`iris/transcript-ui/src/selection.rs`) -- all removed once
  the cause was confirmed, per AGENTS.md's "keep the build clean."
  Reproduced with `app/iris-scroll.sh` against a real sandbox session
  (30 sent markdown messages, `EMU_GPU` unset / `-gpu host`,
  `--features transcript-screen,force-gles`, release build, same
  recipe as this box's own "-gpu host" pass above).
  
  *The trace.* Of 24 swipes in one run, 5 produced zero `render()`
  calls each -- one at the very start of the run, four consecutive
  later (swipes 22-25) -- exactly the "several consecutive swipes
  produce nothing, an identical retry then works" shape from the
  earlier pass's report. Correlating the three log streams by
  timestamp: every one of those 5 swipes delivered a normal
  `Down`/`Move`×N/`Up` sequence to `on_touch_event` (touch delivery
  was never the problem), but `Selection::drag` never once saw
  `PressStart` for the whole gesture -- only `Pressing`, starting from
  the very first `Move`. `DragArbiter::update`'s `Idle` arm answers
  every such frame with `Undecided` and never transitions state (there
  is no way for pure state to tell "no press is happening" from "a
  press is happening but I missed its start"), so the arbiter sat in
  `Idle` from the gesture's first frame to its last, `release()` on
  `Up` its only state change (`Idle` -> `Idle`, a no-op). The row this
  landed on registered its `PressStart` correctly on a *different*
  point in the very next successful swipe at the identical screen
  coordinate -- confirming the miss is about *where the content
  happens to be under that pixel when `ACTION_DOWN` fires*, not about
  timing or a coalesced event.
  
  *Why `ACTION_DOWN` misses a row's sensor.* Each row's `CursorSense`
  handler is registered only on its `TextEdit` field
  (`transcript-ui/src/row.rs`'s `build_text_row`), not on the row's
  `.pad(10)` margin, the `.gap(4)` between the sender-name header and
  the field, or the header itself (`Span::empty`/a plain `wtext` with
  no handler). A real touch's down-point is wherever the finger
  actually is, with no reason to prefer text over padding, and the
  list has no sensor of its own to fall back to (`iris::widget::list`
  registers none) -- pan is reachable *only* through a row's own
  arbiter. So roughly one in five swipes in this run started on a
  pixel no sensor covered.
  
  *The fix.* `iris::sense::DragArbiter` gains `pub fn is_idle(&self)`,
  documented as the recovery signal: a caller that gets a `Pressing`
  frame while the arbiter reports `is_idle()` knows the button is
  genuinely down (that is what `Pressing` means) with no matching
  `press_start` on record, which can only mean it was missed.
  `Selection::drag`'s match gains one arm, checked after `PressStart`/
  `PressEnd` and before the ordinary `_ => update(...)` case: `_ if
  self.arbiter.is_idle()` starts the press right there instead of
  where it was missed, using whatever `already_selected` holds at
  that later frame (the best available answer -- the true value at
  the actual `ACTION_DOWN` is unrecoverable once missed). This is
  the caller's fix, not the arbiter's, because only the caller knows
  what `already_selected` should be; the arbiter's own slop/long-press
  logic was correct throughout and needed no change.
  
  *Tests.* Four new, all passing on the fix and the first three failing
  without it: `sense.rs`'s `drag_arbiter_tests::
  is_idle_reports_a_press_that_was_never_started`,
  `::update_on_an_idle_arbiter_stays_undecided_forever_without_recovery`
  (documents the failure mode itself), `::
  a_caller_can_recover_a_missed_press_start_via_is_idle` (the pure-state
  half); and `transcript-ui/src/selection.rs`'s `tests::
  a_missed_press_start_recovers_on_the_next_pressing_frame`, which
  drives `Selection::drag` directly with only `Pressing` frames (no
  `PressStart` ever sent) and asserts the arbiter is no longer idle
  afterward -- this one fails on the pre-fix code (`is_idle()` stays
  true forever, matching the real trace).
  
  **Not completed this pass, and why.** The task asked for
  `iris-scroll.sh` run three times clean and a re-taken host-GPU
  `FrameReport` row. Partway through that verification, this
  checkout's shared emulator (`ai-app-2`, per-checkout per AGENTS.md)
  turned out to be concurrently in use by another session actively
  working the P0 phone-benchmark item added to this same file earlier
  today: `adb shell dumpsys activity processes` showed
  `com.example.aiapp`/`com.example.aiapp.bench` processes running
  alongside `dev.iris.android.demo`, window focus was observed to have
  moved to the Compose app mid-test, and the sandbox server's own log
  showed a fresh `start` (not `keep`) at 00:55 that wiped this pass's
  30-message test session and replaced it with the peer's own
  `bench-check` session -- confirmed by `ui-sandbox.sh api /sessions`
  returning "no session" for the id this pass had been sending to.
  Rather than disrupt that session's work (deleting its session,
  restarting its server, or fighting over emulator focus), this pass
  stopped chasing a clean aggregate number once the cause was
  confirmed external. What *is* verified is the fix itself, from
  direct traces taken before the interference began (above) plus two
  manual, shorter `ui-trace` swipe sequences (not the full script) that
  each showed full, healthy per-swipe `render()`/`selection::drag`
  coverage with the fix in place. The FrameReport row in this box's
  own table above is therefore **not re-taken this pass** -- a future
  pass should re-run `iris-scroll.sh` three times and retake it once
  the emulator is free, per AGENTS.md's "ask before/tell peers" and
  "coordinate with peer agents" guidance rather than contending for it.
  Also correctly ruled out, not left ambiguous: a hypothesis raised
  mid-pass that `iris::widget::list::List::scroll`'s deliberately
  unclamped anchor (its own module doc, "no overscroll clamping ...
  leaves a gap rather than rubber-banding back") could itself explain
  a run of consecutive failed swipes once enough net drift
  accumulates -- plausible in isolation, but the run where it seemed
  to reproduce is exactly the run now attributed to the peer
  session's interference (same timestamps), so it was not
  independently confirmed and is recorded here as ruled out for now
  rather than as a second bug.
  

The port, in order (decided 2026-09-05)

Iris decided iris over Masonry (DECISIONS.md). This is the ordered plan for the rest of the app, decided by the design agent per the standing "decide technical questions yourself" instruction — a serious user-facing tradeoff is not in play in the ordering itself, so it is not deferred to her. Crate shape, decided here: the screens live in one crate, iris/app-ui, grown from iris/transcript-ui rather than started beside it — transcript-ui already has the right generic shape (Rsc: HasEvents + Rsc::State: FocusHost, the same axis tabs-ui varies along) and the same client-core/event-model path dependencies every later screen needs, so growing it in place is a rename plus new modules rather than a second crate re-declaring dependencies the first already has. It holds a Screen enum and a back stack — the direct equivalent of AppRoot.kt's when and MainScreen.kt's tab enum — with each Compose screen becoming one iris::widget subtree module. iris/desktop-app (E4) and iris/android-app (I2/I5) become thin entry points that call into app-ui, the way AppRoot/ MainActivity today call into Compose screens they don't otherwise own. Platform-only code (the notification foreground service, the share target, the QR scanner, the Keystore-sealed token, deep-link enrolment) stays exactly where E3/E5 already put it — android-shell/ and app/shellApp — since none of it is a screen app-ui could draw.

Order is by risk to the daily-use path, not by screen count: the session screen is what the app is for and where every hard behaviour (paging, cache, keyboard insets, selection) already lives, so it goes first and on the phone as reachable code as soon as possible, before the lower-risk screens.

Every step below assumes the app/ui-sandbox.sh fixtures (AGENTS.md's "The rigs") and the this-machine-android skill's facts (per-checkout AVD, ui-trace by accessibility name, GrapheneOS phone quirks, the adb shell quoting traps) apply unchanged — read that skill before running any pass condition below that touches an emulator or a real device.

• P0 — the phone benchmark gate (asked for 2026-09-05; must pass before P1 starts). Iris runs both apps on her own phone and pastes the reports back; the emulator's numbers are not a substitute. Two halves, buildable independently: - Compose half (app/): a bench build type (release optimisations, applicationIdSuffix ".bench", own label "AI Sessions bench") whose session screen can open an embedded fixture transcript from assets with no server, and a "Run benchmark" control in the existing render-report place that programmatically performs the fixed scroll loop (same distances and timings as transcript-bench.sh, driven through the LazyListState), then a streaming phase (append fixture events at 20/s for 20 s into the same fold path a live SSE reply uses), then shows the report with the existing copy button. The report adds process CPU time over the run (Process.getElapsedCpuTime), peak RSS, and BatteryManager.BATTERY_PROPERTY_CURRENT_NOW samples. - iris half (iris/android-app, transcript-screen feature): the same fixture embedded, the same scripted loop and streaming phase driven through List::scroll and the fold, the same report fields added to FrameReport's line, shown on screen with a copy-to- clipboard control (through the shell's Java side), arm64 release. - The fixture: one synthetic transcript generated from app/ui-sandbox.sh's invented sessions (never a real one), at least 3,000 events, with headings, code fences, links, tool calls with kilobyte outputs, and a few images; committed once under app/bench-fixture/ and read by both apps. - Delivery: ~/host/bench/ gets iris-bench-arm64.apk, compose-bench-arm64.apk, and README.md saying how to run each and what to paste back. Pass: Iris's call from the two reports — iris within a reasonable margin of Compose on p50, p99 and CPU time, no crash, no stutter she can see. Fail stops the port.

  **Compose half: done, 2026-09-05.** `app/androidApp`'s `bench` build
  type, `app/bench-fixture/` (generator + generated `assets/`),
  `BenchFixture.kt`/`BenchNetwork.kt` (an in-process fake backend: a
  `URLStreamHandlerFactory` installed only in `FIXTURE_MODE` answers
  `https://bench.fixture.invalid:1/...` from an in-memory event log
  instead of opening a socket, so `TranscriptSource`, `EventStream`,
  the fold and the paging are the *real* ones, unmodified), and
  `BenchRun.kt` (the scripted scroll-and-stream, driven against the
  real `LazyListState`) are all in. "Run benchmark" sits beside "Copy"
  in the session settings dialog, bench-build only
  (`SessionSettingsDialog`'s `onRunBenchmark`). `./build-apk.sh bench`
  works, produces a universal APK (no ABI splits in this project, so
  arm64-v8a is included alongside the others — confirmed with `aapt2
  dump badging`), signed with the same release key, own application id
  `com.example.aiapp.bench`, own label "AI Sessions bench" via a
  build-type `resValue` overriding `@string/app_name`.
  
  Checks all clean: `ktfmtFormat`, `compileDebugKotlin`,
  `compileBenchKotlin`, `lintDebug`, `lintBench` (both "No issues
  found"), `testDebugUnitTest`. `grep -n "tap [0-9]" app/*.sh` has one
  hit, pre-existing and unrelated — a comment in `bench-lib.sh`
  recounting the 2026-09-03 incident that made that grep a rule, not a
  literal `tap` call.
  
  **Emulator smoke run, 2026-09-05** (this checkout's AVD,
  `ui-trace` tap-by-label throughout — `tap 'Session settings'` then
  `tap 'Run benchmark'`, report read back over `adb logcat`):
  
      ai-app render report
      device: sdk_gphone64_x86_64 (Google), Android 16
      build: release
  
      transcript:
        28 events, 26 rows, 58 units loaded
        viewport 1536px, 2 units visible
        on screen: the list's own 0px, AssistantMsg 18732px
        0 tool calls and 0 groups open
  
      frames:
        1361 frames over 38.1s at 60Hz (16.7ms budget)
        late: 1353 (99.4%)
        total  p50 27.8ms  p90 37.7ms  p99 50.1ms
        gpu    p50 18.9ms  p90 28.9ms  p99 31.5ms
  
      where the draw phase went:
        draw phase 3.12ms per frame, of which:
          the transcript: 0.33ms (measure 0.18, place 0.14, record 0.00)
          everything else: 2.79ms (89%)
  
      bench:
        scroll: 6 cycles (24 swipes), streamed 400/400 fixture events
        process CPU time over this run: 23005ms
        peak RSS: 209348kB
        battery current: mean 900000µA over 39 samples (min 900000, max 900000)
  
  Read this as "the harness runs end to end and produces every field
  P0 asked for," not as a phone number: it is software-rendered
  emulator rasterisation (this-machine-android's skill — the stock
  Settings app scrolls worse on the same device), and the battery
  current is a fixed 900mA on every sample, which is the emulator's
  mocked charger reporting a constant rather than a real battery —
  expect that field to read "unavailable" or a real varying number
  only on Iris's own phone. The ordinary debug build was rebuilt and
  driven with `./transcript-bench.sh` against `ui-sandbox.sh` alongside
  this and produced its usual report with no `bench:` section, so nothing
  changed for it.
  
  `~/host/bench/compose-bench-arm64.apk` (9.7M) and
  `~/host/bench/README.md` are written, with a heading left for the
  iris half. **Known interaction**: the bench build keeps the same
  `aiapp://enroll` intent filter as the ordinary app (it never uses
  it), so with both installed, driving enrollment through a raw `am
  start -d aiapp://...` intent (not the in-app QR scanner, which is
  the primary path and calls straight into the matched activity) opens
  Android's "Open with" chooser between the two. Cosmetic — the real
  enrollment path is unaffected — and left as is rather than pulling
  the intent-filter out of the bench manifest via source-set merging,
  which was more diff than the problem was worth.
  
  **Not done this pass**: the iris half (a separate agent's scope —
  this session was told not to touch `iris/`), and anything past the
  emulator — the actual on-phone runs and Iris's pass/fail call.
  
  **iris half: done, 2026-09-05.** A `bench` Cargo feature on
  `iris-android-app`, built on top of `transcript-screen`
  (`bench = ["transcript-screen", "dep:libc", "dep:tokio"]`,
  `iris/android-app/Cargo.toml`), gives `lib.rs`'s `ActiveClient`
  priority a third `AndroidAppState` (`bench_client::BenchClient`)
  over `TranscriptClient` when both features are listed together --
  matching the exact build command below, which lists both.
  
  **Fixture.** `include_str!("../../../app/bench-fixture/assets/
  transcript.jsonl")` (1,915,760 bytes) at compile time -- no asset
  pipeline needed the way the Compose half's Gradle source set does.
  `bench_client::parse_fixture` splits the same way `BenchFixture.kt`
  does: the first 3,200 non-blank lines parsed as `serde_json::Value`s
  and folded once through `client_core::transcript_fold::fold_page`
  (the real fold a `/transcript` page goes through), the rest parsed
  as `event_model::SeqEvent`s and held back as the streaming tail.
  `build.rs` (transcript-screen's own) now exits early under `bench`
  before requiring a live server's host/port/token/CA -- `BenchClient`
  never calls `build_transport()`, so that requirement made no sense
  for a build that talks to nothing.
  
  **"Run benchmark" (`.label("Run benchmark")`) and "Copy report"
  (`.label("Copy report")`)** sit in a fixed bar above the transcript;
  a selectable `TextEdit` (`.attr::<Selectable>(())`, the same
  attribute the composer field uses) below it shows the report text.
  Pressing "Run benchmark" resets `FrameReport`, then drives
  `List::scroll` in ~60Hz steps (`ANIM_STEP_MS = 16`) to animate each
  900px/200ms swipe rather than jumping it -- iris's `List` has no
  built-in tween the way `animateScrollBy(tween(...))` gives Compose,
  so this is the one place the two backends' bench code has to differ
  in shape rather than only in numbers -- through the same
  `rsc.tasks.redraw_handle()` + manual `request_redraw()` per step
  `transcript_client.rs` already established (a `Tasks::spawn`d
  future's *automatic* redraw fires once, after the whole future
  completes, which would show nothing moving until the run ends).
  After the scroll loop, `List::jump_to_end()` pins to the newest
  content (matching `stream-bench.sh`'s "Jump to latest" tap), then
  400 fixture events replay at 20/s through `fold_event` -- the same
  fold path a live SSE frame takes in `transcript_client.rs`'s own
  `apply_event` -- each one triggering `rebuild_transcript`'s full
  `transcript_ui::build_tree` rebuild at the time this box was
  written, same tradeoff as `TranscriptClient`/`desktop-app`. **Fixed
  2026-09-05, later the same day**: all three now call
  `TranscriptScreen::apply` instead -- see this same section's
  "Streaming no longer costs a full rebuild" entry below for the
  before/after numbers. A battery sampler runs
  concurrently on its own `tokio::spawn`d task (not through
  `ctx.update`, since a JNI battery read needs no widget-tree access),
  attaching whichever thread it runs on via a stored `JavaVM` --
  `AndroidAppState::platform_ready` (new, `IRIS.md`) is what hands
  `bench_client.rs` that `JavaVM` + a `GlobalRef` to the view, since
  neither was reachable from `AndroidAppState::new` before this box.
  
