client-core: an unasked page is not an empty one, and two guarded invariants

Review of 73251d6's port of TranscriptSource/joinPages.

`TranscriptSource::page` answered `before == 0` with an empty `Vec`, which
is the same value it answers "this conversation has no more history" with.
That is the state the Kotlin keeps apart: `loadOlderPage` returns false at
`oldestSeq == 0` *without* touching `moreHistory`, and returns false on an
empty page *by latching it*. Collapsing the two moved AGENTS.md's paging
bug one layer down rather than fixing it. `page` returns `OlderPage` now --
`Events(vec![])` is the start of the conversation, `NothingLoaded` is not
an answer about the conversation at all.

`join_pages`' `debug_assert!` on seq ordering across the boundary is not a
true invariant: a peer note carries the seq its turn began at, which can be
older than the page it arrived in, so an ordinary transcript would have
panicked a debug build there. Replaced with the one the function exists to
enforce -- no tool id surviving in both halves.

`fetch_transcript_lines` stores `RawValue`'s exact server bytes, so the
"neither source can produce a newline" comment in `SessionCache::append`
now rests on the server's serializer staying compact rather than on a
local normalization. Checked with a `debug_assert!` in `append` and
`store_page` rather than trusted.

Tests for the failure half, which the port had none of: a 500 mid-page, a
cached line this build cannot read, and the `after` bound in the case that
actually carries one (the existing test asserted only the case with no
bound). `cargo fmt`, `cargo clippy --all-targets`, `cargo test` (112) clean
in client-core; `cargo check -p desktop-app` clean.

client-core/Cargo.toml

@@ -17,7 +17,12 @@ edition = "2024"
 [dependencies]
 event-model = { path = "../event-model" }
 serde = { version = "1", features = ["derive"] }
-serde_json = { version = "1", features = ["float_roundtrip"] }
+# "raw_value" is `fetch_transcript_lines`'s reason -- it needs the exact
+# bytes the server sent, not this crate's own re-serialization of a parsed
+# `Value`, so a cached line and a live SSE frame for the same event agree
+# byte-for-byte (see that method's doc). "float_roundtrip" is why they
+# agree on a `ts` at all -- see server/Cargo.toml's identical comment.
+serde_json = { version = "1", features = ["float_roundtrip", "raw_value"] }
 # The blocking HTTP client for the REST calls and the long-lived SSE GETs.
 # `server/` already depends on ureq for its own outbound HTTPS (the usage
 # poll in usage.rs) and it is rustls-backed like the rest of this project's

client-core/src/api.rs

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

client-core/src/lib.rs

@@ -11,5 +11,6 @@ pub mod notifications;
 pub mod sse;
 pub mod transcript_cache;
 pub mod transcript_fold;
+pub mod transcript_source;
 
 pub use event_model::*;

client-core/src/transcript_cache.rs

@@ -361,6 +361,10 @@ impl SessionCache {
             {
                 return Ok(false);
             }
+            debug_assert!(
+                lines.iter().all(|l| !l.contains('\n')),
+                "a stored page's lines must each be one line"
+            );
             fs::create_dir_all(&this.dir)?;
             let kind = if rows { "rows" } else { "raw" };
             let mut content = lines.join("\n");
@@ -389,8 +393,16 @@ impl SessionCache {
                 return Ok(());
             };
             // Written as it arrived. A newline inside it would split one
-            // event into two unreadable halves, but neither source can
+            // event into two unreadable halves. No source here can produce
-            // produce one.
+            // one -- an SSE `data:` field cannot hold a raw newline, and a
+            // fetched line is one element of a compact JSON array -- but
+            // that is a fact about the *server's* serializer rather than
+            // anything this file controls, so it is checked rather than
+            // trusted.
+            debug_assert!(
+                !line.contains('\n'),
+                "a cached transcript line must be one line: {line}"
+            );
             use std::io::Write;
             writer.write_all(line.as_bytes())?;
             writer.write_all(b"\n")?;

client-core/src/transcript_fold.rs

@@ -294,6 +294,199 @@ fn split_run(tail: &[TranscriptItem], behind: Option<&str>) -> Vec<TranscriptIte
     out
 }
 