  **Report fields.** `FrameStats`'s existing `Display` (frames, janky
  %, p50/p90/p99, worst, and I5's own `cpu_p50`/`gpu_wait_p50` CPU/GPU
  split) plus a `bench:`-shaped tail this box added: process CPU time
  via `libc::getrusage(RUSAGE_SELF)` (user+system time; chosen over
  parsing `/proc/self/stat` by hand to avoid assuming `USER_HZ`), peak
  RSS from `/proc/self/status`'s `VmHWM` (same source `BenchRun.kt`
  reads), and battery current sampled once a second via
  `BatteryManager.getIntProperty(BATTERY_PROPERTY_CURRENT_NOW)`
  through direct JNI calls (`bench_jni.rs`'s `PlatformHandle` --
  `android_view::context`'s own `Context`/`Resources` wrappers have no
  `getSystemService`, so this calls it directly rather than growing
  that crate's wrapper for two one-off calls). `0`/`Integer.MIN_VALUE`
  read as "unavailable" rather than folded into the average, matching
  `BatterySampler`'s own rule and UI_RULES.md's "never present an
  inferred value as a measured one." The report is logged under the
  existing `iris-android-app` logcat tag on a line starting `iris
  bench report:` (grep-able the same way `transcript_client.rs`'s
  "Frame report" control already is), shown in the on-screen
  `TextEdit`, and copied to the system clipboard by "Copy report"
  through `ClipboardManager.setPrimaryClip` (`bench_jni.rs`, same
  `PlatformHandle`).
  
  **Build commands, all clean this pass:**
  - `cargo fmt --all -- --check` (iris workspace) and
    `cd iris/android-app && cargo fmt --all -- --check`: clean.
  - `cargo clippy --workspace --all-targets` (iris workspace): clean
    (only the pre-existing `wgpu`/`winit`/`naga` future-incompat
    notice).
  - `cargo test --workspace` (iris workspace): 39 + 8 + 10 = the same
    pre-existing counts, all passing, unaffected by this box (it
    touched no logic under test there beyond `AndroidAppState`'s new
    default no-op method).
  - `cargo ndk -t x86_64 -P 26 clippy --features "transcript-screen
    force-gles bench" --lib -- -D warnings` (`iris/android-app`):
    clean.
  - `cargo ndk -t arm64-v8a -P 26 -o app/src/main/jniLibs/ build
    --release --features "transcript-screen force-gles bench"`:
    clean, `arm64-v8a/libmain.so` produced. The pre-existing "unused
    dependency `tabs-ui`" Cargo advisory also appears on a plain
    `--features transcript-screen` build with no `bench` (confirmed
    by building that combination alone with fake env vars) -- not
    something this box introduced, and not a clippy/rustc warning
    (AGENTS.md's "keep the build clean" gate is `cargo clippy`, which
    stays silent on it).
  
  **Packaging.** No `cargo xtask apk` exists for `iris/android-app`
  yet (I2's own Gradle project is the only pipeline), so this reused
  that split rather than inventing one: `cargo ndk --release` above
  builds the cdylib straight into `app/src/main/jniLibs/`, then a new
  `release` build type in `app/build.gradle` (there was previously
  only `debug`) packages and signs it --
  `AI_APP_KEYSTORE=~/.config/ai-app/release.jks` +
  `AI_APP_KEYSTORE_PASSWORD` (the same key `app/build-apk.sh`
  generates for the Compose app) via `gradle :app:assembleRelease`,
  with `applicationIdSuffix ".bench"` so it installs beside the plain
  tabs demo rather than replacing it. `aapt2 dump badging` on the
  result: `package: name='dev.iris.android.demo.bench'`, one native
  library, `lib/arm64-v8a/libmain.so`. `apksigner verify
  --print-certs` shows the same `CN=ai-app` certificate
  `compose-bench-arm64.apk` is signed with.
  
  **Emulator smoke run, 2026-09-05.** This checkout's own AVD
  (`ai-app-2`) was in use by the session recording I5's clean-scroll
  comparison in this same file (its Compose app was in the
  foreground, confirmed via `dumpsys window`/`dumpsys activity
  processes` before touching anything) -- rather than contend for it
  (AGENTS.md's "coordinate with peer agents"), a second,
  differently-named AVD was created (`AVD_NAME=ai-app-2-bench emu
  up`, `pixel_10`/`android-36`/`google_apis`/`x86_64`, cold boot, host
  GPU, no `EMU_GPU=software`), with 12GB of the VM's memory still
  available after both were up (this-machine-android's "two are
  comfortable" guidance). Installed via `adb -s emulator-5556 install
  -r`, launched, driven by `ui-trace record -s emulator-5556 --do
  "tap 'Run benchmark'"` (the control resolved by its accessibility
  label, per AGENTS.md's "no coordinate" rule), then read back over
  `adb logcat`:
  
      iris bench report
      frames=372 janky%=56.99 p50=19.5ms p90=219.5ms p99=284.5ms worst=369.3ms (measures redraw-start to after present() is called, not GPU/compositor completion) cpu_p50=0.4ms gpu_wait_p50=13.9ms (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: 24665ms
        peak RSS: 224600kB
        battery current: mean 900000µA over 21 samples (min 900000, max 900000)
  
  "Copy report" was pressed immediately after and logged `iris bench
  report: copied to clipboard` (`ClipboardManager.setPrimaryClip`
  succeeded). No crash (`adb logcat`'s `FATAL`/`AndroidRuntime` lines
  checked -- only `ui-trace`'s own runtime, unrelated), process alive
  throughout (`dumpsys activity processes`), 400/400 stream events
  confirmed sent.
  
  Read this the same way the Compose half's own box already asks to
  read its number: this is software-rasterised (well, GLES-over-virgl
  under `force-gles`, per I5's "Where iris's frame time goes")
  emulator output, "the harness runs end to end and produces every
  field P0 asked for," not a phone number -- and the battery current
  is again the emulator's fixed 900000µA mocked charger reporting a
  constant, exactly what the Compose box's own run found, not a real
  battery answering. `cpu_p50=0.4ms` (iris's own per-frame CPU work)
  against a much larger `gpu_wait_p50`/`p50` again matches I5's "Where
  iris's frame time goes" finding under real GPU rendering (`-gpu
  host`, `force-gles`) -- the frame-time budget here is dominated by
  the driver/compositor wait, not by iris's layout or primitive
  building, though this run's `janky%`/`p90`/`p99` are considerably
  worse than that earlier isolated pass, most likely the cost of this
  AVD's very first cold boot plus running two emulators on this VM at
  once (a fair comparison against Compose would need both apps run
  back-to-back on the same freshly-booted device, not attempted this
  pass since the second AVD was torn down immediately after per
  AGENTS.md's "stop yours when you are done with it").
  
  Copied to `~/host/bench/iris-bench-arm64.apk` (15,445,468 bytes) and
  `~/host/bench/README.md`'s "iris" section filled in (install, open,
  tap "Run benchmark", read the report from the on-screen text or
  logcat, tap "Copy report", paste back).
  
  **Not done this pass**: the actual on-phone runs and Iris's
  pass/fail call between the two reports (P0's own pass condition) --
  that needs Iris's phone, which this session has no access to.
  `iris/src/android/view.rs` was touched (`AndroidAppState::
  platform_ready`, `new_peer`'s wiring) -- confirmed to not be one of
  the three files the concurrent `device_limits()` work on this
  branch was using (`iris/core/src/render/mod.rs`,
  `iris/src/android/render.rs`, `iris/src/default/render.rs`).
  
  **Streaming no longer costs a full rebuild, 2026-09-05.** The gap
  named above and in E4/I5 (`push_row` can only append; a streaming
  reply is a row that keeps *changing* after it appears) is closed:
  `iris::widget::List` gained `replace_back` (swap the last row's
  widget in place, same slot, so a pinned-to-newest list stays pinned
  and an off-screen replace moves nothing on screen -- two new unit
  tests, `replacing_the_last_row_stays_pinned_to_the_bottom` and
  `replacing_the_last_row_out_of_view_does_not_move_visible_rows` in
  `iris/src/widget/list.rs`) and `clear` (drop every row, the fallback
  path). `transcript_ui::TranscriptScreen::apply(rsc, old_items,
  new_items)` diffs `group_tool_runs(old)`/`group_tool_runs(new)`
  (pure bookkeeping, no widget built doing it) and picks the cheapest
  update: unchanged (no-op), pure append (`push_row`, same as before),
  the common streaming case -- only the last row's content changed --
  rebuilds just that one row and swaps it in with `replace_back`, or
  (rare: `group_tool_runs` regrouping a row before the tail) a full
  `List::clear` rebuild, counted by `TranscriptScreen::take_rebuilds()`.
  Seven new unit tests in `transcript-ui/src/lib.rs`'s `diff_tests`
  cover all three cases directly against synthetic `Vec<FoldedRow>`s
  (no widget/Rsc needed for the decision itself). `bench_client.rs`,
  `transcript_client.rs` and `desktop-app/app.rs` all call `apply` now
  instead of rebuilding per event; `IRIS.md`'s 2026-09-05 entry has
  the full API account. `TextEditCtx` also gained `set_with_spans`
  (`iris/src/widget/text/edit.rs`) -- `set()` plus a fresh span list
  in one call, since a streamed row's re-rendered markdown needs both
  to land together.
  
  **Two new scripts, `iris/android-app/build-apk.sh` and
  `iris/android-app/run-bench.sh`**, written this pass after repeating
  the ANDROID_HOME/NDK-export/`cargo ndk`/Gradle-release/keystore/
  apksigner incantation by hand one too many times. `build-apk.sh
  [debug|release] [--abi arm64-v8a|x86_64] [--features "..."]` builds
  the cdylib and the APK and verifies it (badging, and signing for a
  release build); `run-bench.sh [--apk PATH]` installs on this
  checkout's own emulator (`emu serial`), taps "Run benchmark" by
  label (no coordinates), polls logcat for the report line, and prints
  it. Used for everything below and for the redelivery at the end of
  this box.
  
  **Numbers, this checkout's AVD (`ai-app-2`), release, x86_64,
  `force-gles`, via `run-bench.sh` -- three separate runs, same warm
  AVD (not a fresh cold boot each time):**
  
      frames=690 janky%=78.26 p50=26.9ms p90=60.3ms p99=103.4ms worst=130.1ms cpu_p50=7.4ms  gpu_wait_p50=15.7ms
      frames=691 janky%=84.95 p50=28.3ms p90=60.9ms p99=95.2ms  worst=120.7ms cpu_p50=5.4ms  gpu_wait_p50=18.2ms
      frames=691 janky%=58.32 p50=18.9ms p90=40.3ms p99=75.6ms  worst=101.4ms cpu_p50=3.5ms  gpu_wait_p50=12.5ms
  
  Against this same box's earlier iris-half reading (full rebuild per
  event, a *different*, freshly-booted x86_64 AVD, host GPU):
  `frames=372 janky%=56.99 p50=19.5ms p90=219.5ms p99=284.5ms
  worst=369.3ms cpu_p50=0.4ms gpu_wait_p50=13.9ms`. The tail is what
  moved: `worst` dropped from 369.3ms to 101–130ms and `p99` from
  284.5ms to 76–103ms across all three post-fix runs, consistent with
  removing the periodic full-tree-rebuild stall during the
  20-events/second streaming phase. `p50`/`cpu_p50` are *not* a clean
  comparison -- these three runs share one already-warm AVD instance
  rather than each getting its own fresh cold boot the way the earlier
  reading did, and `cpu_p50` in particular is noisy run to run (3.5 to
  7.4ms here) in a way a controlled A/B would need to separate from
  the code change itself. **What a future pass should do for a clean
  number**: two fresh cold boots of the same AVD, one per build,
  `run-bench.sh` on each, nothing else running.
  
  **The three remaining I5 verifications, closed 2026-09-05.** The
  `server`/`event_model` drift the previous pass hit (`no variant
  named 'LimitReached'`) was already fixed upstream on `rustify` by
  the time this pass started (commit `c07d544`, "carry main's
  `LimitReached` event") -- `cargo build --release` under `server/`
  is clean, `./ui-sandbox.sh start` builds and runs. All three items
  this left open:
  
  1. **Three clean `iris-scroll.sh` runs against a cold `-gpu host`
     boot, all 24/24 swipes scrolling in every run** -- results and
     the per-swipe verification method are recorded just above, in
     this same "Touch-scroll dropout root-caused" subsection's
     "Update, 2026-09-05" paragraph, and the host-GPU table above now
     carries the best-of-three row.
  2. **`EMU_GPU=software` + `force-gles`, cold boot -- still cannot
     isolate SwiftShader-Vulkan from SwiftShader-GL, now for a third
     and structural reason.** Same release build
     (`transcript-screen force-gles`), fresh cold boot under
     `EMU_GPU=software`. The earlier compute-limit abort this same
     box's "Fixed, 2026-09-05, later the same day" paragraph resolved
     (`iris_core::device_limits()` zeroing `max_compute_*`) no longer
     fires -- adapter selection now succeeds and picks up
     SwiftShader's ES 3.0 GL path -- but device creation aborts on a
     *different* limit immediately after:
         Abort message: 'Could not get device!: RequestDeviceError { inner: Core(LimitsExceeded(
         FailedLimit { name: "max_storage_buffer_binding_size", requested: 134217728, allowed: 0 } )) }'
     i.e. SwiftShader's ES 3.0 reports zero storage-buffer capacity at
     all -- SSBOs are an ES 3.1+ feature, the same generation gap the
     compute-limit failure came from, and exactly the trap this box's
     own "Fixed" paragraph flagged when it rejected
     `Limits::downlevel_webgl2_defaults()` for zeroing
     `max_storage_buffers_per_shader_stage` while `shader.wgsl`'s
     vertex stage reads four `var<storage>` buffers unconditionally.
     **Closing the open question, one sentence**: whether
     SwiftShader-Vulkan or GLES-in-general explains the ~80-150ms
     software-mode numbers cannot be answered on this hardware at
     all, because `shader.wgsl`'s storage-buffer reads make a GLES
     path on downlevel (ES 3.0) SwiftShader structurally unreachable
     rather than merely unmeasured -- reaching it is a shader rewrite
     (moving those reads off `var<storage>`), which is real scope, not
     a measurement task, and was not attempted here.
  3. **P0's bench build, cold-boot `run-bench.sh` number**: same
     `-gpu host` cold boot as item 1 (re-cold-booted after the
     `EMU_GPU=software` boot above), `./build-apk.sh release --abi
     x86_64 --features "transcript-screen force-gles bench"`,
     `./run-bench.sh`:
         frames=690 janky%=62.03 p50=19.6ms p90=42.4ms p99=56.5ms worst=62.5ms cpu_p50=4.4ms gpu_wait_p50=12.6ms
         scroll: 6 cycles (24 swipes), streamed 400/400 fixture events
         process CPU time over this run: 15394ms
         peak RSS: 164348kB
         battery current: mean 900000µA over 21 samples (min 900000, max 900000, the emulator's fixed mocked-charger reading, not a real battery -- see P0's own box)
     Against the P0 box's own three same-warm-AVD readings
     (`frames=690/691/691`, `p50` 18.9-28.3ms, `worst` 101-130ms),
     this cold-boot run's `worst` (62.5ms) and `p99` (56.5ms) are
     *lower* than any of the three warm-AVD runs, and its `p50`
     (19.6ms) sits inside their range -- so the P0 box's caveat that
     the warm-AVD numbers might be inflated by AVD staleness does not
     hold up under a fresh cold boot; if anything this run is cleaner.
     `cpu_p50` (4.4ms) is within the 3.5-7.4ms noise band the P0 box
     already flagged as run-to-run noisy on a shared warm AVD.
  
  **Redelivered, 2026-09-05.** `./build-apk.sh release --abi
  arm64-v8a` (arm64-only jniLibs; an earlier step in this same pass
  had left an x86_64 slice in there from the emulator testing above,
  removed before this build so the delivered APK matches P0's
  original arm64-only shape) -- `aapt2 dump badging` confirms
  `native-code: 'arm64-v8a'` and the same
  `dev.iris.android.demo.bench` id, `apksigner verify` the same
  `CN=ai-app` cert as before. Copied over
  `~/host/bench/iris-bench-arm64.apk`; `~/host/bench/README.md` gained
  a one-line build-date/commit note so Iris can tell which build she
  has.
  
  **iris bench crash on the phone, 2026-09-06.** The delivered APK
  (`dev.iris.android.demo.bench`, arm64, release) aborted on Iris's own
  phone (a Pixel, GrapheneOS, Mali GPU) on the very first
  `surface_changed`: `AndroidRenderer::new` -> `UiRenderNode::new` ->
  `create_bind_group_layout` -> wgpu's `default_error_handler` panics
  with `wgpu error: Validation Error`, and Android's crash report
  truncated the message right there, so the actual validation failure
  was unknown. Ran fine on the emulator's Vulkan (SwiftShader) and GLES
  (`force-gles`/virgl) and on the desktop GPU. Nobody on this session
  has the phone or `adb` access to it; this pass worked from the crash
  report alone plus reading wgpu-core's own validation source
  (`wgpu-core-28.0.0/src/{binding_model,device/resource}.rs`, the
  version this workspace pins).
  
  **1. Diagnostic, not guesswork -- what actually happens is now
  visible.** `iris_core::UiRenderNode::new` (`core/src/render/mod.rs`)
  wraps every `create_bind_group_layout`/pipeline call in three nested
  wgpu error scopes (one per `ErrorFilter`: `OutOfMemory`,
  `Validation`, `Internal`), pops them in reverse once creation is
  done, and returns `Result<Self, String>` -- the `String` is wgpu's
  own `Display` text for whichever scope caught something, which is
  already wgpu-core's `format_error` output (`"Validation Error\n\n
  Caused by:\n  ..."`, confirmed by reading
  `wgpu-28.0.0/src/backend/wgpu_core.rs`'s `format_error` -- the exact
  text the panic would have printed, just no longer thrown away).
  `android::render::AndroidRenderer::new` turns a failure into a full
  report: the adapter's name/backend/driver
  (`Adapter::get_info`), the limits bind-group-layout validation
  checks a storage/texture binding against
  (`max_storage_buffers_per_shader_stage`,
  `max_sampled_textures_per_shader_stage`, `max_bind_groups`,
  `max_bindings_per_bind_group`, `max_storage_buffer_binding_size`,
  `min_storage_buffer_offset_alignment`), and
  `DownlevelCapabilities.flags` (`Adapter::get_downlevel_capabilities`)
  -- then wgpu's own error text. `android::view::IrisViewPeer::
  surface_changed` logs it as one logcat line (`iris renderer init
  failed: ...`, newlines replaced with ` | `) and shows the full
  multi-line text on screen: a new `IrisView.showRendererError(String)`
  (an ordinary instance method Rust calls into via JNI, not a `native`
  one -- the direction is Rust reaching into Java, the opposite of
  every `native fn` this view already declares) swaps the activity's
  whole content for a plain, selectable, scrollable `TextView`, opening
  with "Copy this text and send it to Iris" (UI_RULES.md: a failure is
  reported where it happened and says what to do next). Desktop's
  `UiRenderer::new` keeps panicking on failure (no on-screen fallback
  exists there) but now with wgpu's full chain as the message, since it
  no longer relies on wgpu's own handler getting there first.
  
  **A real, separate reentrancy bug turned up while testing this, and
  is fixed alongside it.** Calling `Activity::setContentView` directly
  from inside `surface_changed` deadlocked -- not literally, but hit
  Rust's `RefCell already borrowed` abort: `setContentView` tears the
  old view hierarchy down synchronously, which fires `IrisView`'s own
  `onFocusChanged` *before* `setContentView` returns, straight back
  into the same `IrisViewPeer` through `on_focus_changed` while
  android-view's own dispatch (`with_peer` in its `view.rs`) still
  holds this peer's `RefCell` borrow for the `surface_changed` call in
  progress. Found by deliberately inducing a validation error (see
  below) and watching it abort a different way than the crash this
  pass was fixing. Fixed by moving the Java call into
  `ctx.push_dynamic_deferred_callback`, which android-view already
  runs only after dropping the borrow (confirmed by reading
  `with_peer`'s body) -- the same mechanism `raise_if_enabled` (this
  file's AccessKit push) already relies on for the identical reason.
  Left as a comment at the call site rather than only here, since the
  next thing that reaches into Java from inside a `ViewPeer` callback
  needs the same warning.
  
  **2. The audit -- every bind-group-layout entry, checked against
  wgpu-core's actual validation, not guessed.** `CreateBindGroupLayoutError`
  (`wgpu-core::binding_model`) has seven variants; the ones a static,
  no-`count`, no-feature layout like this crate's can hit are
  `Entry { error: MissingDownlevelFlags(_) }` and
  `Entry { error: MissingFeatures(_) }`. Walked every entry in
  `uniform_layout`, `primitive_layout`, `rsc_layout`, `masks_layout`
  (all four in `UiRenderNode::new`):
  - `uniform_layout` (group 0): one uniform buffer, `VERTEX|FRAGMENT`.
    Uniform buffers need no downlevel flag or feature at any
    visibility. Not it.
  - `primitive_layout` (group 1, `rects`/`glyphs`): two storage
    buffers, both `FRAGMENT`-only (confirmed against `shader.wgsl`:
    `rects`/`glyphs` are read only in `fs_main`). `FRAGMENT`-visible
    storage buffers need `DownlevelFlags::FRAGMENT_STORAGE`, which
    every backend in wgpu-hal grants unconditionally for a
    non-write-only binding (`ty: Storage { read_only: true }` here).
    Not it.
  - `rsc_layout` (group 2, atlas/image/sampler): a `D2Array` texture,
    a `D2` texture, a `NonFiltering` sampler, all `FRAGMENT`, no
    `count`. `Bt::Texture`'s only feature requirement
    (`TEXTURE_BINDING_ARRAY`) gates on `count.is_some()`, which none
    of these set. Not it.
  - `masks_layout` (group 3, `masks`/`move_offsets`): `masks` is
    `FRAGMENT`-only (read only in `fs_main`'s mask lookup). But
    `move_offsets` is `VERTEX | FRAGMENT` -- `shader.wgsl`'s
    `resolve_move` is called from both `vs_main` (a primitive's own
    corners) and `fs_main` (a mask's move chain) -- and it is a
    storage buffer, which is exactly what
    `wgpu-core/src/device/resource.rs` gates on
    `DownlevelFlags::VERTEX_STORAGE` whenever `visibility` contains
    `VERTEX` (confirmed by reading that check directly, not inferring
    it from the flag's name). **This is the one entry among all four
    layouts whose validity is device-dependent rather than static.**
  
  **Named hypothesis: the delivered build forced GLES, and GLES's
  `VERTEX_STORAGE` is not unconditional the way Vulkan's is.**
  Read `wgpu-hal-28.0.0`'s two backends' own downlevel-flag
  construction (`vulkan/adapter.rs`, `gles/adapter.rs`):
  - **Vulkan** grants `Df::VERTEX_STORAGE` unconditionally for any
    Vulkan 1.0 device, alongside `COMPUTE_SHADERS`/`FRAGMENT_STORAGE`
    and others in one unconditional `Df::empty() | ... ` -- there is
    no `.set(VERTEX_STORAGE, <device check>)` call anywhere in that
    file. This is why the emulator's SwiftShader-Vulkan run and the
    desktop's real Vulkan both pass: **on Vulkan, this exact layout
    cannot fail this check, on any conforming device.**
  - **GLES** computes it explicitly:
    `downlevel_flags.set(VERTEX_STORAGE, max_storage_block_size != 0
    && max_storage_buffers_per_shader_stage != 0 &&
    (vertex_shader_storage_blocks != 0 || vertex_ssbo_false_zero))` --
    i.e. it depends on the driver actually reporting a nonzero
    `GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS`. This is a known-weak spot
    on Android GLES drivers specifically (vertex-stage SSBO support
    lags fragment-stage support even on ES 3.1+ hardware), and this
    exact document already found the adjacent failure mode once this
    pass: SwiftShader's ES 3.0 GL path reports zero storage-buffer
    capacity at all (`max_storage_buffer_binding_size: 0`, this box's
    item 2 under "The three remaining I5 verifications"). A phone
    that negotiates a GLES context with no (or driver-buggy)
    vertex-stage SSBO support hits precisely this: `masks_layout`'s
    `move_offsets` entry, `MissingDownlevelFlags(VERTEX_STORAGE)`.
  
    **And the delivered build does force GLES.** `~/repos/
    ai-app-bench`'s own README (committed alongside the P0 APKs)
    says so directly: "`force-gles` matches I5's own ... finding: the
    default Vulkan backend has no adapter under a plain `-gpu host`
    boot on this AVD" -- true and reasoned correctly *for this VM's
    emulator*, but `build-apk.sh`'s default `FEATURES` applied the
    same flag to every arm64 build regardless of target, so the exact
    same cfg-locked GLES-only path that was a deliberate, documented
    emulator workaround shipped to a real Mali phone with no way to
    turn it back to Vulkan short of a rebuild. `force-gles`'s own doc
    comment (`iris/Cargo.toml`) only ever talks about the emulator
    ("the same build can be measured against SwiftShader's software
    Vulkan ICD ... or virgl's GLES path") -- real hardware was never
    the case it was written for.
  
  **What would confirm or kill this, from the on-screen report
  alone**: `backend: Gl` (confirms GLES was in fact what ran) and
  `downlevel flags: DownlevelFlags(...)` *not* containing
  `VERTEX_STORAGE` in the list. If a future report instead shows
  `backend: Vulkan` with `VERTEX_STORAGE` present, this hypothesis is
  wrong and the "Caused by" chain in that same report names the real
  one directly -- which is the entire point of doing (1) first.
  
  **Fix applied**: `build-apk.sh`'s default `--features` dropped
  `force-gles` (now `"transcript-screen bench"`, was
  `"transcript-screen force-gles bench"`), with a comment explaining
  why and telling a future emulator-isolation run to pass it back
  explicitly. This is a build/delivery fix, not a shader rewrite --
  the shader itself is untouched, since moving `move_offsets` off a
  storage buffer is real scope this document already declined once
  this pass for the adjacent SwiftShader-ES-3.0 finding, and the
  actual defect here is that a debug-only backend override reached a
  real device, not that the shader's design is wrong. The rebuilt
  arm64 APK (below) uses the default backend, i.e. Vulkan on a real
  phone -- if it still fails, (1)'s on-screen report is what comes
  back this time, not a truncated abort.
  
  **Verified, this checkout's emulator, cold `emu up`
  (`ai-app-2`):**
  - GLES (`force-gles`, matching every prior P0 GLES reading's
    backend): `run-bench.sh` end to end, no crash, report unchanged
    in shape from the pre-fix readings above
    (`frames=691 janky%=58.90 p50=19.2ms p90=43.7ms p99=62.4ms
    worst=69.3ms cpu_p50=4.7ms gpu_wait_p50=12.6ms`, 24/24 swipes,
    400/400 streamed events) -- the diagnostic wrapper adds no
    measurable cost or behaviour change on the success path.
  - **Induced failure, confirmed the fix works end to end**: added a
    temporary `count: Some(NonZeroU32::new(2))` to `uniform_layout`'s
    one entry (an artificial `ArrayUnsupported`/`MissingFeatures`,
    chosen because it is guaranteed to fail on every backend rather
    than depending on this VM's flaky adapter enumeration), rebuilt,
    installed, launched: logcat showed the full one-line report
    (adapter `Android Emulator OpenGL ES Translator (virgl ...)`,
    every named limit, the downlevel flags, and wgpu's "Caused by"
    chain naming `Binding 0 entry is invalid` / the missing
    `BUFFER_BINDING_ARRAY` feature) and `ui-trace elements` confirmed
    a `TextView` labelled with that exact report text was on screen
    -- process alive, no abort. This is also what caught the
    reentrancy bug above (the first attempt aborted a different way,
    `RefCell already borrowed`, fixed, then reproduced clean). The
    temporary `count: Some(...)` was reverted before anything else.
  - **Vulkan (no `force-gles`) could not be re-verified this pass**:
    this cold `emu up` enumerates zero Vulkan adapters
    (`wgpu_core::instance: enabled backend 'Vulkan' has no adapters`,
    the *unrelated*, already-panicking `.expect("Could not get
    adapter!")` path this fix does not touch) even with the default
    `-gpu host` boot the same README quoted above once relied on --
    matching this document's own prior notes that Vulkan
    availability on this VM's emulator is flaky across cold boots,
    not something this pass's diff caused (confirmed by checking the
    panic message is byte-for-byte the pre-existing
    `RequestAdapterError` shape, not a new one). Not chased further:
    it is orthogonal to the crash this pass fixes, and the real test
    of "does Vulkan work" is the phone itself, not this VM.
  
  **Checks, all clean**: `cargo fmt --all -- --check` and
  `cargo clippy --workspace --all-targets` (iris workspace, zero
  warnings beyond the pre-existing wgpu/winit future-incompat
  notice), `cargo test --workspace` (unchanged counts, all passing --
  nothing here touched logic under test), `cargo ndk -t x86_64 -P 26
  clippy --features "transcript-screen force-gles bench" --lib -- -D
  warnings` (clean), `cargo ndk -t arm64-v8a -P 26 build --release
  --features "transcript-screen bench"` (clean, arm64-only
  `jniLibs`).
  
  **Redelivered, 2026-09-06.** New arm64 APK (Vulkan, no
  `force-gles`), same `dev.iris.android.demo.bench` id, same
  `CN=ai-app` signing cert, copied to `~/host/bench/
  iris-bench-arm64.apk` and `~/repos/ai-app-bench/iris/build/outputs/
  apk/release/iris-bench-arm64.apk`; that repo's own README gained a
  dated entry explaining both changes (the diagnostic and the
  `force-gles` removal) so Iris can tell this build apart from the
  one that crashed. **Not done this pass**: confirming the fix on
  Iris's actual phone -- nobody on this session has it or `adb`
  access to it, so this is read from the crash report and wgpu's
  source, verified as far as this VM's tooling reaches, and handed
  back with a diagnostic that will say the real story on the next
  run either way.
  
  **Benchmark v2 (2026-09-06), asked for by Iris after using the
  Compose build on her phone**: "it doesn't fling like I typically do
  when scrolling up to find old messages. It should travel way faster
  which is better for stress testing. You may also want to add typing
  in the textbox as well and seeing how performant wrapping & pushing
  the transcript up are, and also keyboard performance if possible."
  **This is the one spec** -- written once here so both apps' "Run
  benchmark" implement the identical four phases; a change to a
  constant below has to be made in both `app/`'s `BenchRun.kt` and
  `iris/`'s bench client, together, or the two reports stop measuring
  the same thing while still looking like they do.
  
  1. **fling.** Starting pinned at the newest end
     (`listState.scrollToItem(0)` / iris's equivalent), 8 flings away
     from it (toward older messages) through the list's own real fling
     path -- Compose: `LazyListState.scroll { with(flingBehavior) {
     performFling(velocity) } }` using the screen's actual
     `FlingBehavior` (`ScrollableDefaults.flingBehavior()`, since
     `TranscriptList`'s `LazyColumn` never overrides it -- **not**
     `animateScrollBy`, which can only ever cover the fixed distance
     and time it is given and was Iris's complaint) -- each fling's
     `initialVelocity = 12,000 px/s`. That number is well above a
     moderate tween-swipe's implied speed (v1's `SWIPE_PX`/`SWIPE_MS`
     is roughly 4,500 px/s) and is meant to be a hard, fast flick for
     stress-testing, per Iris's ask. After each fling, wait for
     `isScrollInProgress` to clear (cap 3s; `performFling` already
     suspends until its own decay ends, this is belt-and-suspenders)
     plus 300ms between flings. Then 8 more flings back toward the
     newest end (`-12,000 px/s`). Record the list's first visible
     index/offset at the start, after the 8 outward flings, and at the
     end, so the two apps' *travel* can be compared directly rather
     than just their frame times.
  2. **stream. Unchanged from v1**: 400 tail events at 20/s (20
     seconds), pinned to the newest end before it starts (the same
     "Jump to latest" pin `stream-bench.sh` does).
  3. **type.** Pin to the newest end, focus the composer, show the IME
     if the platform allows it, then insert this **exact 600-character
     string** one character per 50ms through the composer's real
     `TextFieldValue` state (Compose: the same `input` state
     `onValueChange` writes; iris: whatever holds the composer's text
     today), then delete it the same way, one character per 50ms.
     Chosen for long, multisyllabic words specifically so the composer
     wraps across lines and the transcript above it is pushed upward
     by a growing box, which is what Iris asked to see measured:
  
         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!!!
  
     Report whether the IME was actually open during typing (it should
     be, from this phase's own show-IME step -- see phase 4 for what
     to say if the platform refuses to show it at all).
  4. **keyboard.** Show the IME (`WindowInsetsControllerCompat.show
     (WindowInsetsCompat.Type.ime())` against the window/view; iris's
     equivalent through its own shell), wait 1s, hide it, wait 1s;
     five cycles. Confirm each show/hide with the platform's own
     answer (Compose: `ViewCompat.getRootWindowInsets(view)
     ?.isVisible(WindowInsetsCompat.Type.ime())`, i.e. the same
     `WindowInsets.isImeVisible` fact `SessionScreen`'s composer-inset
     bug fix already relies on) rather than assuming the request
     worked -- if it is never confirmed shown even once, the report's
     `keyboard:` line has to say "**keyboard: could not be shown**"
     in words (UI_RULES: never present an inferred value as a
     measured one, and design the unknown/failed state before the
     answer's).
  