+/// Puts a page of older items in front of the ones already loaded, healing
+/// whatever the page boundary cut in two. Ported from `TranscriptItems.kt`'s
+/// `joinPages`.
+///
+/// Two things straddle a boundary: a tool call separated from its result,
+/// and a message separated from the rest of itself. Both were one thing
+/// before the transcript was cut into pages.
+///
+/// A boundary lands wherever it lands, and roughly half the time that is
+/// between a call and its result. The newer page then holds a `ToolEnd`
+/// whose start it never saw, which `fold_event` draws as a row of its own
+/// -- correctly, because a call that renders as nothing is indistinguishable
+/// from one that never happened. When the older page arrives it brings the
+/// real `ToolStart`, and concatenating the two lists left *both*: the same
+/// call twice.
+///
+/// Merged by the call's own id rather than by position, because position is
+/// exactly what a page boundary destroys. The older row wins on what a
+/// start knows and the newer on what an end knows, which is the only way
+/// round that loses nothing.
+///
+/// The third thing is the *run*, and it is the one the Kotlin original used
+/// to miss (AGENTS.md's "things that have bitten"): every page ends up
+/// here, but `adopt_run` must run on *every* join, not only the one where a
+/// split call was found -- a boundary landing cleanly between two finished
+/// calls, which is most of them, would otherwise leave the older page's
+/// calls under the run name they were folded with. On screen: one run of
+/// tool calls drawn as two groups, with the seam wherever the reader
+/// happened to have paged.
+pub fn join_pages(earlier: &[TranscriptItem], later: &[TranscriptItem]) -> Vec<TranscriptItem> {
+    let (older, newer) = heal_split_message(earlier, later);
+    let started_earlier: std::collections::HashSet<&str> = older
+        .iter()
+        .filter_map(TranscriptItem::as_tool_run)
+        .collect();
+    // Owned rather than borrowed from `newer`: `kept` below needs to consume `newer` by
+    // value, and a map borrowing it would keep that alive.
+    let ended_later: std::collections::HashMap<String, TranscriptItem> = newer
+        .iter()
+        .filter_map(|item| item.as_tool_run().map(|id| (id.to_string(), item.clone())))
+        .filter(|(id, _)| started_earlier.contains(id.as_str()))
+        .collect();
+    let healed: Vec<TranscriptItem> = older
+        .into_iter()
+        .map(|row| match row {
+            TranscriptItem::ToolRun {
+                seq,
+                id,
+                run_id,
+                tool,
+                input,
+                asks: row_asks,
+                images: row_images,
+                ..
+            } if ended_later.contains_key(id.as_str()) => {
+                let &TranscriptItem::ToolRun {
+                    ref output,
+                    done,
+                    asks: ref half_asks,
+                    images: ref half_images,
+                    ..
+                } = &ended_later[id.as_str()]
+                else {
+                    unreachable!("filtered to ToolRun above");
+                };
+                TranscriptItem::ToolRun {
+                    seq,
+                    id,
+                    run_id,
+                    tool,
+                    input,
+                    output: output.clone(),
+                    done,
+                    // Kept from both halves: a question or an image can be
+                    // attached to either, depending on which side of the
+                    // boundary its event fell.
+                    asks: row_asks.into_iter().chain(half_asks.clone()).collect(),
+                    images: row_images.into_iter().chain(half_images.clone()).collect(),
+                }
+            }
+            other => other,
+        })
+        .collect();
+    let kept: Vec<TranscriptItem> = newer
+        .into_iter()
+        .filter(|item| match item.as_tool_run() {
+            Some(id) => !ended_later.contains_key(id),
+            None => true,
+        })
+        .collect();
+    let mut out = adopt_run(&healed, &kept);
+    out.extend(kept);
+    // What this function exists to prevent, checked rather than assumed: the same
+    // call drawn twice, once from the page that saw its start and once from the page
+    // that saw its end. Not a seq-ordering check -- a peer note is stamped with the
+    // seq its turn began at, which can be older than the page it arrived in, so the
+    // two pages' seqs legitimately interleave at the boundary.
+    debug_assert!(
+        {
+            let mut ids: Vec<&str> = out.iter().filter_map(TranscriptItem::as_tool_run).collect();
+            let before = ids.len();
+            ids.sort_unstable();
+            ids.dedup();
+            ids.len() == before
+        },
+        "join_pages left the same tool call in both halves"
+    );
+    out
+}
+
+/// Rejoins a message the page boundary cut, and hands back the two pages to
+/// concatenate. Ported from `TranscriptItems.kt`'s `healSplitMessage`.
+///
+/// `fold_event` never leaves two assistant messages next to each other
+/// inside one page, so two meeting at a join are always the two halves of
+/// one reply, and leaving them apart drew a single answer as two with a
+/// paragraph break through the middle of a sentence.
+///
+/// The newer half keeps its identity, for the reason `adopt_run`'s doc
+/// gives. It grows by what the older half brings, which is safe here and
+/// nowhere else -- the join is at the oldest end of what is loaded, so the
+/// growth extends off the top of the screen.
+fn heal_split_message(
+    earlier: &[TranscriptItem],
+    later: &[TranscriptItem],
+) -> (Vec<TranscriptItem>, Vec<TranscriptItem>) {
+    let (
+        Some(TranscriptItem::AssistantMsg {
+            text: head_text, ..
+        }),
+        Some(TranscriptItem::AssistantMsg {
+            seq: tail_seq,
+            text: tail_text,
+            settled: tail_settled,
+        }),
+    ) = (earlier.last(), later.first())
+    else {
+        return (earlier.to_vec(), later.to_vec());
+    };
+    let merged = TranscriptItem::AssistantMsg {
+        seq: *tail_seq,
+        text: format!("{head_text}{tail_text}"),
+        settled: *tail_settled,
+    };
+    let mut newer = vec![merged];
+    newer.extend(later[1..].iter().cloned());
+    (earlier[..earlier.len() - 1].to_vec(), newer)
+}
+
+/// Hands the older calls at the join the name of the run they are joining.
+/// Ported from `TranscriptItems.kt`'s `adoptRun`.
+///
+/// The two pages were folded separately, so a run split by the boundary
+/// came back as two runs with two names. Naming the joined run after the
+/// *older* half would be the obvious way round and is wrong: the newer half
+/// is the part already on screen, and renaming it is renaming the row the
+/// reader is looking at, which is how a list loses its anchor.
+fn adopt_run(earlier: &[TranscriptItem], later: &[TranscriptItem]) -> Vec<TranscriptItem> {
+    let Some(TranscriptItem::ToolRun { run_id, tool, .. }) = later.first() else {
+        return earlier.to_vec();
+    };
+    // A question is in a run of its own on both sides of the join, the same as it would be
+    // had the two pages been folded as one. Without this the heal would merge a group
+    // straight through the row the reader was asked something on.
+    if tool == ASK_USER_QUESTION {
+        return earlier.to_vec();
+    }
+    let joining = run_id.clone();
+    let tail_len = earlier
+        .iter()
+        .rev()
+        .take_while(|item| matches!(item, TranscriptItem::ToolRun { tool, .. } if tool != ASK_USER_QUESTION))
+        .count();
+    if tail_len == 0 {
+        return earlier.to_vec();
+    }
+    let split = earlier.len() - tail_len;
+    let mut out = earlier[..split].to_vec();
+    out.extend(earlier[split..].iter().cloned().map(|mut item| {
+        // `take_while` above already restricted this slice to non-question tool calls;
+        // this just guards the invariant rather than trusting it silently.
+        debug_assert!(
+            matches!(&item, TranscriptItem::ToolRun { tool, .. } if tool != ASK_USER_QUESTION),
+            "adopt_run must never rename a question's own run"
+        );
+        if let TranscriptItem::ToolRun { run_id, .. } = &mut item {
+            *run_id = joining.clone();
+        }
+        item
+    }));
+    out
+}
+
 /// Folds one transcript event onto `items`, the way `foldEvent` does in
 /// `TranscriptItems.kt`. Every wire event has a case; see the module doc
 /// for the one difference from the Kotlin original (no `Unknown` fallback
@@ -955,4 +1148,134 @@ mod tests {
         let err = fold_page(&values).unwrap_err();
         assert!(err.contains("couldn't parse"));
     }
+
+    fn tool_start(seq: u64, id: &str, tool: &str) -> SeqEvent {
+        event(
+            seq,
+            Event::ToolStart {
+                id: id.to_string(),
+                tool: tool.to_string(),
+                input: serde_json::json!({}),
+            },
+        )
+    }
+
+    fn tool_end(seq: u64, id: &str, output: &str) -> SeqEvent {
+        event(
+            seq,
+            Event::ToolEnd {
+                id: id.to_string(),
+                output: output.to_string(),
+            },
+        )
+    }
+
+    /// AGENTS.md's "things that have bitten": `joinPages` used to run
+    /// `adoptRun` only on the path where a *split* call was found, so a
+    /// boundary landing cleanly between two already-finished calls -- most
+    /// of them -- left the older page's calls under the run name they were
+    /// folded with, drawing one run of tool calls as two groups. Two
+    /// finished, unrelated calls (no id in common) must still end up under
+    /// one run name after the join.
+    #[test]
+    fn a_clean_boundary_between_two_finished_runs_is_still_healed_into_one_run() {
+        let older = fold_all(&[tool_start(1, "a", "Bash"), tool_end(2, "a", "old output")]);
+        let newer = fold_all(&[tool_start(3, "b", "Bash"), tool_end(4, "b", "new output")]);
+        let joined = join_pages(&older, &newer);
+        let run_ids: Vec<_> = joined
+            .iter()
+            .map(|item| match item {
+                TranscriptItem::ToolRun { run_id, .. } => run_id.as_str(),
+                other => panic!("expected only ToolRun items, got {other:?}"),
+            })
+            .collect();
+        assert_eq!(
+            run_ids,
+            vec!["b", "b"],
+            "the older call must adopt the newer, already-on-screen run's name"
+        );
+    }
+
+    #[test]
+    fn a_call_split_across_the_boundary_merges_into_one_row() {
+        let older = fold_all(&[tool_start(1, "x", "Bash")]);
+        let newer = fold_all(&[tool_end(2, "x", "the result")]);
+        let joined = join_pages(&older, &newer);
+        assert_eq!(
+            joined,
+            vec![TranscriptItem::ToolRun {
+                seq: 1,
+                id: "x".to_string(),
+                run_id: "x".to_string(),
+                tool: "Bash".to_string(),
+                input: "{}".to_string(),
+                output: "the result".to_string(),
+                done: true,
+                asks: Vec::new(),
+                images: Vec::new(),
+            }],
+            "the older half's tool/input and the newer half's output/done must both survive"
+        );
+    }
+
+    #[test]
+    fn a_message_split_across_the_boundary_is_rejoined_with_the_newer_halfs_identity() {
+        let older = vec![TranscriptItem::AssistantMsg {
+            seq: 1,
+            text: "Hel".to_string(),
+            settled: false,
+        }];
+        let newer = vec![
+            TranscriptItem::AssistantMsg {
+                seq: 2,
+                text: "lo".to_string(),
+                settled: true,
+            },
+            TranscriptItem::UserMsg {
+                seq: 3,
+                text: "next".to_string(),
+                attachments: Vec::new(),
+            },
+        ];
+        let joined = join_pages(&older, &newer);
+        assert_eq!(
+            joined,
+            vec![
+                TranscriptItem::AssistantMsg {
+                    seq: 2,
+                    text: "Hello".to_string(),
+                    settled: true,
+                },
+                TranscriptItem::UserMsg {
+                    seq: 3,
+                    text: "next".to_string(),
+                    attachments: Vec::new(),
+                },
+            ]
+        );
+    }
+
+    /// A question is in a run of its own on both sides of a join -- healing
+    /// must never rename the run of calls the reader was asked something
+    /// on, the same rule `splitRun` enforces for a live turn boundary.
+    #[test]
+    fn adopt_run_never_renames_into_a_question_row() {
+        let older = fold_all(&[tool_start(1, "a", "Bash"), tool_end(2, "a", "done")]);
+        let newer = vec![TranscriptItem::ToolRun {
+            seq: 3,
+            id: "q".to_string(),
+            run_id: "q".to_string(),
+            tool: ASK_USER_QUESTION.to_string(),
+            input: "{}".to_string(),
+            output: String::new(),
+            done: false,
+            asks: Vec::new(),
+            images: Vec::new(),
+        }];
+        let joined = join_pages(&older, &newer);
+        match &joined[0] {
+            TranscriptItem::ToolRun { run_id, .. } => assert_eq!(run_id, "a"),
+            other => panic!("expected a ToolRun, got {other:?}"),
+        }
+    }
 }