  **Frame accounting**: one recorder, not two. Mark each phase's start
  in the existing per-frame recorder (Compose: `FrameStats.markPhase
  (name)`, a list of `(name, frameIndexAtStart, wallClockAtStart)`
  alongside the existing `total`/`waited`/... arrays) and slice the
  same `FrameMetrics` samples by phase afterward
  (`FrameStats.phaseLines`) rather than running a second listener.
  
  **Report shape**: a `per phase:` block appears once any phase marks
  exist (empty/absent on an ordinary "Copy" press, which never marks a
  phase), one entry per phase: frame count, the phase's wall-clock
  duration, late count/percent (against the same refresh-rate budget
  the whole-run section uses), p50/p90/p99, and the worst single
  frame. Then **every existing whole-run section stays, unchanged in
  shape** -- `frames:`, `where the draw phase went:`, `work since this
  was last copied:` -- because that is what the emulator-baseline and
  phone-baseline numbers already on record in this file were read
  against, and a report that dropped or renamed one of those lines
  would silently stop being comparable to them. Finally `bench:` gains
  new lines beside the existing CPU/RSS/battery ones: the fling
  phase's total travel (start/outward/end index+offset), the typed
  character count, and the keyboard phase's shown/hidden-confirmed
  counts (or the "could not be shown" line).
  
  **Compose half: done, 2026-09-06.** `FrameStats.markPhase`/
  `phaseLines` (`app/androidApp/src/main/kotlin/com/example/aiapp/
  FrameStats.kt`), `debugReport`'s new `phaseFrames` parameter
  (`DebugStats.kt`), and `BenchRun.kt`'s four-phase `run` (fling via
  `ScrollableDefaults.flingBehavior()` captured in `SessionScreen` and
  passed down since it needs a `@Composable` call site; type via a new
  `composerFocus: FocusRequester` attached to the composer's
  `OutlinedTextField` plus a `setComposerText` callback that writes
  `input` the same way a keystroke does; keyboard via
  `WindowInsetsControllerCompat` against `LocalView.current`) are all
  in. `BenchRun.TYPE_TEXT` is the exact 600-character constant quoted
  above (verified `.length == 600`). A pre-existing, unrelated break
  in `MainActivity.kt`'s `benchSessionSummary()` (missing several
  `SessionSummary` constructor arguments added by a change this pass
  did not otherwise touch -- confirmed pre-existing by reproducing the
  same compile failure after stashing this pass's own diff) was fixed
  alongside this, since it blocked `compileBenchKotlin` outright and
  is in this session's own `app/` scope.
  
  Checks all clean: `ktfmtFormat`, `compileDebugKotlin`,
  `compileBenchKotlin`, `lintDebug`, `lintBench` (both "No issues
  found"), `testDebugUnitTest`. `grep -n "tap [0-9]" app/*.sh` still
  has its one pre-existing, unrelated hit.
  
  **Compose bench v2, emulator smoke run, 2026-09-05** (this
  checkout's AVD, cold `emu up`, `ui-trace` tap-by-label throughout --
  the dialog needed a swipe to reach "Run benchmark" below the fold,
  report read back over `adb logcat`):
  
      ai-app render report
      device: sdk_gphone64_x86_64 (Google), Android 16
      build: release
  
      transcript:
        124 events, 26 rows, 58 units loaded
        viewport 1714px, 2 units visible
        on screen: the list's own 0px, AssistantMsg 18732px
        0 tool calls and 0 groups open
  
      per phase:
        fling: 1620 frames over 32.3s
          late: 1537 (94.9%)
            total  p50 20.5ms  p90 29.2ms  p99 45.9ms
          worst 61.8ms
        stream: 1079 frames over 20.6s
          late: 1037 (96.1%)
            total  p50 21.0ms  p90 33.5ms  p99 39.3ms
          worst 51.2ms
        type: 3568 frames over 61.4s
          late: 3536 (99.1%)
            total  p50 23.8ms  p90 32.1ms  p99 38.5ms
          worst 50.3ms
        keyboard: 215 frames over 10.0s
          late: 212 (98.6%)
            total  p50 21.3ms  p90 37.6ms  p99 48.4ms
          worst 50.2ms
  
      frames:
        6482 frames over 124.3s at 60Hz (16.7ms budget)
        late: 6322 (97.5%)
        total  p50 21.7ms  p90 33.1ms  p99 45.3ms
        gpu    p50 17.4ms  p90 27.0ms  p99 30.6ms
  
      where the draw phase went:
        draw phase 1.27ms per frame, of which:
          the transcript: 0.16ms (measure 0.09, place 0.07, record 0.00)
          everything else: 1.10ms (87%)
  
      bench:
        fling: 8 flings out + 8 back at 12000px/s, travel start=idx=0/off=0px outward=idx=218/off=73px 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: 61192ms
        peak RSS: 195716kB
        battery current: mean 900000µA over 125 samples (min 900000, max 900000)
  
  Read this the same way the v1 emulator smoke run above is read: it
  proves the harness runs end to end and produces every field this
  spec asked for, not a phone number -- software rasterisation, and
  the fixed 900mA battery reading is the emulator's mocked charger
  again. Two things worth carrying forward: the **fling phase reached
  index 218** in 8 flings (against v1's `animateScrollBy` loop, which
  never moved past a handful of indices in the same 8-swipe count),
  which is the direct evidence the new fling travels "way faster" as
  asked; and **the emulator's software keyboard toggled and was
  confirmed by `isImeVisible` all 10 times**, so phase 4 is not a
  guaranteed "could not be shown" on every platform, only where the
  IME genuinely refuses. `frames:`'s 6,482-frame, 124.3s total matches
  the sum of the four phase durations (32.3+20.6+61.4+10.0 ≈ 124.3s),
  confirming the phase marks partition the whole run rather than
  overlapping or dropping frames between them.
  
  **Iris's first real phone report, 2026-09-06** (the redelivered,
  no-`force-gles` APK above): no crash. Two screenshots, before any
  touch: headings/links/code/table all render correctly. Four defects
  found and worked this pass:
  
  1. **Every glyph disappears on the first tap or scroll; rectangles
     stay drawn** (the keyboard case is the same thing -- a tap on the
     composer). **Not root-caused this pass.** Audited `GpuTextures`'
     atlas-grow/patch path, `ArrBuf`'s resize-on-length-change
     contract, and the masks/move_offsets/rsc bind-group rebuild logic
     in `core/src/render/mod.rs` against wgpu's queue-ordering
     contract -- everything read as spec-correct (a `queue.write_texture`/
     `write_buffer` issued before a later `queue.submit` is guaranteed
     visible to it on the same queue, and a dropped `Buffer`/`Texture`/
     `BindGroup` still in flight is kept alive by wgpu's own tracker).
     No violation found by static reading; reproducing needs either
     the phone or a Mali driver trace, neither available this pass.
     Instrumented for the next report instead: `Device::
     on_uncaptured_error` is now installed on the Android device
     (`WgpuErrorLog`, `android::render::AndroidRenderer`), and
     `IrisViewPeer::render` logs masks/moves-resized, atlas
     pages-grown and image bind-group creates for the first 10 frames
     after every `surface_changed` -- exactly the window this bug
     lands in. The bench screen's new "Diagnostics" button (below)
     surfaces the error log and adapter identity on demand.
  2. **Bold words render as blank gaps of the correct advance width**
     (regular, links, inline code render fine). Fixed by bundling Noto
     Sans/Noto Sans Mono (regular/bold/italic/bold-italic, static
     cuts, OFL) into `iris-core` and registering them ahead of the
     platform's own fonts -- `core/src/primitive/text.rs`'s
     `TextData::register_bundled_fonts`. Named hypothesis, not
     confirmed on the phone: the system "Roboto" on a modern Android
     device is the variable "Roboto Flex," and this crate's glyph
     path (`TextData::place`) does not apply `Synthesis`/variable-axis
     correction at all -- a bundled *static* per-style face sidesteps
     the question rather than answering it. `TextData::font_diagnostics`
     reports what got resolved; logged once at startup and shown on
     the Diagnostics page.
  3. **Text far too small** -- iris had no device-pixel-ratio handling
     on *either* platform before this pass (grepped for `scale_factor`
     across the whole crate: zero hits). `DisplayMetrics.density`
     (Android) / `Window::scale_factor()` (desktop) now divides every
     physical-pixel number (window size, touch coordinates, the
     shader's window uniform) down to logical units before it reaches
     layout, so a `font_size: 16.0` is 16 dp rather than 16 raw device
     pixels on a ~3x-density phone. Cost a second, real bug found only
     by measuring on this checkout's emulator after the first fix
     landed: `android::view::IrisViewPeer::surface_changed`'s call
     into `UiRenderState::resize` (the layout engine's own notion of
     the canvas, which every widget's absolute `PixelRegion` is
     computed against) was still being handed raw physical
     `width`/`height`, while `AndroidRenderer`'s side of the same
     resize had already switched to logical -- splitting layout and
     the shader into two different units. A fixed-size widget (the
     bench screen's `.height(56)` button row) exposed it at ~40
     physical px against the ~147px `56 * content_scale` predicts;
     a proportional (`rest(n)`) size hid it by adapting to whichever
     total it was given. Both are logical now. **Not fully verified**:
     a fresh-install emulator screenshot after both fixes shows
     visibly larger, readable text (`docs/bench/` has neither
     screenshot committed -- see AGENTS.md on transcripts/screenshots
     not going in this repo -- but the before/after is described in
     the commit), and the button row's own height still isn't
     obviously matching `56 * content_scale` on this run -- worth a
     second look with `ui-trace show --field box` once there's time,
     but not a blocker for the magnitude of the original bug (3x too
     small).
  4. **Status-bar inset not applied** -- confirmed nothing in this
     app ever read `insets().top` at all (`android/insets.rs` has
     carried `Insets.top` since it was written; nothing consumed it).
     Fixed with a new, generic hook: `AndroidAppState::
     on_insets_changed(rsc, LogicalInsets)`, called from `render()`
     exactly when `AndroidUiState::insets()` changes, in logical units
     matching everything else `content_scale` now divides.
     `BenchClient::on_insets_changed` rebuilds the root tree with
     `Padding::top(insets.top)` on the button row -- rebuilding the
     whole tree rather than one `WidgetPtr` slot's content, because
     the first attempt (a `Pad` dropped into an unrelated `WidgetPtr`
     slot with no height override of its own) did not propagate the
     wrapped span's fixed height correctly, which is what surfaced
     finding 3's `UiRenderState::resize` bug in the first place.
     Verified via `ui-trace show --field box`: the button row's top
     (150 physical px) sits 8px below `statusBarBackground`'s bottom
     edge (142px) on this checkout's emulator.
  
  **A named `Diagnostics` control now exists** (RUST.md's own earlier
  ask): a third button on the bench screen's top row, filling the
  existing benchmark-report `TextEdit` with adapter identity/backend/
  driver, font resolution, the atlas's live view count, every
  uncaptured wgpu error since surface creation, and the frame report
  -- `android::render::AndroidRenderer::diagnostics_report`. Uses the
  existing "Copy report" button/clipboard path rather than a second
  one.
  
  **Verified this pass, this checkout's emulator** (`EMU_GPU` default,
  `--features force-gles` -- this cold `emu up` again enumerated zero
  Vulkan adapters, the same pre-existing flakiness earlier boxes
  documented, not something this pass's diff caused): `cargo fmt --all
  -- --check`, `cargo clippy --workspace --all-targets` (zero warnings
  beyond the pre-existing wgpu future-incompat notice), `cargo test
  --workspace` (all passing, unchanged pure-logic counts), `cargo ndk
  -t x86_64 -P 26 check` clean, `./run-bench.sh` end to end
  (`frames=691`, 24/24 swipes, 400/400 streamed events, no crash),
  fresh-install screenshots and `ui-trace` box readouts for the four
  items above. **Not verified this pass**: the actual phone (no
  access), and item 1's root cause (needs either the phone's next
  Diagnostics-page report or a Mali trace).
  
  **Recorded but not fixed this pass** (a follow-up agent takes these,
  to avoid colliding with this pass's `bench_client.rs`/`view.rs`
  changes) -- see `IRIS_TODO.md`'s "From the phone, 2026-09-06":
  swiping has no momentum (stops exactly where the finger releases,
  unlike Compose's fling), and scrolling down sometimes jitters the
  text.
  
  **Redelivered, 2026-09-06, later the same day.** New arm64 APK
  (Vulkan, no `force-gles`, bundled fonts, content-scale fix,
  Diagnostics control), same `dev.iris.android.demo.bench` id, same
  `CN=ai-app` signing cert, copied to `~/host/bench/
  iris-bench-arm64.apk` and `~/repos/ai-app-bench/iris/build/outputs/
  apk/release/iris-bench-arm64.apk`; that repo's own README gained a
  dated entry. Still not confirmed on Iris's actual phone.
  
  **Redelivered again, 2026-09-06, a later pass.** Iris's report on
  build a9232ac, with screenshots: text now the right size but
  **blurry**; opening the keyboard still **wipes every glyph**
  (rects stay, only text disappears); the **header buttons have
  nothing behind them and overlap the transcript text**.
  
  **1. The keyboard wipe.** Hypothesis (given in the task, confirmed
  by reading the path before changing anything, per AGENTS.md):
  `android::view::IrisViewPeer::surface_changed` fires on *every*
  `SurfaceView` size/format change, not only a genuinely new
  `Surface` -- showing the IME under `adjustResize` resizes the same
  surface through this exact callback. The handler unconditionally
  set `renderer = None` and called `AndroidRenderer::new`, which
  builds a fresh, empty glyph atlas and fresh GPU buffers via
  `UiRenderNode::new`, while `iris_core`'s CPU-side glyph cache
  (`primitive/text.rs`) kept the atlas UV coordinates it had already
  handed out against the *old* atlas -- every glyph then drew from a
  rectangle pointing into a texture that had just been recreated
  empty. Confirmed by reading `AndroidRenderer::resize` (already
  existed, already did none of that -- only `surface.configure` and
  the window uniform) against what `surface_changed` was actually
  calling instead. **Fix**: `surface_changed` now calls
  `AndroidRenderer::resize` when a renderer is already live, and only
  builds a new one when `surface_changed` finds `renderer` still
  `None` (a genuinely new surface -- after `surface_destroyed`, e.g.
  backgrounding). Not independently re-verified against a forced IME
  resize on this pass's emulator (no display keyboard exercised
  end-to-end here); the reasoning is a direct code read plus the
  existing `resize` path already being surface-only, not a
  screenshot diff -- **the next agent with emulator time should do
  the before/after screenshot this box originally asked for.**
  
  **2. The blur.** Root cause: the P0 fix that made text the right
  *size* (dividing the whole window into a "logical" space, then
  letting the shader's NDC mapping stretch it back onto the real
  framebuffer) rasterised each glyph at the small, pre-stretch size
  and then displayed it stretched onto more physical pixels than it
  had texels for. **Fix, and the density-independent length unit
  Iris asked for the same day (IRIS_TODO.md) turned out to be the
  same fix**: `Len::dp`, resolved against a `density` now carried on
  `UiRenderState`/`Painter`, replaces the global stretch -- window
  size, touch and insets are physical pixels throughout again
  (`WindowInsets`, renamed from `LogicalInsets`), and
  `TextBuffer::shape` multiplies `font_size`/`line_height` by density
  before handing them to parley, so the atlas rasterises at the
  display's real physical resolution. Full design in docs/LAYOUT.md's
  "Density: `Len::dp`" section and the public-API summary in
  docs/IRIS.md's 2026-09-06 entry.
  
  **3. The header.** Only each button's own `rect(...)` painted
  anything, so the gaps between/around them and the status-bar strip
  above showed `CLEAR_COLOR` (black) one layer back, and the row's
  reserved height was three `abs` (now-physical-pixel) button boxes
  -- smaller than the dp-correct size the transcript below uses,
  which is what read as "overlap" once the two disagreed. Fixed with
  a `HEADER_SURFACE` rect stacked behind the whole row and every
  header size moved onto `dp(...)`.
  
  **4. Keyboard diagnostics, so Iris can report back even if a
  keyboard-triggered regression persists.** `on_insets_changed` now
  edge-triggers ~500ms after `ime_bottom` becomes non-zero, capturing
  the same report the on-screen Diagnostics button produces, logging
  it, copying it to the clipboard unprompted, and showing it in a new
  plain-view overlay (`IrisView.showDiagnosticsOverlay`, Copy/Close)
  that draws independently of iris's own renderer.
  
  **Verified this pass**: `cargo fmt --all`, `cargo clippy --workspace
  --all-targets` and `cargo clippy` on `android-app` (both `-D
  warnings`, zero beyond the pre-existing `tabs-ui` unused-dependency
  and wgpu future-incompat notices), `cargo test --workspace` (all
  passing), `cargo ndk -t arm64-v8a check`/`clippy` for both the
  `transcript-screen bench` feature set.
  
  **Then run on this checkout's own emulator** (x86_64 debug,
  `--features "transcript-screen force-gles bench"` -- this AVD has no
  Vulkan adapter under a plain `-gpu host` boot, matching every prior
  emulator finding in this file): `run-bench.sh` end to end, no crash,
  `frames=534 janky%=79.03 ... cpu_p50=1.3ms`, 24/24 swipes, 400/400
  streamed events -- unchanged in shape from prior readings, so the
  diff cost nothing on the success path. **Header background**:
  screenshot confirms the `HEADER_SURFACE` panel now sits behind all
  three buttons (`/tmp/bench-after-run.png` this pass). **Keyboard
  wipe**: forced a real `surface_changed` two ways -- `adb shell wm
  size 1080x1900` (screenshot before/after, text intact) and actually
  opening the soft keyboard via `settings put secure
  show_ime_with_hard_keyboard 1` + tapping the message field
  (ui-trace confirmed a real resize, elements moved -547px; keyboard
  visible in the screenshot, text still fully rendered, not wiped).
  Both are real evidence the reuse-renderer fix works, though neither
  is the literal before/after diff this box originally asked for --
  **still worth a deliberate side-by-side screenshot pair in a future
  pass.**
  
  **Found during this same verification, not fixed, needs a follow-up
  pass**: after the keyboard-triggered resize, the top button row
  appeared to render a **second time**, well below its real position,
  inside the transcript's scroll area (same colours/text, unmistakably
  the same three buttons) -- and a tap aimed at the composer's
  "Message" field landed on "Run benchmark" instead (a second
  benchmark run started, visible in logcat as two `iris bench report:`
  lines from one session). Only seen after a resize with the keyboard
  genuinely open; the plain `wm size` resize screenshot pair did not
  show it, nor did the fresh-install screenshot before either resize.
  **Not root-caused this pass** -- time ran out before isolating
  whether this is the `Span::DOWN` two-phase draw (LAYOUT.md's
  provisional-then-real placement) leaving a phase-1 primitive
  retained somewhere it should have been moved from, something
  specific to the keyboard's `on_insets_changed` rebuild racing a
  redraw, or unrelated to this pass's changes entirely (not verified
  against a build predating this session's commits, so do not treat
  "caused by this pass" as established -- MACHINE.md's pinned rule
  about not attributing without measuring applies here too). Also
  noteworthy: `capture_keyboard_diagnostics` never fired in this
  session (no "iris keyboard diagnostics" log line) despite the
  keyboard visibly opening -- `on_insets_changed`'s `ime_bottom` may
  not be populated the way expected on this emulator/API level, or
  the duplicate-row state above interfered; **also needs a follow-up
  pass** before relying on the auto-capture on a real phone.
  
  **Not verified this pass**: anything on Iris's real phone, the
  two-density crispness check IRIS_TODO.md's unit item asks for, and
  the two open items just above.
  
  **Fling and jitter, 2026-09-06.** The two `IRIS_TODO.md` "From the
  phone" items this box's own text names as follow-ups are fixed --
  `List::fling`/`VelocityTracker`/`FlingCalculator` (IRIS.md's
  2026-09-06 entry) and the `DragArbiter` slop-release jump (fixed by
  applying only the excess past `DRAG_SLOP` on the crossing frame,
  not the whole pre-threshold drag) -- both wired through
  `Selection::drag`'s release path, both covered by new unit tests in
  `iris/src/sense.rs` and `iris/src/widget/list.rs`. **Root-caused by
  reading `DragArbiter::update` and testing it directly, not by an
  emulator trace** -- this pass did not open an emulator, so the
  "trace the list's offset per frame" verification this box's own
  todo asked for is still open, as is a feel-check of the fling on
  real touch input.
  
  **Benchmark v2, iris half, done 2026-09-06, later the same day.**
  `bench_client.rs` implements all four phases against the identical
  constants this box's "Benchmark v2" spec names: fling (8 out + 8
  back at 12,000px/s through `List::fling`, waiting for
  `!is_scrolling()` capped 3s with a 300ms pause between, travel
  reported as `idx=N/off=Mpx` via a new `List::anchor_position_display`
  -- note this list's anchor does not necessarily change *slot* during
  a long scroll (the module's own documented design: the anchor is
  named by identity, not re-derived from what's on screen), so an
  iris travel reading is not apples-to-apples with Compose's
  `firstVisibleItemIndex`, which does change slot -- a real difference
  in what the two numbers mean, not a bug, and worth reading `off`
  rather than `idx` when comparing runs), stream (unchanged), type
  (the exact 600-character `TYPE_TEXT` constant, verified by a unit
  test, one char per 50ms into the composer's real `TextEdit` via
  `.set()` -- the same whole-string-replace shape `BenchRun.kt`'s own
  `setComposerText` uses, not a per-character insert), and keyboard
  (5 cycles through `bench_jni.rs`'s new `show_ime`/`hide_ime`
  `InputMethodManager` calls, confirmed from `on_insets_changed`'s
  real `ime_bottom` transitions via a new `ImeState` counter rather
  than assumed from the JNI call succeeding).
  
  `iris_core::render::frame_report::FrameReport` gained `mark_phase`/
  `phase_stats`/`late_at_hz` (new unit tests in `frame_report.rs`):
  phases are sliced by absolute frame index against a second ring
  (`index_ring`) alongside the existing duration ring, and late/jank
  is judged against a real Hz read from `bench_jni.rs`'s new
  `refresh_rate_hz` (`View::getDisplay().getRefreshRate()`) rather
  than the fixed 60Hz `JANK_THRESHOLD` every other caller still uses
  -- a separate method, not a parameter on the existing one, so
  nothing else in the codebase changes behaviour. `RING_CAPACITY`
  4096->16384 since one full v2 run is 3,000+ frames.
  
  **A real deadlock, found and fixed while wiring this up.** Getting
  a value back out of a task spawned via `rsc.spawn_task` has no
  built-in return channel (`ctx.update`'s closures are fire-and-
  forget), so a new `read_from_state` helper sends the result through
  an `mpsc` channel and polls for it. Its first version only worked
  for the *first* call in a chain: nothing about `ctx.update` drains
  itself, so unless something calls `redraw.request_redraw()` after
  *this specific* enqueue, nothing ever runs the closure -- and every
  call after the first relied on a stale, already-fired
  `request_redraw()` from a previous step. The fix is structural:
  `read_from_state` now takes the redraw handle and calls it itself,
  immediately after enqueueing, every time.
  
  **Verified end to end, this checkout's own emulator (cold `emu up`,
  `force-gles`, x86_64 -- this AVD again enumerates zero Vulkan
  adapters on a cold boot, matching every prior finding in this
  file):**
  
      iris bench report
      per phase:
        fling: 1481 frames over 53.2s
          late: 158 (10.7%)
            total  p50 11.2ms  p90 16.9ms  p99 26.5ms
          worst 43.3ms
        stream: 401 frames over 20.8s
          late: 342 (85.3%)
            total  p50 26.1ms  p90 49.1ms  p99 57.2ms
          worst 61.3ms
        type: 1202 frames over 63.1s
          late: 89 (7.4%)
            total  p50 12.8ms  p90 15.1ms  p99 23.3ms
          worst 26.7ms
        keyboard: 9 frames over 9.1s
          late: 2 (22.2%)
            total  p50 6.3ms  p90 25.8ms  p99 25.8ms
          worst 25.8ms
  
      frames:
        3093 frames over 146.3s at 60Hz (16.7ms budget)
        late: 591 (19.1%)
        total  p50 12.4ms  p90 22.2ms  p99 51.2ms
        worst 61.3ms
        cpu_p50 0.7ms  gpu_wait_p50 11.6ms
  
      bench:
        fling: 8 flings out + 8 back at 12000px/s, travel start=idx=651/off=1336px outward=idx=651/off=101672px end=idx=651/off=1427px
        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: 43303ms
        peak RSS: 193152kB
        battery current: mean 900000µA over 146 samples (min 900000, max 900000)
  
  Read this the same way every prior emulator smoke run in this box
  is read: software rasterisation, not a phone number, and the
  battery line is the emulator's fixed mocked-charger constant again.
  **Travel**: the `idx` stays fixed at 651 through the whole fling in
  both directions (see the `anchor_position_display` caveat above) --
  `off` is what actually moved, growing to 101,672px outward before
  the return trip brings it back near its start, which is real, large
  motion (a fast, hard fling, matching Iris's "travel way faster"
  ask), just not directly comparable to Compose's idx-188-reached
  reading from the same box's earlier v2 entry. **`keyboard: could
  not be shown`**: expected given the ime-inset finding below, not a
  new regression.
  
  Redelivered: `./build-apk.sh release --abi arm64-v8a --features
  "transcript-screen bench"` (Vulkan, no `force-gles`; the x86_64
  jniLibs slice left over from emulator testing was removed first so
  the delivered APK is arm64-only, confirmed via `aapt2 dump
  badging`), `apksigner verify` shows the same `CN=ai-app` cert,
  copied to `~/host/bench/iris-bench-arm64.apk` and `~/repos/
  ai-app-bench/iris/build/outputs/apk/release/iris-bench-arm64.apk`;
  that repo's own README gained a dated entry. `run-bench.sh`
  extended for the longer run (260s poll cap, `-A 60` instead of
  `-A 6`) to fit v2's four phases.
  
  **Redelivered, 2026-09-06, the defect pass.** `./build-apk.sh
  release --abi arm64-v8a --features "transcript-screen bench"`
  (Vulkan, no `force-gles`; the x86_64 `jniLibs` slice from this
  pass's emulator work was removed first, confirmed arm64-only by
  listing the APK's `lib/` entries), `apksigner verify` showing the
  same `CN=ai-app` cert, copied to `~/host/bench/
  iris-bench-arm64.apk`; that README gained a dated entry naming what
  to look for. What changed: the composer sits at the bottom of the
  screen at launch again (the black third was the bench shell's empty
  report pane, not an inset -- see the plan box above), typing into it
  works at all (the empty-field caret bug), the keyboard no longer
  throws up an undismissable diagnostics overlay, and every surface
  and insets event is logged so a phone `logcat` can answer the
  app-switch text loss. Not fixable from here and still open: the
  duplicated `Compacted:` row.
  
  **(a) The header-duplicate bug (found by a concurrent pass on this
  branch): investigated, not fixed.** Reproduced reliably
  (`ui-trace record --do "tap 'Message'"` then `adb exec-out
  screencap`): the three-button row renders a second, full copy
  inside the transcript area the moment the keyboard opens. Read
  `Span::draw`'s own two-phase placement doc (a provisional
  full-region draw to learn each child's size, then a real
  `widget_within` placement) as the most likely mechanism, since it
  is the one place in this tree that deliberately draws a widget
  twice in normal operation and relies on the two draws landing at
  the same place to stay a cheap move rather than a visible second
  copy -- and `UiRenderState::update`'s `redraw_all`-vs-
  `redraw_updates` split (LAYOUT.md) means a `.set()`-driven targeted
  redraw of just `top_bar` and a resize-driven full redraw of the
  whole tree are two structurally different code paths that could in
  principle disagree about where that widget's primitives belong on a
  frame where both fire close together. **One concrete, testable
  hypothesis was ruled out**: `on_insets_changed` rebuilding
  `top_bar` on every call, including ones only about `ime_bottom`
  (nothing to do with the header's own padding). Added a guard
  (`last_top_pad`, skips the rebuild unless `insets.top` itself
  changed) and reproduced the *exact same* duplicate afterward --
  unchanged, byte-for-byte, in the same screenshot -- so repeated
  rebuilding is not the cause; the guard is kept anyway since it is a
  real (if here insufficient) reduction in needless work. **Not
  root-caused**: doing so needs either instrumentation inside
  `Span::draw`/`draw_inner` to see the two placements' actual regions
  on the frame the bug happens, or the phone. Left for a follow-up
  pass rather than guessed at further.
  
  **(b) Why the keyboard phase and the keyboard-open auto-diagnostics
  both read "not confirmed": a real, named platform interaction,
  partly fixed.** `MainActivity.java`'s manifest declares
  `windowSoftInputMode="adjustResize"` (AGENTS.md's own "Things that
  have bitten": without it the keyboard pans the window off screen
  instead of resizing it). Under `adjustResize`, `WindowInsets.
  Type.ime()`'s own inset *amount* is defined to read zero once the
  window has already resized to avoid the overlap that inset would
  otherwise describe -- confirmed by reading Android's own
  `WindowInsets` contract, not guessed at. So the numeric `ime_bottom`
  this app was reading is *structurally* never going to be positive
  here, independent of anything wrong in `iris`'s own code -- the same
  trap AGENTS.md already names for the Compose side
  (`WindowInsets.isImeVisible` "does not share the failure mode").
  **Fixed**: `MainActivity.java`'s `OnApplyWindowInsetsListener` now
  reads `insets.isVisible(WindowInsets.Type.ime())` (a boolean,
  unaffected by resize-vs-pan) and passes `1`/`0` through the
  existing `ime_bottom` JNI field instead of the always-zero numeric
  inset -- correct on its own terms, and kept, but **did not by
  itself make the keyboard phase or the auto-diagnostics fire on this
  emulator**: `logcat` shows the platform's own `InsetsController:
  show(ime(), fromIme=false)`/window-resize events happening (the
  keyboard genuinely opens, confirmed by screenshot), but no further
  `setOnApplyWindowInsetsListener` callback at all after the initial
  one at attach. Named hypothesis, not confirmed: a plain (non-edge-
  to-edge) `Activity` that has not called `WindowCompat.
  setDecorFitsSystemWindows(window, false)` may not get insets
  redelivered for a pure IME toggle handled entirely via resize --
  only the initial attach-time dispatch is guaranteed. Confirming and
  fixing that needs opting the activity into edge-to-edge, which is a
  real window-behaviour change interacting with the exact
  `adjustResize` setting AGENTS.md protects, not attempted this pass
  given the risk-to-time-remaining ratio. Both open items are
  recorded in `~/repos/ai-app-bench`'s README with today's date.
  
  **Composing text, the tap-vs-swipe focus rule, and app-switch text
  loss, 2026-09-06.** Iris's report on this same dc01f88 build: typing
  doesn't enter text or move the caret until a space is hit; typed
  text doesn't visibly appear and there is empty black space below the
  composer bar; text disappears again after leaving and returning to
  the app; and (a follow-up message the same day) swiping over the
  composer bar wrongly summons the keyboard.
  
  1. **The caret/composing bug's cause**: `android/ime.rs`'s
     `InputConnection` never called `InputMethodManager.updateSelection`
     after an edit -- confirmed by reading android-view's own demo
     (`~/src/android-view/demo/src/lib.rs`'s `render()`), which calls it
     every time its editor's generation changes. Without it, Gboard has
     no confirmation the app is keeping up and holds keystrokes back
     rather than trusting a screen it believes is stale -- exactly
     "doesn't enter it until I hit space." **Fix**: `IrisViewPeer::
     update_ime_selection` (new, `ime.rs`) reports the real selection
     and (an approximation, `compose_len` chars back from the caret)
     the composing region, called from `after_input`'s existing tail so
     every touch/key/IME callback already runs it. The buffer-level
     half (`replace`/`insert_str` correctly advancing the caret) was
     already correct and is now covered by four new unit tests in
     `iris/src/widget/text/edit.rs` (composing, `commitText`,
     `deleteSurroundingText`, `setSelection`). **Verified**: on the
     emulator (`force-gles`, no Vulkan adapter on this AVD), tapping a
     real Gboard key now shows a real, single-character-appropriate
     suggestion strip ("H | How | Hey") rather than stale state, and a
     `render()` log line fires for every keystroke -- both confirm the
     `InputConnection` calls are landing and are being processed, which
     a hand-typed `adb shell input text` did *not* reliably exercise on
     this AVD (no `render()` at all followed one such call -- most
     likely a modern `input text` no longer round-trips through
     `commitText` the way older docs assume; Gboard-key taps are the
     real path and the one this fix was verified against).
  