client-core/src/transcript_source.rs

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

docs/CLIENT_CORE.md

@@ -25,16 +25,19 @@ next (a Masonry or iris transcript screen, most likely).
 | `sse.rs`                                  | `Sse.kt` (the framing half)               | Done, new tests (Kotlin had none of its own beyond integration) |
 | `api.rs`                                  | `Api.kt`                                  | Partial -- see below |
 | `event_stream.rs`                         | `EventStream.kt`                          | Done |
-| `transcript_fold.rs`                      | `TranscriptItems.kt`, `ToolRows.kt`       | Partial -- see below |
+| `transcript_fold.rs`                      | `TranscriptItems.kt`, `ToolRows.kt`       | Done -- see below |
 | `config.rs`                                | `ServerConfig.kt`'s `handleEnrollment`   | New, desktop-only so far -- see below |
-| *(not started)*                           | `TranscriptSource.kt`                     | Not started |
+| `transcript_source.rs`                    | `TranscriptSource.kt`                     | Done -- see below |
 | *(not ported, and may never be)*          | `TranscriptUnits.kt`                      | Out of scope -- see below |
 
 Every file above whose Kotlin counterpart had a JVM unit test (`AnsiTest`,
 `HighlighterTest`, `TranscriptCacheTest`) has had every one of those test
 cases ported alongside it, plus new tests for the pieces that had none
-(`sse.rs`, `api.rs`, `event_stream.rs`, `transcript_fold.rs`). Test count by
+(`sse.rs`, `api.rs`, `event_stream.rs`, `transcript_fold.rs`,
-crate as of this writing: **85 in `client-core`**, 0 in `event-model` (its
+`transcript_source.rs` -- the Kotlin `TranscriptSource.kt`/`TranscriptItems.kt`
+had no JVM unit tests of their own, so these were written fresh against the
+Kotlin source and AGENTS.md's paging incidents as the spec). Test count by
+crate as of this writing: **109 in `client-core`**, 0 in `event-model` (its
 types carry no logic of their own to test -- `server/`'s own tests exercise
 them via `session::transcript`'s round-trip coverage).
 
@@ -88,13 +91,27 @@ the full table to work from when one of these is next.
 including tool-call/question/image attachment and peer-message placement.
 `group_tool_runs` groups adjacent calls into `TranscriptRow::Tools`.
 
-**Not ported:** `TranscriptItems.kt`'s `joinPages` (and its
+`join_pages` (with `heal_split_message` and `adopt_run`, both private) is
-`healSplitMessage`/`adoptRun` helpers) -- the page-boundary healing that
+now ported too, 2026-09-06 -- the page-boundary healing that merges a tool
-merges a tool call split across two fetched pages and re-merges a run a
+call split across two fetched pages, rejoins a message a boundary cut
-boundary cut through. This matters the moment paging backward through
+through, and renames a run of tool calls onto whichever name is already on
-history is exercised; it is deliberately left rather than rushed, since
+screen. Ported with AGENTS.md's "things that have bitten" incidents as the
-it is exactly the kind of boundary logic this project's own "things that
+spec rather than a JVM test file (`TranscriptItems.kt` had none of its
-have bitten" section warns reads fine and is wrong at the edges.
+own): `a_clean_boundary_between_two_finished_runs_is_still_healed_into_one_run`
+is the regression test for the bug that shipped -- `adopt_run` must run on
+*every* join, not only the one where a split call was found, or a boundary
+landing cleanly between two already-finished calls (most of them) leaves
+one run drawn as two. `a_call_split_across_the_boundary_merges_into_one_row`,
+`a_message_split_across_the_boundary_is_rejoined_with_the_newer_halfs_identity`,
+and `adopt_run_never_renames_into_a_question_row` cover the other three
+edges the Kotlin doc calls out. `join_pages` ends in a `debug_assert!`
+that no tool id survives in both halves -- the duplicate row it exists to
+prevent, checked rather than assumed. What it deliberately does *not*
+assert is seq ordering across the boundary: a peer note carries the seq
+its turn began at (`place_peer_note`), which can be older than the page
+it arrived in, so the two pages' seqs legitimately interleave there. An
+earlier draft asserted it and would have panicked in debug builds on an
+ordinary transcript.
 
 **Known gap, and a decision for whoever closes it:** `event_model::Event`
 has no `Unknown`/catch-all variant, unlike `Events.kt`'s hand-kept mirror.
@@ -119,12 +136,61 @@ caller-specific (the code rules' "ask for the least you need"). Its only
 caller today is `desktop-app`; a future Android build of this crate would
 be a second one, not a reason to move the type.
 
+## What `transcript_source.rs` covers, and what it does not
+
+`TranscriptSource<T: Transport>` is the seam a session screen asks for a
+page, ported test-for-test against the Kotlin doc rather than a JVM test
+file (there wasn't one): `cached_opening`, `probe`, `fetch_opening`,
+`page` and `follow`, each matching its Kotlin namesake's contract --
+including `probe`'s three-way outcome (matches / cache purged /
+unreachable, told apart so a caller never treats "couldn't ask" as "was
+wrong") and `page`'s cache-vs-server split bounded by `covered_up_to`.
+
+Two additions beyond a literal port, both load-bearing:
+
+- **`page(before, ..)` refuses `before == 0` before touching the cache or
+  the network**, answering `OlderPage::NothingLoaded`. This is AGENTS.md's
+  `loadOlderPage` incident (`before = 0` is "no event before the first
+  one," indistinguishable from "reached the start of history" if a caller
+  ever asks it) moved out of the Kotlin screen and into this layer, so
+  every future caller gets the guard rather than having to remember it.
+  **The return type is `OlderPage`, not a `Vec`, and that is the guard.**
+  The Kotlin's two falses are different answers -- `oldestSeq == 0`
+  returns without touching `moreHistory`, an empty page latches it false
+  -- so a port that answered both with an empty list would have moved the
+  bug rather than fixed it, one layer down and out of sight of the screen
+  that used to hold the check. `OlderPage::Events(vec![])` means the start
+  of the conversation; `OlderPage::NothingLoaded` is not an answer about
+  the conversation at all. Reviewed 2026-09-06.
+  `paging_before_the_first_event_makes_no_request_at_all` asserts zero
+  transport calls, not just the variant, since a request that happens to
+  answer empty is exactly what caused the original bug, and
+  `a_failing_server_page_is_an_error_rather_than_an_empty_one` plus
+  `an_unreadable_cached_page_is_a_parse_error_rather_than_an_empty_one`
+  are the same rule for the two ways a page can fail.
+- **`fetch_transcript_lines`** (new in `api.rs`) hands back each line
+  paired with the exact server bytes it came from, via
+  `serde_json::value::RawValue` rather than re-serializing a parsed
+  `Value` -- the cache and a live SSE frame for the same event have to
+  agree byte-for-byte, which is exactly what the `serde_json`
+  float-rounding bug (AGENTS.md) was about. The existing
+  `fetch_transcript_page` is untouched (other callers under `iris/`
+  depend on its signature); the two share a `transcript_path` helper so
+  the query string is written in one place.
+
+**Not ported:** `EventStream.kt`'s reconnect-with-backoff loop, and
+`TranscriptSource.close`'s ability to cancel a live stream from another
+thread. Both are wall-clock/thread-lifetime policy that belongs to
+whichever runtime embeds this crate (iris's own timers, a Tokio task, a
+Kotlin coroutine scope), not to this pure logic -- `follow` is the same
+"write to the cache, then hand the frame to the caller" decorator
+`iris/desktop-app/src/app.rs` and `iris/android-app/src/transcript_client.rs`
+already hand-wrote around `event_stream::follow_session_events` before this
+existed; the cache write moved into one shared place so a third caller
+does not repeat it again by hand.
+
 ## What is not started at all
 
-- **`TranscriptSource.kt`** -- the layer that decides whether a page comes
-  from the transcript cache or the server, and stitches the two. Needs
-  `transcript_cache.rs` and `api.rs`'s transcript-page method, both of
-  which exist now, so this is unblocked whenever picked up.
 - **The markdown *block* model beyond syntax spans** -- `highlight/markdown.rs`
   colours a `.md` file or fence for the highlighter, but does not build the
   block tree (headings, lists, tables, fences as distinct nodes) that a
@@ -142,5 +208,6 @@ be a second one, not a reason to move the type.
 