  2. **A second, deeper bug found while verifying (1), not root-caused
     this pass**: composed text never becomes visible on screen at
     all -- the grey composer bar stays empty, with no glyph anywhere
     in the frame, confirmed on repeated Gboard-key taps and across a
     keyboard-resize. **Ruled out**: the widget tree's own layout math.
     A new unit test, `layout_tests::
     composing_text_after_a_keyboard_resize_lands_in_the_bars_own_region`,
     builds the composer's exact tree shape (`Stack{rect, Span{Pad{
     TextEdit}}}` inside an outer `Span::DOWN`) with no GPU or window,
     resizes it the way a real keyboard-triggered `surface_changed`
     does, edits the field both before and after, and asserts the
     field's `window_region` stays a small box near the bottom of
     whichever window size is current -- it passes, both before and
     after this pass's composer rebuild (item 3 below), so the CPU-side
     region a redraw lands at is provably correct. The bug is
     therefore downstream of that -- most likely something specific to
     the GPU-side redraw a content-only edit takes (`UiRenderState::
     redraw`, which redraws a single dirtied widget directly at its
     stored region rather than re-running its ancestors' layout) or to
     this AVD's forced `force-gles` backend (the only one available
     here; Iris's phone deliveries have used real Vulkan) -- neither
     isolated this pass. **Not attributable to this pass's changes**:
     reproduced identically before touching `composer.rs` (the very
     first build tested, before the composer rebuild below, already
     had it) and the render-engine files this pass did not touch
     (`core/src/render/mod.rs`, `core/src/ui/render_state.rs`) are the
     likely next place to look -- specifically `UiRenderState::redraw`'s
     reuse of a widget's own last-drawn region versus a full tree walk.
     **Needs**: either a Vulkan-capable emulator boot or the real phone
     to rule `force-gles` in or out, and a GPU-side primitive dump
     (the existing `frame diagnostics` log line, extended to name which
     primitives a frame actually wrote) to see whether the glyph quads
     are emitted at all or emitted somewhere off-screen.
  
  3. **The composer bar rebuilt as one widget**, per this box's own
     ask: `transcript_ui::composer::build_composer` (unchanged
     `Stack{background, Span{Pad{TextEdit}}}` idiom, the same one the
     header row's `HEADER_SURFACE` already uses) now also caps the
     field at roughly six lines (`MaxSize` + `.scrollable()` for a
     wheel/trackpad overflow scroll -- a real touch-drag scroll on
     overflowing composer text is not wired and is a follow-up) and
     wraps the whole bar in one `Pad` whose `bottom` a new
     `Composer::set_bottom_inset(rsc, inset)` rewrites in place
     whenever the platform's insets change, called from
     `bench_client.rs`'s existing `on_insets_changed` with
     `insets.bottom.max(insets.ime_bottom)` -- the IME's own inset
     while it is open, the navigation bar's otherwise. Rewritten in
     place rather than rebuilt through a `WidgetPtr` swap (`top_bar`'s
     own pattern) because the field is strongly owned inside this tree
     and cannot be re-added to a new wrapper without panicking
     ("was already added") -- rebuilding would also drop focus,
     selection and in-progress text on every keyboard toggle.
     **Verified**: `ui-trace` box readouts before/after a keyboard
     open on the emulator (the field's row correctly reports a
     547px move matching the real IME-triggered resize); the
     known-separate "top row renders twice after a keyboard resize"
     bug this box already recorded is unrelated and still open. **Not
     fixed by this alone**: item 2 above -- the text still does not
     render, so the "empty space at the bottom" symptom's other half
     (nothing filling the space the bar itself now correctly reserves)
     needs item 2's fix first before a real before/after screenshot is
     worth taking.
  
  4. **App-switch text loss, fixed and verified.** `surface_destroyed`
     (backgrounding) drops the whole `AndroidRenderer` -- device,
     atlas, buffers -- and a subsequent `surface_changed` with no live
     renderer builds a genuinely new one (`AndroidRenderer::new`,
     distinct from the keyboard-resize path this box already fixed by
     *reusing* the renderer). But `iris_core::TextData::atlas` (the
     CPU-side glyph cache) and `UiData::textures` (the CPU-side texture
     bookkeeping the atlas is built on) live on `AndroidRsc`, which
     outlives any one `AndroidRenderer` -- so both kept pointing at the
     *old*, now-destroyed device's textures across the switch, the
     exact "rectangles stay, glyphs disappear" shape, just triggered by
     backgrounding instead of the keyboard. **Fix**: new
     `GlyphAtlas::clear()` and `Textures::reset()` (`iris/core/src/
     render/atlas.rs`, `iris/core/src/primitive/texture.rs`), called
     together from `surface_changed`'s "genuinely new renderer" branch
     only -- the same `already_live` check that already decides
     reuse-vs-new, so this is one mechanism gated on the one condition
     that needs it, not a second ad hoc check. **Verified on the
     emulator**: backgrounded via `KEYCODE_HOME`, reopened via
     `am start`, screenshotted -- every pre-existing glyph (headings,
     body text, the whole diagnostics report) is intact, `frame_count`
     resets to 1 confirming a genuinely new renderer was built, no
     crash.
  
  5. **Swipe-vs-tap focus, fixed and verified** (Iris's follow-up the
     same day: "if I swipe over the input bar it brings up the
     keyboard... scrolling should be pinned"). `attr.rs`'s `Selector`/
     `Selectable` registered `CursorSense::click_or_drag()`, which
     calls `select()` -- and so grants focus and requests the IME --
     on the *first* frame of any press, before it is known whether the
     gesture will end up a tap or a drag. Rewritten around a shared
     `on_press` dispatcher over `PressStart`/`Pressing`/`PressEnd`: a
     field that is **already** focused behaves exactly as before
     (every frame updates the selection, so dragging inside a focused
     field to select text still works); a field that is **not**
     focused records where the press began (`TextEdit::press_origin`,
     new field) and only grants focus on `PressEnd` if no intervening
     frame crossed `sense::DRAG_SLOP` -- a drag recognised early simply
     clears the pending tap and does nothing further, so it is never
     consumed and whatever is behind the field still sees every frame
     of it. New `FocusHost::is_focused` (both platform impls) is what
     lets `on_press` tell the two cases apart. **Verified on the
     emulator**: `dumpsys input_method`'s `mInputShown` reads `false`
     after a `swipe` gesture starting on the composer bar (`ui-trace`
     confirms the field's own box never moved, i.e. no keyboard-driven
     resize happened), and reads `true` after an ordinary `tap` on the
     same field. **Coordination note**: a concurrent pass is moving
     drag arbitration into `sense.rs` behind a new `Drop` event: this
     fix touches only `attr.rs` (new `press_track`/`on_press`) and
     `iris/src/widget/text/edit.rs` (the new `press_origin` field), not
     `sense.rs` itself, so it should merge cleanly, but the next agent
     through here should check whether `Selector`/`Selectable`'s
     `Pressing`-frame delivery still arrives the way this code assumes
     once that lands.
  
  **Checks this pass**: `cargo fmt --all` clean, `cargo clippy
  --workspace --all-targets` and `cargo ndk -t x86_64 -P 26 clippy
  --features "transcript-screen bench force-gles"` both zero warnings
  beyond the pre-existing `tabs-ui` unused-dependency notice, `cargo
  test --workspace` all passing (new tests: four in `edit.rs`, one in
  `layout_tests.rs`). **Not done**: item 2's root cause; a real
  before/after screenshot pair for item 3 (blocked on item 2); anything
  on Vulkan or the real phone.
  

• P1 — session screen parity. Started 2026-09-06, on Iris's word: "just continue with the plan for now; try to move towards feature parity for the transcript screen so that the test can be more fair." So P0's "must pass before P1 starts" is lifted — the phone bench continues alongside, and parity is what makes its comparison fair. Sub-order, decided by the design agent, by what the bench fixture exercises and Compose already draws (each is one agent; tick and date in place): - [x] P1a — markdown block rendering parity. Done 2026-09-06. Each top-level block is drawn in one of three frames (transcript-ui::markdown::BlockFrame, mapped from BlockKind by the pure frame_of): Plain (a paragraph, heading, list or rule -- text and spans, no extra widget), Verbatim { fill } (a fence or a table -- a rounded panel that does not wrap and pans sideways, CodeFence.kt's horizontalScroll), and Quote (a bar behind text padded past it). Everything else markdown can say is expressed in SpanStyles, which cost no widgets. What each block looks like now, against Markdown.kt: - Headings -- Material's own ladder, the six sizes markdownTypography picks (24/22/16/14/12/11 at a 16pt body), bold. Was a three-step 28/24/21/19. - Fences -- monospace on Mocha Crust, rounded, with client_core::highlight's spans by language in the same Catppuccin palette Theme.kt's catppuccinSyntax() uses. An unknown language is plain rather than coloured by the nearest one. A fence being streamed into re-renders only the last block (RowBlocks::apply_delta), so earlier fences are never re-scanned. - Lists -- the bullet ladder MarkdownPieces.kt draws (disc/ring/square by depth) and ordered lists counting from the number written, markers in Lavender. - Tables -- padded monospace columns measured from the cells, header bold, a rule under it, on Surface 0. A real grid was rejected; docs/DECISIONS.md, 2026-09-06, has why. - Quotes -- a Surface 2 bar down the left, text one shade back from body. - Links -- coloured and underlined as before, and now tappable: GestureOutcome::Tapped (a press that committed to neither a pan nor a selection), TextEditCtx::byte_at for which byte, and iris::platform::OpenUrl for the platform (xdg-open/ open/start; on Android an ACTION_VIEW intent deferred to after_input, the shape pending_show_keyboard uses). Screenshots: docs/bench/p1a-2026-09-06/. compose-heading-fence-table.png and compose-fence-table.png are the Compose bench build on this checkout's AVD against app/bench-fixture/; iris-blocks.png is iris rendering the same heading, paragraph, link, fence and table source (plus a list and a quote, which the fixture has neither of) from transcript-ui's own transcript example. Why the iris half is not from the emulator, which the pass condition asked for: the emulator cannot draw iris's glyphs at all. Every character comes out as a solid filled box of the right width -- iris-emulator-gles-glyphs.png. Established as not this change's doing and not the app's: the previous commit (20303e0) draws the same boxes, and the Compose bench build on the same AVD in the same minute draws text perfectly. The atlas sample's alpha reads as 1 under -gpu host + -feature -Vulkan (Mesa 26.2.2 / virgl), which is what an incomplete GL texture returns (0,0,0,1). Both ways out were tried and both fail: GPU_HOST_FEATURES=" " still dies at boot with gfxstream's documented "Format VK_FORMAT_R8G8B8A8_UNORM is not supported ... Failed to find memory type for ColorBuffers", and EMU_GPU=software does give the guest SwiftShader Vulkan but iris SIGSEGVs inside surface_changed on it. So the appearance half of this box is taken on the desktop/winit backend, which renders on the host's real GPU through iris/run-headless.sh. What still differs, pair by pair: 1. Colour, on the desktop shot only. The winit surface is sRGB and the shader writes the palette's bytes as linear, so every fill reads ~4x lighter: Crust (17,17,27) comes out (73,73,91), measured. Not a palette error and not present on Android, where the previous pass measured the composer bar at rgb(41,40,49) for a declared (40,40,46). Worth its own item; it makes the desktop build a poor colour reference until fixed. 2. A list's wrapped line. Compose lays an item out as a marker column beside a text column, so a second line stays indented; iris writes the marker into the same buffer, so a wrapped line returns to the left margin. Needs per-line indent in TextAttrs. 3. A table. Compose draws a real grid, cells wrapping at a 136dp floor; iris draws padded monospace columns. Same information, different picture. 4. Inline code. Compose draws a chip behind it; iris gives the range a monospace face and the code colour. Unchanged by this box -- still blocked on per-range glyph geometry (IRIS_TODO). 5. A user message. Compose draws it in a rounded card; iris draws a sender label above plain text. That is the row's own styling, P1's rather than P1a's. One real defect found and fixed on the way, and it is not a small one: Rect::is_size_independent() answered true. A Rect fills whatever region it is given, so its content is the region -- and draw_inner's size-independent fast path, which rewrites a widget's primitives with r.outside(&from).within(&region) instead of redrawing, cannot reproduce that remap once a region carries both rel and abs. The visible result: a fenced block's background kept the height of the provisional full-region draw Span does in its first phase, so one fence's panel covered every block below it and every row below that, while the text underneath was laid out correctly. It answers false now (iris/src/widget/rect.rs, with the account at the definition). This is very likely the same family as this file's older "the composer bar's grey background is not drawn" note and any other .background(rect(..)) tint. One defect found and left open, with its repro: UiRenderState::reposition's debug assert -- "widget ... is both moved by its parent's own layout (mov) and repositioned within it" -- fires from List::place when a transcript row's blocks wrap. Reproduce in one line: change .wrap(!verbatim) to .wrap(true) in transcript-ui/src/row.rs's build_block and run iris/run-headless.sh transcript --shot /tmp/x.png -- -p transcript-ui. It is not caused by the Rect fix above (it survives it) and not by any one block kind (bisected: it appears once the row is tall enough). The shipping configuration does not reach it -- verbatim blocks do not wrap -- and neither does the Android bench, which ran clean with the assertions live. It should be the next thing looked at under P1, because it is a real inconsistency about who owns a widget's move slot, not a false alarm. Bench, stream phase, this checkout's AVD, debug x86_64 force-gles, assertions live, no abort: stream: 294 frames over 21.0s, late 283 (96.3%), p50 53.0ms p90 108.6ms p99 132.0ms against the pre-P1a p50 52.8ms p90 108.1ms p99 137.3ms -- unchanged, which is the point: block styling is span work, not layout work. The worst figure is the one number that moved and it does not reproduce: 567.3ms, 140.1ms and 664.5ms across three runs of the same build, against 148.9ms before. Unexplained; it is a single frame in 294 and the percentiles are flat, so it reads as an emulator hiccup rather than a cost, but it is written down rather than rounded off. Checks: cargo fmt --all --check clean in both workspaces; cargo clippy -p iris -p iris-core -p transcript-ui -p desktop-app -p tabs-ui --all-targets warning-free; cargo test 85 (iris, +4) + 13 (iris-core) + 31 (transcript-ui, +11) + 123 (client-core). 2026-09-06, after P1a: "the emulator cannot draw iris's glyphs" was iris's bug, not the emulator's. The finding recorded in the box above -- that every glyph is a solid filled box under -gpu host GLES and that this is what an incomplete GL texture returns -- had the mechanism right and the attribution wrong. It is a real defect on any adapter that is GL rather than Vulkan. - Reproduced off the emulator entirely, which is what made it cheap: default/render.rs now honours the same force-gles feature android/render.rs did, so ./run-headless.sh transcript --shot /tmp/x.png -- -p transcript-ui --features iris/force-gles draws the boxes on this machine's own GPU in seconds. Two shader probes then said what the sample was: return vec4(texel.rgb, 1.0) drew black boxes and return vec4(texel.a, texel.a, texel.a, 1.0) drew white ones, so the atlas sample was exactly (0, 0, 0, 1) -- GL's answer for an incomplete texture unit, and not the null texture (which is zeroed, alpha 0). - Root cause: the glyph atlas array was created with one layer. GpuTextures::new started array_capacity at 1 and grow_array only doubles once a page needs a layer past it, so the ordinary case -- one atlas page -- is a one-layer array. wgpu-hal picks the GL texture target from the descriptor alone (gles::Texture::get_info_from_desc: (false, 1) => TEXTURE_2D), so that array is created as a GL_TEXTURE_2D and then bound to the shader's sampler2DArray. wgpu has a name for this (log_failing_target_heuristics, its issues #1614/#1574); the result is an incomplete unit, texel.a == 1, and draw_glyph's color.a *= texel.a paints the whole glyph quad. - Fix: MIN_ARRAY_LAYERS = 2 in iris/core/src/render/texture.rs -- the array is never created with fewer, with the account at create_array_texture and a debug_assert! there so a future capacity arithmetic change fails at the mistake rather than as boxes on a screen. Cost: one page of texture memory, which the next atlas page uses anyway. - Not a regression from 3e72a4e..20303e0, and the bisect was not run. The defect is a function of the layer count, not of any commit in that range: it has been there since the atlas became a texture_2d_array (TEXTURES.md, 2026-09-04) and it reproduces at HEAD and disappears at HEAD with the one-line capacity change. The claimed "visible text at 3e72a4e" is a misreading of its own evidence -- /tmp/final-typing.png, the screenshot that entry cites, is boxes; what the agent read was the iris text render: chars=5 glyphs=5 log line, which reports what parley shaped, not what reached the screen. The earlier genuinely-good emulator shots (/tmp/after3.png, 2026-09-05 19:56) predate the APK being built with force-gles at all, so they were the Vulkan path. - The phone build is not affected and does not need withdrawing. android-app/build-apk.sh's default features are deliberately without force-gles (d73db97's comment), so a phone build takes Backends::PRIMARY -> Vulkan, where a one-layer array is an ordinary one-layer array and glyphs draw correctly -- which is also what Iris's phone reports have shown all along. The fix matters for any device that falls back to GLES, which is why it is not just an emulator convenience. - Appearance testing on Android is back. build-apk.sh debug --abi x86_64 --features "transcript-screen bench force-gles" on this checkout's AVD draws the transcript legibly -- docs/bench/p1a-2026-09-06/iris-emulator-gles-fixed.png -- so P1b onwards can be checked here rather than only on the desktop backend or Iris's phone. 2026-09-06: the move slot has one owner. IRIS_TODO's open "a wrapped transcript row trips reposition's debug assert" is fixed rather than suppressed. mov accumulates a delta on a widget's move slot and 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) -- measured with a probe on the .wrap(true) repro: h=1604.7 height=548.7, the row's own draw having updated active.size without the list's height cache. The slot now means move_applied + repositioned (both on ActiveData), so reposition adds the move instead of dropping it and stays idempotent, and the old assert is replaced by a debug_assert_eq! that nothing but those two ever wrote the slot. Test: a_widget_moved_by_its_parent_and_then_placed_inside_it_lands_at_the_placement (layout_tests.rs), which draws the child at the offered position rather than the placement without the fix. Checks: cargo fmt --all --check clean, cargo clippy --workspace --all-targets warning-free, cargo test --workspace 86 (iris, +1) + 13 (iris-core) + 31 (transcript-ui), the .wrap(true) repro drawing correctly, and an emulator bench run with assertions live and no abort (2438 frames over 147.7s, p50 27.2ms).