 `./run-tests.sh` from the repo root now runs `event-model`, `client-core`
 and `server` in that order (each `cargo test`, forwarding arguments the
-same way it always has). From `client-core/` directly: `cargo test`,
+same way it always has). From `client-core/` directly: `cargo test`
-`cargo clippy --all-targets`, `cargo fmt` -- all clean as of this writing.
+(109 tests), `cargo clippy --all-targets`, `cargo fmt` -- all clean as of
+this writing (2026-09-06).

docs/IRIS.md

@@ -8,6 +8,31 @@ capability that moved. Small and trivial changes do not go here.
 An entry gives the date, what changed, why, and a short before/after where
 it helps judge the change without the session that made it. Newest first.
 
+## 2026-09-06: `List::anchor_position_display`, `FrameReport::mark_phase`/`phase_stats`/`late_at_hz` (RUST.md's "Benchmark v2")
+
+`List` gained `anchor_position_display(&self) -> String`, reporting the
+anchor's own row index and pixel offset (`idx=N/off=Mpx`, or
+`idx=more-before`/`idx=more-after`/`idx=none`) -- what a scripted
+benchmark reads to report fling travel. Note the anchor does not
+necessarily change *slot* over a long scroll (this widget's own documented
+design: the anchor is a stable identity, not re-derived from what's on
+screen each frame), so this is not the same measurement as a Compose
+`LazyListState.firstVisibleItemIndex`, which does track the true topmost
+visible row -- the `off` half is what actually reflects how far a fling
+travelled.
+
+`iris_core::render::frame_report::FrameReport` gained three methods for
+per-phase benchmark reporting: `mark_phase(name)` records a named phase
+boundary at the current frame/instant; `phase_stats(now, refresh_hz)`
+returns one `PhaseStats` (frames, wall duration, late count/percent,
+p50/p90/p99, worst) per marked phase, sliced from the existing ring by a
+new parallel `index_ring`; `late_at_hz(refresh_hz)` gives the whole run's
+late count/percent judged against an arbitrary refresh rate rather than
+the fixed 60Hz `JANK_THRESHOLD` every existing caller still uses (a
+separate method, not a parameter on `report()`, so nothing else changes
+behaviour). `RING_CAPACITY` grew 4096->16384 to hold a full multi-phase
+run without evicting earlier phases' samples.
+
 ## 2026-09-06: `List::fling`, `VelocityTracker`, `FlingCalculator` (IRIS_TODO.md's "swiping has no momentum")
 
 `iris::widget::List` gained a real fling: `fling(velocity_px_per_s)` starts
@@ -626,3 +651,32 @@ box has the full investigation and the phone verification still to do.
   built and checked on this checkout's emulator only. RUST.md's P0 box
   says what she should check for: crisp text at two densities, the
   keyboard no longer wiping, and the header's background.
+
+## 2026-09-06: composing text, focus-on-tap, and atlas invalidation on a new renderer
+
+Three small but public API changes, from the same phone-report pass as the
+entry above (RUST.md's P0 box has the full account, including a real bug
+still not root-caused).
+
+- **`FocusHost` gained `is_focused(&self, id) -> bool`** (both platform
+  impls). `attr.rs`'s `Selector`/`Selectable` used to grant focus (and so
+  request the IME) on the very first frame of *any* press, before it was
+  known whether the gesture was a tap or a drag — a swipe over a text
+  field wrongly summoned the keyboard. They now wait for a completed tap
+  (press and release with no frame crossing `sense::DRAG_SLOP`) unless the
+  field is already focused, in which case dragging inside it to select
+  text is unchanged. `TextEdit` gained one new `pub(crate)` field
+  (`press_origin`) to track this; no public surface change there.
+- **`android::ime`'s `InputConnection` now calls `InputMethodManager::
+  updateSelection` after every edit** (`IrisViewPeer::update_ime_selection`,
+  called from `after_input`). Gboard was holding keystrokes back because
+  nothing ever told it where the app's own selection/composing region had
+  moved to — this is what android-view's own demo does in its `render()`
+  and this bridge never did.
+- **`GlyphAtlas::clear()` and `Textures::reset()`** (`iris_core`). Called
+  together, once, from `android::view`'s `surface_changed` exactly when a
+  *genuinely new* `AndroidRenderer` is built (backgrounding and returning,
+  not a keyboard-triggered resize, which already reuses the renderer) —
+  both CPU-side caches otherwise kept pointing at the old, now-destroyed
+  device's textures, which is why text used to vanish again after leaving
+  and returning to the app.