  - [x] **P1b — tool-call cards and grouping.** Done 2026-09-06.
        `ToolRows.kt`/`ToolInput.kt` ported to
        `iris/transcript-ui/src/tool.rs` plus two new pure modules in
        `client-core`. **Screenshots:
        `docs/bench/p1b-2026-09-06/iris-tools-collapsed.png` and
        `iris-tools-expanded.png`**, both from
        `iris/run-headless.sh transcript -- -p transcript-ui` on the
        desktop/winit backend (the emulator was not touched this pass
        -- another agent held this checkout's AVD). The expanded one
        is taken with `IRIS_TOOLS_EXPANDED=1`, which the example reads
        to call `TranscriptScreen::expand_tail_tools` -- the expanded
        appearance is otherwise unreachable on a machine with no
        display and no finger.
        **What the cards look like, against `ToolRows.kt`:**
        - *A collapsed card* -- a mark, the tool's name (14pt), the
          one-line summary `parse_tool_input` derives (12pt, Subtext
          0, one line, clipped), and the state word at the far right.
          Same as Compose, except that Compose ellipsises the summary
          and iris clips it: there is no overflow-ellipsis in
          `TextAttrs` yet (IRIS_TODO).
        - *An open card* -- the timeout at the top right, the tool's
          own description, the subject in a `Verbatim` panel with
          `client_core::highlight`'s spans, the leftover input fields
          under it, then the output. Same order as Compose.
        - *A group* -- "Called N tools" (Compose's exact wording, and
          so the name a `ui-trace` script taps), the cards on a Mantle
          surface, and a chevron bar at the foot that closes it from
          the end the reader is looking at.
        - *States* -- `client_core::transcript_fold::ToolState`, five
          of them, each with its own word and colour: nothing for
          `Succeeded`, "running" (Subtext 0), "your turn" (Peach, the
          Compose card's own wording and colour), "failed" (Red) and
          **"no result" (Yellow)**. The last two are new -- Compose
          can say neither.
        **Two things the port had to add to be able to say "it
        broke".** `event_model::Event::ToolEnd` gained `is_error`
        (`#[serde(default)]`), read from the CLI's own `tool_result`
        by one function used by both the live translator and the
        import replay (`import::tool_result_is_error`); without it a
        result was all a card had and a failed call drew exactly as
        confidently as one that worked. And `ToolState` separates
        `Succeeded`-with-empty-output from `NoResult`: both leave the
        same empty string, and only the session's own status tells
        them apart, which is why `TranscriptScreen::
        set_session_working` exists and why only the *newest* row can
        be "running" (every row behind it belongs to a turn that has
        ended).
        **Pass condition, met**:
        `collapsed_cards_shape_only_their_summary_lines`
        (`transcript-ui/src/lib.rs`) opens a group of three cards
        whose calls carry 88 kB of output each and asserts the
        text-shape count equals the same group's over three bytes.
        **17 either way.** Confirmed to be a real test, not a
        tautology, by pushing the output block into the collapsed
        branch: **17 against 20**.
        `a_result_arriving_redraws_one_card_whatever_the_run_holds` is
        the second: one `ToolEnd` costs the same number of
        `Widget::draw` calls in a twelve-call run as in a three-call
        one.
        **Three defects found on the way, all by looking at the
        render rather than at the diff:**
        1. **A `Span` of `Pad`ded children inside another `Span`
           places those children a slot out of step.** Every card drew
           its content one card's height below its own box, so the
           group read as empty bars with somebody else's summary in
           them. Bisected against
           `IRIS_TOOLS_EXPANDED=1 iris/run-headless.sh transcript`:
           removing the inner `Span` fixes it, and so does removing
           the cards' own `Pad`; the card background, the `Sized`
           wrappers and the per-card `WidgetPtr` all make no
           difference. Worked around by building the group as **one**
           `Span` (header, cards, collapse bar), which costs the 4dp
           inset Compose holds its cards off the group's edge by. The
           framework defect is still open -- IRIS_TODO has it, and it
           is not the `mov`/`reposition` one f5b8893 fixed (it
           survives that commit).
        2. **`scrollable_on(Axis::X)` on a non-editable `Text` draws
           nothing at all** -- an empty panel where the command should
           be. A markdown fence does the same thing to a `TextEdit`
           and is fine. So a card's verbatim block is `masked()` and
           clips rather than panning; when this is fixed the pan
           belongs there too, because the long command is the one
           being read closely.
        3. **`NotoSans-Regular.ttf` has no U+25B8/25BE/25B4** (read
           out of the bundled `cmap`s) while `NotoSansMono-Regular`
           does, so the expander mark is set in the monospace face at
           the one place the character is written. The old
           `build_tools` summary drew that codepoint in the sans face,
           which was a missing glyph nobody had looked closely enough
           to see.
        **Checks**: `cargo fmt --all --check` clean in both
        workspaces; `cargo clippy -p iris -p iris-core -p
        transcript-ui -p desktop-app -p tabs-ui --all-targets` and
        `cargo clippy --all-targets` in `client-core`/`server`/
        `event-model` warning-free; tests 86 (iris) + 13 (iris-core) +
        36 (transcript-ui, +5) + 137 (client-core, +11) + 160
        (server, +1).
        **Not done**: nothing on the emulator or the phone (the AVD
        was another agent's this pass, so no frame times were taken);
        a card's text is not selectable, unlike Compose's, since
        `Selection` is keyed per markdown block and a card has none
        (IRIS_TODO); no per-corner radius, so the "connected stack"
        shape `connectedShape` draws is a 2dp gap instead;
        `AskUserQuestionBody`/`PermissionAsk`'s answer buttons are not
        ported -- an unanswered ask forces its card open and says
        "your turn", but there is nothing to press yet, which is P1d's
        modal/controls work.
  - [ ] **P1c — history paging and jump-to-latest.** Wire
        `client-core::transcript_source` into `transcript-ui`:
        the opening page, paging back on scroll with the cushion
        measured in on-screen viewports (`HISTORY_SCREENS`, IRIS_TODO
        "Build (for the port)"), the `NothingLoaded`/empty/error
        states drawn distinctly (UI_RULES: design the unknown state
        first), `join_pages` at each seam, and a jump-to-latest
        control that pins to the newest end. Pass condition: the
        P1 pass condition below, against `ui-sandbox.sh` with
        `AI_SANDBOX_BIG_MB` and `--delay`.
  - [ ] **P1d — images, the session settings dialog, attachments,
        usage bar.** `SessionImage` thumbnails (the scaled image
        widget), the modal primitive and `SessionSettingsDialog`/
        `UsageDialog`, `PendingAttachments` over the attachments
        route (`api.rs` gap), `SessionUsageBar` (the gauge widget).
  - [ ] **P1e — keyboard and insets behaviours** from AGENTS.md's
        "Things that have bitten", re-verified on the phone build:
        composer never left floating after the keyboard closes
        mid-stream, `adjustResize` + edge-to-edge together, one
        recomposition-equivalent per keyboard toggle (the
        `iris insets:` log line count).
  
  History paging backward (with the
  page-boundary healing `client-core` does not have yet, below),
  `TranscriptSource`-backed cache/server stitching, jump-to-latest,
  tool-call cards and grouping, the session settings dialog, composer
  attachments, and the keyboard/insets behaviours AGENTS.md's "Things
  that have bitten" names (the floating-composer bug, `adjustResize`,
  the `imePadding`-vs-raw-inset rule). This is the highest-risk step:
  it is the screen the app is used for, every hour of the day.
  
  **Kotlin it replaces**: `SessionScreen.kt`, `TranscriptList.kt`,
  `SessionSettingsDialog.kt`, `ToolInput.kt`, `ToolRows.kt`,
  `AskQuestion.kt`, `Compaction.kt`, `SessionUsageBar.kt`,
  `PendingAttachments.kt`, `Attachment.kt`, `Attachments.kt`,
  `SessionImage.kt`, `MemoryNote.kt`, `PeerMessage.kt`, `RawBlock.kt`,
  `CodeFence.kt`, `MarkdownLinks.kt`, `MarkdownPieces.kt`,
  `Markdown.kt`, `Bubble.kt`, `ScrollAnchor.kt`, `Drafts.kt`,
  `UsageDialog.kt`, `Chevron.kt`, `Dividers.kt`. (`transcript-ui`
  already covers the row/markdown/selection/composer core these sit
  on top of or beside.)
  
  **`client-core` needed, and what is not yet covered and must be
  ported first** (`CLIENT_CORE.md`): `TranscriptSource.kt` (deciding
  cache vs. server per page and stitching them — "not started"),
  `TranscriptItems.kt`'s `joinPages`/`healSplitMessage`/`adoptRun`
  (page-boundary healing — "not ported," and paging backward is
  exactly what exercises it), the markdown *block* model beyond
  syntax spans (headings/lists/tables/fences as distinct nodes —
  "not started," needed for `CodeFence`/`MarkdownPieces`' equivalents),
  and the attachments route (`/sessions/{id}/attachments` — "not
  covered" in `api.rs`, needed for `PendingAttachments`/`Attachment`).
  
  **iris widgets missing, → `IRIS_TODO.md`'s new "Build (for the
  port)" section**: row-level accessibility names and the tappable
  link / background-chip primitive (both already listed under I5's
  leftovers — this step is what needs them, not a new ask); a
  history-paging cushion measured in on-screen viewports rather than
  a row count (the `HISTORY_SCREENS` lesson in "Things that have
  bitten," which iris's `List` has no equivalent of yet); a scaled
  thumbnail/image widget for `SessionImage`'s in-transcript images; a
  modal/dialog primitive for the session settings dialog and
  `UsageDialog` (iris has none today — check before building a second
  one for P3/P5); a horizontal gauge/bar widget for
  `SessionUsageBar`.
  
  **Pass condition**: `app/ui-sandbox.sh`'s fixtures driven by
  `ui-trace record --do "tap '<label>'"` — a session with the big
  transcript (`AI_SANDBOX_BIG_MB`), a paused/slow-spawning one
  (`AI_SANDBOX_SPAWN_DELAY`), and `--delay` on the server — exercising
  the four states UI_RULES.md says to design first: unknown (a page
  that hasn't loaded), empty (a session with no messages yet), error
  (a failed send/interrupt), and too-long (the big transcript,
  paged). Re-take the I5 `FrameReport` (`iris frame report` in
  logcat, same as I5's box) once this screen has real paging and
  compare it against I5's own numbers, not against Compose's — the
  three measurement sources still are not comparable per
  `DECISIONS.md`'s DEFERRED item.
  

• P2 — the shell merge and a real phone install. Merge this screen's cdylib into the E3/E5 shell (android-shell + app/shellApp) behind the same feature-flag pattern I5 used to extend iris-android-app (DECISIONS.md, 2026-09-05), so there is one app — notification service, share target and the real screen — rather than a demo shell and a service shell side by side. Package with cargo xtask apk (E5) and get it onto the real GrapheneOS phone, not just the emulator: arm64-v8a is the ABI that matters there (the emulator here is x86_64), and the this-machine-android skill's facts apply for the first time in this port — no System Tracing on that phone (frame numbers have to come from FrameReport itself), the local-network permission is required there even though AOSP's docs say VPN traffic is excluded, and ui-trace/adb target this checkout's own emulator by default so a real-device command needs -s <serial> explicitly.

  **Kotlin it replaces**: nothing further than E3 already did
  (`Notifications.kt` → `notifications.rs`, `Share.kt` → `share.rs`,
  `ServerConfig.kt`'s Keystore half → JNI calls into `wg-app-link`) —
  this step is wiring P1's screen in as the shell's real content
  instead of E3's placeholder, plus getting a signed APK onto a
  physical device for the first time in this port.
  
  **`client-core` needed**: none new; E3 already covers what the
  shell itself needs. Attachments (P1's gap) matter here too if a
  real photo share is exercised.
  
  **iris widgets missing**: none — this step is integration, not new
  widgets.
  
  **Pass condition**: `cargo xtask apk`, install on the real phone
  over adb, enroll via the deep link, background the app and get a
  real notification, share a text snippet into a session, and
  confirm `ui-trace` can still find controls by name on real
  hardware (accessibility names are not guaranteed to survive a real
  device's TalkBack/AccessKit wiring the way they do in the
  emulator — this is the first time that gets checked for real).
  

• P3 — root tabs. Screen/MainTab in app-ui: the sessions list, import, models and setups tabs, plus spawn and the app's one level of back-stack navigation (AppRoot.kt's when).

  **Kotlin it replaces**: `AppRoot.kt`, `MainScreen.kt`,
  `SessionListScreen.kt`, `ImportScreen.kt`, `ModelsScreen.kt`,
  `SetupsScreen.kt`, `SpawnScreen.kt`, `BusyItem.kt`,
  `UniqueItems.kt`, `SessionAlerts.kt`.
  
  **`client-core` needed, not yet covered**: setups/machine/provider
  discovery, the models routes (`/models*`, HuggingFace browsing and
  downloads), and importing (`/setups/{id}/importable*`) — all three
  listed "not covered" in `api.rs`'s table and none started; each is
  real work, not a stub, per `CLIENT_CORE.md`'s own caveat.
  
  **iris widgets missing**: a `BusyItem` equivalent — a row dimmed,
  drained of colour, labelled with the operation in progress, that
  does **not** block the list's own scroll/drag the way an overlay
  did on the Compose side (AGENTS.md's "Shared appearance"); a
  `uniqueItems` equivalent is logic, not a widget, and ports directly
  into `app-ui` itself; a confirmation dialog with a toggle switch,
  for the delete-with-`deleteForeign` flow, needs the same modal
  primitive P1 flagged — build it once, here or in P1, whichever
  lands first.
  
  **Pass condition**: `ui-trace` tap-by-name on all four tabs against
  `ui-sandbox.sh`'s fixtures; the two-copies-of-one-session-id
  fixture (AGENTS.md's "Importing") does not crash the list — this is
  the regression `uniqueItems` exists for and it must be exercised
  here, not assumed; the delete dialog's paragraph reads correctly
  both with and without `deleteForeign` toggled (its own text, not
  appended, per AGENTS.md).
  

• P4 — file explorer. The viewer, the editor with its EDIT_LIMIT, and the 409 conflict.

  **Kotlin it replaces**: `FilesScreen.kt`, `FileViewer.kt`,
  `FileEditor.kt`, `FileLines.kt`.
  
  **`client-core` needed, not yet covered**: `/setups/{id}/dir|file`
  — not in `api.rs`'s covered list, real work, port first.
  
  **iris widgets missing**: nothing beyond what P1 needs (a
  virtualised line-numbered text view is `iris::widget::List` reused,
  per I3's box) — the open question is whether the editor's
  `BasicTextField`-equivalent cost (`docs/EXPLORER.md`'s "what the
  measurements said") reproduces in iris's `TextEdit` at the same
  `EDIT_LIMIT`, which this step has to re-measure rather than assume.
  
  **Pass condition**: `app/ui-sandbox.sh`'s `~/files` fixture tree
  (empty dir, tab/apostrophe names, binary, over `FILE_LIMIT`,
  `chmod 000`, symlinks good and broken, one source file per
  language, `edit-32k.rs`/`edit-128k.rs`/`big-source.rs`) driven by
  name; the 409 reproduced by editing the file on the machine between
  opening it and saving, per AGENTS.md's own recipe.
  

• P5 — settings, enrollment, notifications permission.