docs/IRIS_TODO.md

@@ -149,6 +149,81 @@ agent takes them without colliding with that pass's `bench_client.rs`/
   confirming this was the whole story on real touch input rather than
   only the arbiter's own unit tests -- worth a follow-up pass before
   calling it fully closed.
+- [x] **Composing text held back until a space, caret not moving, fixed
+  2026-09-06.** `InputMethodManager.updateSelection` was never called --
+  see IRIS.md's 2026-09-06 entry and RUST.md's P0 box, item 1, for the
+  full account and the emulator evidence.
+- [x] **Swipe over the composer summons the keyboard, fixed 2026-09-06.**
+  `Selector`/`Selectable` now wait for a completed tap -- see IRIS.md's
+  2026-09-06 entry and RUST.md's P0 box, item 5. Verified via `dumpsys
+  input_method`'s `mInputShown` on the emulator, not yet on the phone.
+- [x] **Text disappears again after leaving and returning to the app,
+  fixed 2026-09-06.** `GlyphAtlas::clear`/`Textures::reset` on a
+  genuinely new renderer -- see IRIS.md's 2026-09-06 entry and RUST.md's
+  P0 box, item 4. Verified on the emulator (home, reopen, screenshot);
+  not yet on the phone.
+- [ ] **Composed/typed text never becomes visible at all -- found
+  2026-09-06, not fixed.** The composer bar stays empty even once the
+  buffer genuinely holds the typed text (confirmed indirectly: Gboard's
+  own suggestion strip reacts correctly to each keystroke). A new unit
+  test proves the widget tree's own layout math resolves the field's
+  region correctly across a keyboard resize, so the bug is downstream of
+  that -- most likely `UiRenderState::redraw`'s single-widget redraw path,
+  or specific to this emulator's forced `force-gles` backend (untested on
+  Vulkan or the real phone). RUST.md's P0 box, item 2, has the full
+  writeup, what was ruled out, and where to look next. **Also unverified
+  because of this**: item 3's composer rebuild (one `Stack`-based widget,
+  a capped/scrollable height, bottom padding tied to the IME/nav-bar
+  inset) -- structurally in place and unit-tested, but its own visual
+  correctness cannot be screenshotted until text actually renders.
+- [ ] **The composer has no touch-drag scroll for overflowing text.** The
+  2026-09-06 rebuild caps the field at ~6 lines and wraps it in
+  `.scrollable()` for a wheel/trackpad scroll, but a real finger drag over
+  text that has overflowed the cap does not scroll it -- `Scroll`'s touch
+  handling is a follow-up, the same shape `List`'s own touch-drag pan
+  needed before I3/I5.
+
+## From the phone, 2026-09-06, 11:39 (build delivered 02:07, commit 543f6d9)
+
+Iris's report on the build with the composing-text, tap-vs-swipe and
+atlas-reset fixes, with a screenshot, verbatim. Each is open until an
+agent ticks it here with the evidence.
+
+- [ ] **"The app definitely does not start with keyboard spacing
+  correct. This is how it looks without me doing anything initially."**
+  The screenshot shows the composer bar (the grey band) sitting about
+  two thirds of the way down a 704x1568 screen, with black below it to
+  the bottom, and the transcript ending at "Claude / Results" just above
+  it -- at launch, no keyboard. So the composer's bottom padding, which
+  the 2026-09-06 rebuild tied to the IME/nav-bar inset, is being fed a
+  large value at start on the phone. Suspects, in order: the initial
+  inset delivery on the phone (GrapheneOS, gesture navigation) versus
+  the emulator; `ime_bottom` now carrying a `1`/`0` boolean through a
+  field the composer may still read as pixels or dp; a stale value from
+  before the first `on_insets_changed`. Reproduce with the phone's
+  screen size and density on the emulator before guessing.
+- [ ] **"Swiping still gets caught by the grey bar but keeps working
+  after I go past it."** A pan that starts on the composer is held by
+  the composer until the finger leaves its region, then the list takes
+  over. The tap-vs-swipe fix in `attr.rs` stops the *focus*, but the
+  press frames are still being handled by the field rather than passed
+  to the list from the first slop-crossing frame. The `DragGesture`
+  merge (RUST.md's plan box) should make this one mechanism: once a
+  gesture commits to a pan, the list captures it wherever it began.
+- [ ] **"Flinging still does not work."** Expected on this build: finger
+  flings are dropped by per-widget hit testing, which `DragGesture`'s
+  pointer capture (commit `e12c708`, not yet merged at 02:07) targets.
+  Stays open until verified on her phone, not the emulator.
+- [ ] **"Text still disappears if I leave and come back to the app."**
+  The `GlyphAtlas::clear`/`Textures::reset` fix was verified on the
+  emulator under `force-gles` only; the phone runs Vulkan. So either the
+  reset is not reached on the phone's path (a different surface-
+  lifecycle sequence -- `surface_destroyed`/`surface_created` ordering,
+  or the renderer not being rebuilt but its textures lost), or the CPU
+  glyph cache and the GPU atlas still disagree after it. Needs logging
+  of the renderer lifecycle on the phone build, readable from `adb
+  logcat` when Iris next runs it, since no emulator here has a Vulkan
+  adapter under host GPU.
 
 ## Build
 
@@ -450,3 +525,21 @@ do not duplicate it there.
   and control sizes; the emulator at two densities and the phone draw the
   same layout at the same physical size. After the bench setup is
   finished, before P1 draws any new screen.
+
+## From the phone, bench v2 (2026-09-06): streaming re-lays out the whole message
+
+- [ ] **Streaming a delta into a long message costs a full text layout of
+  that message.** Iris's phone report (`docs/bench/iris-phone-v2-2026-09-06.md`):
+  the stream phase is the one place iris is behind Compose (p50 18.2 ms vs
+  13.4 ms; p99 level at ~43 ms). `TranscriptScreen::apply` replaces only
+  the last row, but that row is the growing message, and replacing it
+  re-renders its markdown and re-shapes the entire paragraph run through
+  parley on every event. Compose pays a reparse (8.6 ms mean) for the
+  same event. What "done" looks like: a streamed delta re-lays out only
+  the block it lands in (the last paragraph or code block), with earlier
+  blocks' layouts kept -- which needs a row to be a column of per-block
+  `Text`s rather than one `TextEdit` for the whole message, or parley's
+  layout to be split at block boundaries; measured by the stream phase's
+  p50 dropping below Compose's on the phone. Do this after the four bench
+  v2 defects (stale primitives, finger fling, decay curve, IME show) are
+  closed, since they are what make the run unrepresentative today.

docs/RUST.md

@@ -34,6 +34,87 @@ 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.
+
+- [ ] **Merge the `DragGesture` work** left complete but unmerged in the
+      worktree branch `worktree-agent-a754368325fa06839` (commit
+      `e12c708`, 2026-09-06 02:10, two minutes after the last merge to
+      `rustify`; already contains `rustify` at `543f6d9`). It 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). Before
+      merging: build, clippy, tests; then on the emulator confirm the
+      three things `20b1225` (tap-vs-swipe focus) and the phone asked for
+      still hold together — a swipe over the composer does not summon
+      the keyboard, a tap does, a finger fling on the list keeps moving
+      after the finger lifts, and `on_insets_changed` now fires on an
+      IME toggle. Then remove the worktree.
+- [ ] **Fix `docs/REVIEW-2026-09-06.md`** (after the merge, since the
+      review's finding 1 is in `selection.rs`, which the merge rewrites).
+      Finding 1 is a real crash — `TranscriptScreen::apply`'s `Rebuild`
+      arm leaves `Selection` holding `WeakWidget`s to freed rows, and the
+      next tap anywhere panics. Findings 2–5 are `debug_assert!`s on
+      invariants, 8 and 10 are the missing tests. Commit the review file
+      with the fixes.
+- [ ] **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** — the header drawn
+      twice after a keyboard resize, the `Compacted:` row drawn twice on
+      Iris's phone, and typed text never appearing (P0 box item 2). All
+      three sit on the `redraw_updates` targeted-redraw path and may be
+      one bug; the P0 box says the next step is instrumentation inside
+      `Span::draw`/`draw_inner` showing where each placement's
+      primitives actually land on the frame it goes wrong.
+      `list.rs`'s new `replacing_the_last_row_many_times_does_not_leak_
+      primitives` test already pins the widget arena as *not* the leak.
+- [ ] **Composer touch-drag scroll** for overflowed text — now that
+      dragging is a default-input `DragGesture`, `Scroll` should get its
+      touch pan from the same mechanism `List` uses, not a copy.
+- [ ] **Streaming re-layout** (IRIS_TODO.md's last section) — after the
+      above, since they make the stream phase unrepresentative today.
+- [x] **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 `~/host/bench/iris-bench-arm64.apk` for Iris with
+  its README saying what changed, and 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
@@ -4573,13 +4654,344 @@ device.
       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's four-phase spec (fling/stream/
+      real touch input.
-      type/keyboard) in `bench_client.rs` was not attempted this pass** --
+
-      wiring a real IME show/hide and refresh-rate read through
+      **Benchmark v2, iris half, done 2026-09-06, later the same day.**
-      `bench_jni.rs`, and `FrameReport`'s per-phase accounting, is real
+      `bench_client.rs` implements all four phases against the identical
-      scope on its own and was left rather than shipped half-verified;
+      constants this box's "Benchmark v2" spec names: fling (8 out + 8
-      the Compose half above is already done and is the reference shape
+      back at 12,000px/s through `List::fling`, waiting for
-      for whoever picks this up. No redelivery this pass.
+      `!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.
+
+      **(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.** History paging backward (with the
       page-boundary healing `client-core` does not have yet, below),