  **Kotlin it replaces**: `SettingsScreen.kt`, `ServerConfig.kt`'s
  remaining non-Keystore parts, `DebugStats.kt`, `FrameStats.kt`,
  `CrashLog.kt`. The QR scanner (`EnrollmentScanActivity`, in
  `wg-app-link`) is platform-only and is **not** replaced — it stays
  a Java/Kotlin activity per decision 1 above, called into from
  `app-ui` the way it is called into from Compose today.
  
  **`client-core` needed**: none new — `config.rs`'s
  `EnrolledServer`/`parse_link` already cover the deep-link half; the
  Keystore half stays the JNI call E3 already wired.
  
  **iris widgets missing**: none identified yet — a plain form screen.
  
  **Pass condition**: enroll via the same `aiappshell://enroll?...`
  link `ui-sandbox.sh`'s banner prints; the local-network-not-allowed
  banner (AGENTS.md's standing-condition text) reads by name when the
  permission is off; `POST_NOTIFICATIONS` request flow checked on the
  real phone from P2, not just the emulator.
  

• P6 — desktop parity. Root tabs, explorer and settings on iris/desktop-app, matching P3–P5 there. Not a Kotlin replacement (the desktop app has no Compose original) — this is closing the gap E4 deliberately left (session list + transcript only).

  **`client-core` needed**: the same P3/P4 gaps, once closed there.
  
  **Pass condition**: `run-headless.sh` screenshots of each tab and
  the explorer against `app/ui-sandbox.sh`, the same way E4's did.
  

• P7 — the switch of ai-app's main. Point ai-app's production Android build at app-ui/android-shell instead of app/androidApp; decide then whether app/androidApp stays as a reference or is retired — a load-bearing decision (AGENTS.md's "ask before changing load-bearing decisions") to bring to Iris rather than make here.

  **Pass condition**: the full set of pass conditions above, re-run
  once more against a real `ai-server` (not the sandbox) on a real
  phone, side by side with the Compose build until it holds.
  

For the next agent

What to do when you pick this up, in order, so nothing here has to be re-derived:

1. Read this file, then AGENTS.md and PLAN.md. The rules there (measure, do not read; fix the rig before accepting its limits; the emulator is this checkout's own) all apply.
2. Work on the rustify branch of this clone (ai-app-2), not on main and not in ai-app. Nothing on this branch is production until Iris says so. Commit and push as you go.
3. The E- and I-steps (the framework decision) are done — iris won, DECISIONS.md 2026-09-05. Take the next unchecked P-box in "## The port, in order (decided 2026-09-05)"; P1 — session screen parity — is next.
4. Every step ends with its measurement written into this file beside the box, and the box ticked or the reason it could not be written in its place. A step that is blocked says by what, not "later". Write it as you go rather than at the end — see "Keep this file current as you work".
5. Run the existing rigs rather than inventing new ones: ui-sandbox.sh for a server with fixtures, transcript-bench.sh for the scroll baseline, ui-trace for anything positional, emu up for the emulator, iris/run-headless.sh EXAMPLE --shot PNG for an iris example on this displayless machine, and rigs/gpu-probe to ask a device (this VM, the emulator, or a real phone over adb push) what wgpu features and limits it actually has before building anything on the assumption it does. The Vulkan section below says how to get a Vulkan path in the emulator when a wgpu backend needs one.
6. Bound anything heavy at the moment you start it. An emulator or a long build gets a deadline — timeout, or a watchdog scoped to the pid you just started — rather than a plan to stop it later. Scope it to that pid: a watchdog written as sleep N; emu down fired into a later experiment here and made a working Vulkan build look like a crash. And stop the emulator when the work needing it is done rather than between tasks.
7. Decisions belong here with a date and what was rejected, the way PLAN.md does it. Do not put design into commit messages alone.

cargo test --workspace was taking half an hour, and it was debug info (2026-09-08)

Iris asked why. It is worth reading before concluding that this machine needs more disk, because that was the first guess here too and it is the smaller half.

What it was. rustc's default for dev is debug = true, and this workspace has eight test binaries, each statically linking the whole wgpu + naga + winit + parley graph. Every one of them therefore gets its own full copy of that graph's DWARF written into it at link time. Measured: the linkers for one cargo test --workspace had written ~54 GB between them and were still going at 30 minutes, the worst single one 16.9 GB for one test binary, leaving an 88 GB target/.

What it was not: CPU. The machine was 87% idle with 8-12% iowait throughout, and every rustc was in __futex_wait with no busy thread. An earlier guess in this same session -- that concurrent agents' builds were loading the machine -- was wrong, and checking ps for a busy process would have killed it in a minute. rust-lld's worker threads were in D state in btrfs handle_reserve_ticket: blocked on space reservation, because the filesystem was at 83% full. So the write volume is the disease and the full disk is what turned slow into stalled -- they move about 20 MB/s between them in that state.

How to tell it apart from a busy build, since the two look identical from the outside (cargo printing Compiling ... and not finishing):

ps -eo pid,stat,etime,pcpu,comm | grep -E 'rustc|rust-lld'
cat /proc/<lld-pid>/wchan          # handle_reserve_ticket = btrfs, not you
for l in $(pgrep -x rust-lld); do awk '/^write_bytes/{print $2/1048576}' /proc/$l/io; done

A linker at 0.5% CPU having written gigabytes is not compiling.

The fix, in iris/Cargo.toml. debug = "line-tables-only" on both profile.dev and profile.test -- both, because cargo test builds dependencies under dev and the test targets themselves under test, so setting only dev leaves the eight big binaries at the default. It keeps the file and line of every frame, which is what a panicking test prints and what gdb needs to name the frames of a segfault (both were used today); it gives up inspecting variables in a debugger, which is worth asking for per-run when it is wanted:

RUSTFLAGS="-C debuginfo=2" cargo test -p iris --test mask_sdf

Measured after, cold (cargo clean first), in iris/: the whole workspace compiles and runs all 19 test suites in 69 s, with a 3.7 GB target/. Before, an incremental run — everything already compiled, only the eight binaries left to link — had not finished after thirty minutes. cargo clean alongside the profile change freed 104 GB (83% -> 67% full).

One trap while measuring this, worth not repeating: the session's working directory is the repo root, not iris/, so a du -sh target appended to a build command reports the root workspace's 29 MB and makes a correct iris build look like it built the wrong thing. Read the test names in the log to tell two workspaces apart, not a path-relative size.

Venus went away for an hour, and nothing said so (2026-09-08)

It came back on its own and nothing in the host config changed. Iris asked whether something had, since she still passes Venus as true. What happened, and what was ruled out, so the next occurrence is not re-investigated from scratch:

The symptom, around 02:10 on 2026-09-08 with the VM at load 68 and several agents building: vulkaninfo reported Failed to detect any valid GPUs in the current config and vkEnumeratePhysicalDevices failed with ERROR_INITIALIZATION_FAILED; Mesa printed No virgl contexts available on host; and wgpu reported NotFound { active_backends: VULKAN, no_adapter_backends: VULKAN, supported_backends: VULKAN | GL }. So both paths through the virtio-gpu died at once -- Venus for Vulkan and virgl for GL -- and GL then fell through to llvmpipe, which is what actually rendered that hour's layer-2 screenshots.

That string is Mesa's virgl DRM winsys, next to DRM_IOCTL_VIRTGPU_ CONTEXT_INIT failed with %s in libgallium: the host refused a new context. After a reboot the same host config gives Virtio-GPU Venus (AMD Radeon RX 7900 XT (RADV NAVI31)), Mesa 26.2.2, driverID MESA_VENUS.

What was ruled out, by measurement rather than by reasoning:

• A guest-side context cap. 48 concurrent short-lived Vulkan clients all succeed, and 120 concurrent long-lived ones (each holding a VkDevice open at once, a throwaway holder) all succeed. So the ceiling, if there is one, is not near the handful of GPU-using processes that were running.
• A Mesa upgrade. mesa 1:26.1.7 -> 1:26.2.2 landed 2026-09-05, two days before Venus was last seen working here.
• Anything in iris. It was vulkaninfo's answer too, from a process that has never linked against this repo.

What this host's virtio-gpu actually offers, asked of the kernel rather than assumed (rigs/virtgpu-probe): bitmask 0x16 -- VIRGL, VIRGL2 and VENUS. Capset 6, the DRM "native context", is not offered. That is the answer to "is there something to do with qemu instead of Venus", asked by Iris 2026-09-08: native context is the thing worth wanting -- RADV running in the guest against a passed-through DRM context instead of Venus proxying every Vulkan call, and it is where Mesa's effort has gone. Whether it would avoid the teardown crash below is untested -- it is a different driver stack, so it is a reasonable thing to try rather than a known fix -- but it needs the host side to offer it (virglrenderer built with its amdgpu DRM renderer, and a qemu that exposes context_types=drm; crosvm has it further along). The guest would also need vulkan-radeon installed, which it does not have today -- only vulkan-virtio. The other two options are VFIO passthrough (complete, but the host loses the GPU) and dropping venus=true (leaves virgl/GL only, i.e. no Vulkan at all, which is the wrong direction since Vulkan is the phone's path).

So it is host-side and transient, and this VM cannot see the host to say more: no dmesg (the guest's kernel buffer is not readable to this user), and no view of the host's amdgpu. The honest state is "we do not know which host-side resource ran out" -- worth capturing the host side of it if it recurs, since that is the half that would answer it.

A second, unrelated Venus fault, found the moment it came back (2026-09-08). iris/tests/mask_sdf.rs had passed the day before and now SIGSEGVd -- and it had passed because Venus was down, so it silently ran on GL. What it actually is, narrowed by measurement:

• The test's work completes and its answer is right (worst CPU/shader disagreement 5.8e-6). The crash is at process teardown, dropping wgpu's device: a call through an unmapped address on a wgpu-created thread, per gdb.
• It is wgpu's teardown, not Venus's device lifecycle. A plain Vulkan program creating and destroying five VkDevices and its instance on the same adapter exits cleanly (rigs/virtgpu-probe).
• It is Venus-specific. The same test binary, with Vulkan hidden (VK_DRIVER_FILES=/nonexistent) so wgpu falls back to GL, exits cleanly.

Worked around in the test rather than fixed, at Gpu::leak with the reason written there: one device for the whole test, handed to the process instead of dropped. Compute was investigated and is not involved -- an early version of that test used a compute pass, which was wrong for its own reason (the paragraph after this one), but the render-pass rewrite crashes identically, and Venus here reports full compute anyway (maxComputeWorkGroupInvocations 1024, maxComputeSharedMemorySize 65536, Vulkan 1.4 -- rigs/virtgpu-probe again). The 2026-09-05 "no compute" finding is about the Android emulator's SwiftShader GL path reporting ES 3.0, which is a different machine; it says nothing about this VM. This was got wrong out loud first, so it is written down: the compute pass was blamed for the crash before the rewrite showed the crash was not about compute at all.

The test was rewritten to a render pass regardless, and that part is not a workaround: it now asks for iris_core::device_limits() -- what iris itself requests -- and calls the function from the fragment stage, which is where the renderer calls it. A test that needs a capability the thing under test has never needed is testing the wrong device.

What was fixed, because the failure was silent. Two things, both the rule that a degraded state must be distinguishable from a healthy one:

1. default::render::UiRenderer::new had the defect the Android backend was fixed for in 85869d0 -- Backends::PRIMARY and an .expect -- so layer 2 aborted with Could not get adapter! instead of falling back. It now probes and rebuilds on Backends::GL, as Android does.

2. The desktop had no logger at all, so every log:: call on that side -- including that new fallback warning -- went to log's no-op default. DefaultApp::run installs a stderr logger now (src/default/logging.rs, no new dependency), and the renderer says which adapter won at info:

   INFO iris::default::render: iris renderer: Virtio-GPU Venus
   (AMD Radeon RX 7900 XT (RADV NAVI31)) (Vulkan, venus Mesa
   26.2.2-arch1.1) on Backends(VULKAN | METAL | DX12 | BROWSER_WEBGPU)
   

   That line is the point: with the fallback in place and no logger, a run-headless.sh screenshot rendered by llvmpipe and one rendered by the host's GPU are the same PNG, and the difference is exactly what a screenshot is being taken to judge. Check it before trusting a layer-2 screenshot or any frame number from that window.

Vulkan in the emulator (measured 2026-09-04)

Settled 2026-09-04: the guest gets Vulkan from SwiftShader, and the missing step was a cold boot. -feature Vulkan plus VK_DRIVER_FILES=$HOME/Android/Sdk/emulator/lib64/vulkan/vk_swiftshader_icd.json gets the host side to select SwiftShader, but the guest keeps reporting zero devices until -no-snapshot-load is added, because it boots from a snapshot saved under the previous GPU config — -no-snapshot-save is worth adding too, so the Vulkan-configured snapshot does not then break the next ordinary boot. With that, cmd gpu vkjson reports SwiftShader Subzero and wgpu takes its Vulkan path (E1). EMU_GPU=software in emulator-tools gets the same guest Vulkan with no GPU use at all, for work where the emulator's frame rate is not what is being measured.

The rest of this section stands as the record of why host Vulkan is not available. A wgpu app in this emulator was going to get GLES only, because host Vulkan is switched off in emulator-tools. Retried on Mesa 26.1.7: Venus still fails the same way — gfxstream picks externalMemoryMode: OpaqueFd, probes VK_FORMAT_R8G8B8A8_UNORM for an exportable colour buffer, and Venus says the format is unsupported (Failed to find memory type for ColorBuffers, fatal before adb sees the device). Venus does advertise VK_KHR_external_memory_fd and VK_EXT_external_memory_dma_buf, so the gap is specifically opaque-fd image export. gfxstream has a string-valued VulkanExternalMemoryMode setting ("overrides what would otherwise be determined automatically"), but -feature Name=Value is rejected as a bad feature name, and the only mode words compiled into this emulator's libgfxstream_backend.so (37.1.11) are OpaqueFd, Metal and none — there is no dma-buf mode in this build to switch to. So Venus is blocked by the emulator, not by Mesa; retry when the emulator package updates, since upstream gfxstream does have dma-buf external memory.

What does work: pointing the emulator's Vulkan loader at the software ICDs the emulator ships itself, with the feature enabled:

VK_DRIVER_FILES=$HOME/Android/Sdk/emulator/lib64/vulkan/vk_swiftshader_icd.json \
GPU_HOST_FEATURES="-feature Vulkan" emu up

The guest then reports Vulkan 1.3 (cmd gpu vkjson, SwiftShader Subzero) while GLES still runs on the real GPU through virgl — so a Vello/wgpu app can take its real Vulkan path here, with compute shaders, CPU-rasterised. That is enough to test correctness of the Vulkan path in the emulator; GPU performance of it is a phone measurement either way, exactly as MACHINE.md already says about frame times. lavapipe (lvp_icd.json, the other ICD the emulator ships) selected llvmpipe and booted, then the emulator died right after loading the default_boot snapshot with nothing in the log; a snapshot saved under a different Vulkan device is the suspect, and -no-snapshot-load is the untested next step. SwiftShader is the one that works today. EMU_GPU=software is now in emulator-tools (agreed with the ai-app session and with Iris, default unchanged, since -gpu host was measured and the Compose scroll benchmarks depend on it). The cold-boot flags are not a knob there: that wants snapshot invalidation as well, which is a bigger design question in shared tooling.

Things a Rust app changes elsewhere

• wg-app-link's :link (pinned TLS, enrollment store, QR activity) is Kotlin shared with Dev Updater. The certificate code already exists on the Rust side of the submodule; the pinned-CA build step (generatePinnedCert) becomes a build.rs reading the same path. The QR scanner stays a Kotlin activity, since the camera is a platform feature.
• Tooling becomes cargo for everything but packaging: cargo test, clippy, fmt cover the whole client, which is the motivation. Gradle remains for the APK, signing (~/.config/ai-app/release.jks) and Dev Updater's build modes; build-apk.sh would call cargo ndk first.
• The bench scripts (ui-trace by accessibility label) keep working only if the framework exposes names through AccessKit on Android; that is part of E2's pass condition, not a nicety.
• Icons stay Nerd Font glyphs from the committed subset; Parley/Fontique loads a font file directly, so build-icon-font.sh is unchanged.

Sources

• iced: repo, 0.14 release, Android thread, markdown selection request
• Linebender: 2026 Q1 report, xilem, parley, vello, vello_hybrid
• android-view: repo; android-activity PR #214
• winit Android IME: #1823, #2766, #2305
• egui on Android: discussion #2053
• Slint: Android guide, 1.15 release, licensing, rich text #1325, markdown #6684
• Makepad: repo, makepad-widgets, Robrix, Robrix releases
• AccessKit: releases
• Build tools: cargo-ndk, cargo-apk, rust-mobile
• uniffi: repo, KMP bindings fork
• GPUI mobile: gpui-mobile
• The earlier Dioxus spike's findings on wgpu/Vulkan in this emulator: ~/repos/tdep-survey/app-dioxus/README.md