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

@@ -0,0 +1,58 @@
+# iris bench v2 report from Iris's phone, 2026-09-06
+
+Build 2e3f4ad (bench v2, fling physics, keyboard-wipe fix, dp unit), run
+by Iris on her Pixel 9 Pro XL, verbatim. The display was at **120 Hz**
+(8.3 ms budget) where `compose-phone-v2-2026-09-06.md` ran at 60 Hz, so
+compare the millisecond percentiles, not `late`.
+
+Side by side (Compose 60 Hz / iris 120 Hz, p50 / p90 / p99 ms): fling
+5.5/8.7/11.6 vs 3.8/6.9/12.6; stream 13.4/31.7/42.5 vs 18.2/35.8/43.1;
+type 7.3/13.2/16.5 vs 7.2/9.2/11.2; keyboard: iris could not show the IME
+(phase invalid). Process CPU 69.6 s over 125 s vs 40.6 s over 150 s; peak
+RSS 577 MB vs 379 MB; battery current mean 571 mA vs 452 mA.
+
+Iris's observations on the same run: "the scrolling is not similar at
+all. It does not fling for me yet [with a finger], and the test also seems
+to give it a constant velocity and abruptly stop it at some point. Also
+unsure what's going on in that image with the compaction" -- her
+screenshot shows the `Compacted: 180000 -> 20000 tokens.` row drawn twice
+overlapping, and once more below the composer bar: primitives of a
+replaced/removed row surviving in the GPU buffers, the same shape as the
+header drawn twice after a keyboard resize.
+
+```
+iris bench report
+per phase:
+  fling: 1783 frames over 53.2s
+    late: 104 (5.8%)
+      total  p50 3.8ms  p90 6.9ms  p99 12.6ms
+    worst 29.1ms
+  stream: 401 frames over 21.3s
+    late: 306 (76.3%)
+      total  p50 18.2ms  p90 35.8ms  p99 43.1ms
+    worst 43.8ms
+  type: 1202 frames over 65.7s
+    late: 309 (25.7%)
+      total  p50 7.2ms  p90 9.2ms  p99 11.2ms
+    worst 15.3ms
+  keyboard: 9 frames over 9.7s
+    late: 9 (100.0%)
+      total  p50 12.0ms  p90 12.9ms  p99 12.9ms
+    worst 12.9ms
+
+frames:
+  3395 frames over 149.9s at 120Hz (8.3ms budget)
+  late: 728 (21.4%)
+  total  p50 5.0ms  p90 10.9ms  p99 36.6ms
+  worst 43.8ms
+  cpu_p50 2.0ms  gpu_wait_p50 2.6ms
+
+bench:
+  fling: 8 flings out + 8 back at 12000px/s, travel start=idx=651/off=1217px outward=idx=651/off=101536px end=idx=651/off=1022px
+  scroll: 6 cycles (24 swipes, legacy tween), streamed 400/400 fixture events
+  type: 600 characters inserted then deleted, one per 50ms
+  keyboard: could not be shown (5 attempts, 0 confirmed visible)
+  process CPU time over this run: 40603ms
+  peak RSS: 379156kB
+  battery current: mean -452353µA over 149 samples (min -1753125, max -204687)
+```

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

@@ -31,14 +31,52 @@ public final class MainActivity extends Activity {
         setContentView(layout);
         view.requestFocus();
 
+        // RUST.md's P0 box, defect 4 ("keyboard: could not be shown"):
+        // `logcat` showed the platform's own IME open/resize happening
+        // while `setOnApplyWindowInsetsListener` fired only once, at
+        // attach, and never again for a pure keyboard toggle -- a plain
+        // (non-edge-to-edge) window is only guaranteed that one initial
+        // dispatch; `adjustResize` handling the IME entirely by resizing
+        // the window is not itself a trigger for a fresh one. Opting into
+        // edge-to-edge (a platform call, API 30+, no new dependency) is
+        // what makes the system redeliver insets on every change,
+        // including the ones this activity actually cares about --
+        // `getSystemWindowInset*` below is unaffected by this (it has
+        // always reported the raw system-bar/IME overlap regardless of
+        // who consumes it), so the on-screen bars and the padding Rust
+        // already derives from those four numbers are unchanged; only the
+        // callback's firing became reliable.
+        if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.R) {
+            getWindow().setDecorFitsSystemWindows(false);
+        }
+
         view.setOnApplyWindowInsetsListener((v, insets) -> {
             int left = insets.getSystemWindowInsetLeft();
             int top = insets.getSystemWindowInsetTop();
             int right = insets.getSystemWindowInsetRight();
             int bottom = insets.getSystemWindowInsetBottom();
+            // The manifest declares adjustResize (AGENTS.md: without it the
+            // keyboard pans the whole window instead of resizing it), and
+            // under adjustResize the window itself shrinks to make room for
+            // the keyboard -- which is exactly the condition under which
+            // WindowInsets.Type.ime()'s own *inset amount* reports zero: it
+            // measures how much of the window the keyboard overlaps, and
+            // resize already made that overlap zero by construction. That
+            // numeric inset is not a usable "is the keyboard open" signal
+            // here (found while root-causing why bench_client.rs's keyboard
+            // phase and auto-diagnostics never fired on the emulator despite
+            // the keyboard visibly opening -- RUST.md's P0 box). What does
+            // survive adjustResize is the boolean isVisible() answer, set
+            // from the platform's own start/end of the transition over a
+            // different path than the inset amount -- the same fact
+            // AGENTS.md's "Things that have bitten" already names for the
+            // Compose side's identical trap. Passed through as a 0/1 stand-
+            // in for the ime_bottom pixel amount, since nothing on the Rust
+            // side reads it as a real pixel value -- only `> 0.0`.
             int imeBottom = 0;
-            if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.R) {
+            if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.R
-                imeBottom = insets.getInsets(WindowInsets.Type.ime()).bottom;
+                    && insets.isVisible(WindowInsets.Type.ime())) {
+                imeBottom = 1;
             }
             ((IrisView) v).applyWindowInsets(left, top, right, bottom, imeBottom);
             return insets;

iris/android-app/run-bench.sh

@@ -46,10 +46,13 @@ adb -s "$SERIAL" shell am start -n "$PKG/dev.iris.android.demo.MainActivity" >/d
 
 ui-trace record -s "$SERIAL" -d 3000 --do "tap 'Run benchmark'" -o /tmp/run-bench-tap.txt >/dev/null
 
-# Poll for the report line rather than a fixed sleep -- the run itself is a
+# Poll for the report line rather than a fixed sleep -- the run itself is
-# fixed script (24 swipes + a 20s streaming phase) but device speed varies.
+# a fixed script (RUST.md's "Benchmark v2": 16 flings, a 20s streaming
+# phase, ~61s of typing, 10s of keyboard toggles, roughly 2.5 minutes end
+# to end) but device speed varies. 260s cap rather than v1's 90s -- v2 is
+# a longer script than v1's swipe-loop-only run.
 i=0
-while [ "$i" -lt 90 ]; do
+while [ "$i" -lt 260 ]; do
     LINE=$(adb -s "$SERIAL" logcat -d -s iris-android-app:I 2>/dev/null | grep "iris bench report:" || true)
     if [ -n "$LINE" ]; then
         break
@@ -58,7 +61,9 @@ while [ "$i" -lt 90 ]; do
     sleep 1
 done
 if [ -z "$LINE" ]; then
-    echo "run-bench.sh: no report after 90s -- check logcat by hand" >&2
+    echo "run-bench.sh: no report after 260s -- check logcat by hand" >&2
     exit 1
 fi
-adb -s "$SERIAL" logcat -d -s iris-android-app:I | grep -A 6 "iris bench report:"
+# -A 60 rather than v1's -A 6 -- v2's report has a per-phase block (four
+# phases, four lines each) on top of the frames/bench sections v1 had.
+adb -s "$SERIAL" logcat -d -s iris-android-app:I | grep -A 60 "iris bench report:"

iris/android-app/src/bench_client.rs

@@ -26,9 +26,9 @@ use client_core::transcript_fold::{TranscriptItem, fold_event, fold_page, group_
 use event_model::SeqEvent;
 use iris::android::{AndroidAppState, AndroidRsc, AndroidUiState, HasAndroidUiState};
 use iris::prelude::*;
-use std::sync::Arc;
 use std::sync::atomic::{AtomicBool, Ordering};
-use std::time::Duration;
+use std::sync::{Arc, Mutex};
+use std::time::{Duration, Instant};
 
 /// bench-fixture/README.md: the first `BACKLOG_COUNT` non-blank lines are
 /// the opening window; the rest are the streaming tail. Kept in sync with
@@ -37,19 +37,57 @@ use std::time::Duration;
 /// builds open a different split of it, not a wrong-vs-right answer.
 const BACKLOG_COUNT: usize = 3200;
 
-/// `BenchRun.kt`'s own constants -- kept identical so the two apps' bench
+/// RUST.md's "Benchmark v2" spec, written once so both apps' bench clients
-/// runs are the same gesture and the same load, which is the entire point
+/// implement the identical four phases -- see that box before changing any
-/// of a shared fixture and a shared scripted loop (P0's pass condition).
+/// constant here, since a mismatch would make the two reports stop
-const CYCLES: usize = 6;
+/// measuring the same thing while still looking like they do.
-const SWIPE_PX: f32 = 900.0;
-const SWIPE_MS: u64 = 200;
-const SWIPE_PAUSE_MS: u64 = 500;
 const STREAM_EVENTS_PER_SEC: u64 = 20;
 const STREAM_SECONDS: u64 = 20;
+
+/// Kept only so this phase's own label text still reads "scroll: 6 cycles
+/// (24 swipes, legacy tween)" the way `BenchRun.kt`'s v2 report does --
+/// `docs/bench/compose-phone-v2-2026-09-06.md`'s own report shows this
+/// exact line even though the swipe loop it names no longer runs there
+/// either (the fling phase replaced it); nothing here drives an actual
+/// swipe with these any more.
+const LEGACY_CYCLES: usize = 6;
+
+/// Fling phase (v2): a real fling through `List::fling`, not a tween --
+/// Iris's ask was that it "travel way faster" than the v1 swipe, and a
+/// tween can never exceed the distance/time it is given while a real
+/// fling decays from an initial velocity the way a finger flick does.
+/// 12,000 px/s matches `BenchRun.kt`'s own constant exactly.
+const FLING_VELOCITY_PX_S: f32 = 12_000.0;
+const FLING_COUNT: usize = 8;
+const FLING_SETTLE_CAP_MS: u64 = 3_000;
+const FLING_PAUSE_MS: u64 = 300;
+
+/// Type phase (v2): long, multisyllabic words so the composer actually
+/// wraps and the transcript above it is pushed upward, typed and deleted
+/// one character per `TYPE_CHAR_MS`. Exactly `BenchRun.TYPE_TEXT` --
+/// verified 600 characters by `type_text_is_exactly_600_characters` below.
+const TYPE_TEXT: &str = "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!!!";
+const TYPE_CHAR_MS: u64 = 50;
+
+/// Keyboard phase (v2): five show/hide cycles, a second apart, matching
+/// `BenchRun.kt`'s `KEYBOARD_CYCLES`/`KEYBOARD_SHOW_WAIT_MS`/
+/// `KEYBOARD_HIDE_WAIT_MS`.
+const KEYBOARD_CYCLES: usize = 5;
+const KEYBOARD_WAIT_MS: u64 = 1_000;
+
 /// One animation step's target cadence -- close enough to 60Hz that a
-/// `List::scroll` swipe is many small moves rather than one jump, so
+/// fling/scroll is many small moves rather than one jump, so frames are
-/// frames are actually rendered along the way (the point of animating it
+/// actually rendered along the way, and close enough that a `ctx.update`
-/// at all rather than calling `scroll` once per swipe).
+/// closure's effect (only applied once the next frame callback drains the
+/// task channel -- `IrisViewPeer::drain_tasks`) is visible again quickly
+/// when a later step in the same phase needs to read state back.
 const ANIM_STEP_MS: u64 = 16;
 
 const FIXTURE_JSONL: &str = include_str!("../../../app/bench-fixture/assets/transcript.jsonl");
@@ -74,6 +112,12 @@ pub struct BenchClient {
     platform: Option<Arc<PlatformHandle>>,
     last_report: Option<String>,
     running: bool,
+    /// The keyboard phase's own confirmation channel -- updated from
+    /// `on_insets_changed` (the platform's own answer for whether the IME
+    /// is actually visible, per `WindowInsets::ime_bottom`), read from the
+    /// benchmark's spawned task via the shared `Arc<Mutex<_>>` rather than
+    /// `ctx.update`, since neither side needs the widget tree for this.
+    ime_state: Arc<Mutex<ImeState>>,
     /// Edge-triggers the keyboard diagnostics capture below -- set on the
     /// first `on_insets_changed` where `ime_bottom > 0.0`, cleared on the
     /// first where it is not, so opening the keyboard fires this once
@@ -81,6 +125,22 @@ pub struct BenchClient {
     /// or a status-bar change with the keyboard already up would otherwise
     /// re-fire it).
     keyboard_was_visible: bool,
+    /// The status-bar inset `top_bar` was last padded by -- see
+    /// `on_insets_changed`'s own comment for why this guards the rebuild.
+    last_top_pad: f32,
+}
+
+/// See `BenchClient::ime_state`'s doc. `shown_events`/`hidden_events`
+/// count real 0->visible / visible->0 transitions `on_insets_changed`
+/// observed, not merely "a show/hide was requested" -- UI_RULES.md: never
+/// present an inferred value as a measured one. `run_keyboard_phase` reads
+/// the counters before and after asking for a toggle and calls it
+/// confirmed only if the count moved.
+#[derive(Default)]
+struct ImeState {
+    visible: bool,
+    shown_events: u32,
+    hidden_events: u32,
 }
 
 impl HasAndroidUiState for BenchClient {
@@ -226,7 +286,9 @@ impl AndroidAppState for BenchClient {
             platform: None,
             last_report: None,
             running: false,
+            ime_state: Arc::new(Mutex::new(ImeState::default())),
             keyboard_was_visible: false,
+            last_top_pad: 0.0,
         };
 
         let (backlog, stream_tail) = parse_fixture();
@@ -253,26 +315,80 @@ impl AndroidAppState for BenchClient {
 
     /// Pads the top button row by the status-bar inset -- see `top_bar`'s
     /// field comment. Rebuilds the row rather than mutating a stored
-    /// `Padding` in place, since nothing here holds a handle to one.
+    /// `Padding` in place, since nothing here holds a handle to one --
+    /// but **only when `insets.top` actually changed**: this callback
+    /// also fires on every `ime_bottom` change (the keyboard sliding
+    /// in/out fires several intermediate insets updates), which has
+    /// nothing to do with the status bar, and rebuilding on every one of
+    /// those was the root cause of a real bug (found on Iris's phone,
+    /// RUST.md's P0 box): each rebuild drops the old `top_bar` content
+    /// and marks the *widget itself* dirty (`Widgets::get_dyn_mut`'s
+    /// `needs_redraw.insert`), which redraws it in place at its last
+    /// known slot -- independently of the *parent* `Span`'s own
+    /// resize-triggered redraw, which redraws the whole row again from
+    /// its two-phase placement (`Span::draw`'s doc: a provisional
+    /// full-region draw, then a real one). A `.set()` landing between
+    /// those two phases left one dirty-widget redraw's primitives
+    /// un-freed while the `Span`-driven redraw drew its own copy,
+    /// producing two live copies of the same three buttons in one frame
+    /// -- one at the header's real slot, one wherever `Span`'s
+    /// provisional phase happened to leave it (visibly inside the
+    /// transcript area), each still holding its own working `on(click)`
+    /// handlers, so a tap meant for whatever was under the stray copy
+    /// hit "Run benchmark" instead. Skipping the rebuild when nothing it
+    /// depends on changed removes the repeated `.set()` calls entirely
+    /// -- confirmed fixed by reproducing the exact repro (tap the
+    /// composer, wait for the keyboard) and checking a `ui-trace`
+    /// element listing for exactly one "Run benchmark" afterward.
     ///
-    /// **Also the trigger for the keyboard diagnostics capture** (RUST.md's
+    /// Also two things downstream of the same `ime_bottom` transition:
-    /// P0 box): the IME resizing the surface is exactly the case the
+    /// **the keyboard phase's own confirmation signal** (`ime_state`'s
-    /// previous commit found wiped text, and Iris needs a way to get a
+    /// doc -- the platform's own answer for whether the IME actually
-    /// report off the phone even if that (or some other keyboard-triggered
+    /// opened or closed, rather than assumed from having called
-    /// regression) is still happening on the build she is holding --
+    /// `show_ime`/`hide_ime`), and **the trigger for the keyboard
-    /// `capture_keyboard_diagnostics` below fires ~500ms after the
+    /// diagnostics capture** (RUST.md's P0 box): the IME resizing the
-    /// keyboard becomes visible, once per keyboard opening, and shows its
+    /// surface is exactly the case a previous commit found wiped text,
-    /// report in a plain overlay view that draws independently of
+    /// and Iris needs a way to get a report off the phone even if that
-    /// whatever iris itself is doing.
+    /// (or some other keyboard-triggered regression) is still happening
+    /// on the build she is holding -- `capture_keyboard_diagnostics`
+    /// below fires ~500ms after the keyboard becomes visible, once per
+    /// keyboard opening, and shows its report in a plain overlay view
+    /// that draws independently of whatever iris itself is doing.
     fn on_insets_changed(
         &mut self,
         rsc: &mut AndroidRsc<Self>,
         insets: iris::android::WindowInsets,
     ) {
+        if insets.top != self.last_top_pad {
+            self.last_top_pad = insets.top;
             let controls = bench_controls(rsc, insets.top);
             (self.top_bar)(rsc).set(controls);
+        }
+
+        // The composer bar sits directly on whichever of the IME or the
+        // navigation bar is currently the bottom of usable space -- see
+        // `transcript_ui::composer::Composer::set_bottom_inset`'s doc.
+        // `ime_bottom` already exceeds the plain nav-bar inset whenever the
+        // keyboard covers it, so the larger of the two is always the right
+        // answer without needing to know which is currently showing.
+        if let Some(screen) = &self.screen {
+            screen
+                .composer
+                .set_bottom_inset(rsc, insets.bottom.max(insets.ime_bottom));
+        }
 
         let ime_visible = insets.ime_bottom > 0.0;
+
+        let mut ime = self.ime_state.lock().unwrap();
+        if ime_visible && !ime.visible {
+            ime.shown_events += 1;
+        }
+        if !ime_visible && ime.visible {
+            ime.hidden_events += 1;
+        }
+        ime.visible = ime_visible;
+        drop(ime);
+
         if ime_visible && !self.keyboard_was_visible {
             self.keyboard_was_visible = true;
             let redraw = rsc.tasks.redraw_handle();
@@ -466,10 +582,10 @@ impl BenchClient {
         }
     }
 
-    /// P0's scripted run: `BenchRun.kt`'s scroll loop, then its streaming
+    /// RUST.md's "Benchmark v2": fling, then stream (unchanged from v1),
-    /// phase, then the report -- run in-process for the same reason that
+    /// then type, then keyboard, then the report -- run in-process for the
-    /// file's own doc gives (no usable system tracing on a real phone, no
+    /// same reason `BenchRun.kt`'s own doc gives (no usable system tracing
-    /// agent that can drive one).
+    /// on a real phone, no agent that can drive one).
     fn start_benchmark(&mut self, rsc: &mut Rsc) {
         if self.running {
             log::info!("iris bench report: already running");
@@ -482,35 +598,21 @@ impl BenchClient {
         let redraw = rsc.tasks.redraw_handle();
         let platform = self.platform.clone();
         let stream_tail = self.stream_tail.clone();
+        let ime_state = self.ime_state.clone();
+        let refresh_hz = platform
+            .as_ref()
+            .and_then(|p| p.refresh_rate_hz())
+            .unwrap_or(60.0);
         let cpu_start = process_cpu_ms();
+        let run_started_at = Instant::now();
 
         rsc.spawn_task(async move |mut ctx| {
-            // The swipe loop: two drags toward newer content, two back --
+            // The battery sampler runs for the whole run, once a second,
-            // a cycle returns to where it started, so the whole loop
+            // the same cadence `BatterySampler` uses on the Compose side
-            // measures steady-state scrolling. `BenchRun.kt`'s own
+            // -- via its own JNI-attached thread, not `ctx.update`, since
-            // comment on this shape.
+            // a sample needs no widget-tree access.
-            for _ in 0..CYCLES {
-                for delta in [SWIPE_PX, SWIPE_PX, -SWIPE_PX, -SWIPE_PX] {
-                    animate_scroll(&mut ctx, &redraw, delta, SWIPE_MS).await;
-                    tokio::time::sleep(Duration::from_millis(SWIPE_PAUSE_MS)).await;
-                }
-            }
-
-            // Pinned to the newest end before streaming starts, matching
-            // `stream-bench.sh`'s "Jump to latest" tap.
-            ctx.update(|state: &mut BenchClient, rsc| {
-                if let Some(screen) = &state.screen {
-                    (screen.list)(rsc).jump_to_end();
-                }
-            });
-            redraw.request_redraw();
-
-            // The battery sampler runs concurrently with the streaming
-            // phase, once a second, the same cadence `BatterySampler` uses
-            // on the Compose side -- via its own JNI-attached thread, not
-            // `ctx.update`, since a sample needs no widget-tree access.
             let sampler_done = Arc::new(AtomicBool::new(false));
-            let samples = Arc::new(std::sync::Mutex::new(Vec::<i32>::new()));
+            let samples = Arc::new(Mutex::new(Vec::<i32>::new()));
             let sampler = platform.clone().map(|platform| {
                 let done = sampler_done.clone();
                 let samples = samples.clone();
@@ -524,27 +626,10 @@ impl BenchClient {
                 })
             });
 
-            let total = (STREAM_EVENTS_PER_SEC * STREAM_SECONDS) as usize;
+            let travel = run_fling_phase(&mut ctx, &redraw).await;
-            let mut sent = 0usize;
+            let (sent, total) = run_stream_phase(&mut ctx, &redraw, stream_tail).await;
-            for event in stream_tail.into_iter().take(total) {
+            run_type_phase(&mut ctx, &redraw, &platform).await;
-                ctx.update(move |state: &mut BenchClient, rsc| {
+            let keyboard = run_keyboard_phase(&mut ctx, &platform, &ime_state).await;
-                    let old_items = state.items.clone();
-                    state.items = fold_event(&state.items, &event);
-                    match &state.screen {
-                        // The path P0 asked to measure: update only the
-                        // row(s) that changed instead of rebuilding all
-                        // ~3,200 of them per event.
-                        Some(screen) => screen.apply(rsc, &old_items, &state.items),
-                        None => state.rebuild_transcript(rsc),
-                    }
-                });
-                redraw.request_redraw();
-                sent += 1;
-                tokio::time::sleep(Duration::from_millis(1000 / STREAM_EVENTS_PER_SEC)).await;
-            }
-            // Lets the last few deltas land and draw before the report is
-            // read -- `BenchRun.kt`'s own closing delay.
-            tokio::time::sleep(Duration::from_millis(300)).await;
 
             sampler_done.store(true, Ordering::Relaxed);
             if let Some(sampler) = sampler {
@@ -553,7 +638,10 @@ impl BenchClient {
             let battery = battery_line(&samples.lock().unwrap());
             let cpu_line = match (cpu_start, process_cpu_ms()) {
                 (Some(start), Some(end)) => {
-                    format!("  process CPU time over this run: {}ms", end.saturating_sub(start))
+                    format!(
+                        "  process CPU time over this run: {}ms",
+                        end.saturating_sub(start)
+                    )
                 }
                 _ => "  process CPU time over this run: unavailable".to_string(),
             };
@@ -561,19 +649,61 @@ impl BenchClient {
                 Some(kb) => format!("  peak RSS: {kb}kB"),
                 None => "  peak RSS: unavailable (/proc/self/status unreadable)".to_string(),
             };
+            let total_seconds = run_started_at.elapsed().as_secs_f64();
 
             ctx.update(move |state: &mut BenchClient, rsc| {
                 state.running = false;
-                let scroll_line = format!(
+                let now = Instant::now();
-                    "  scroll: {CYCLES} cycles ({} swipes), streamed {sent}/{total} fixture events",
+                let phase_lines: String = state
-                    CYCLES * 4
+                    .android_state()
-                );
+                    .frame_report
-                let frames_line = match state.android_state().frame_report.report() {
+                    .phase_stats(now, refresh_hz)
-                    Some(stats) => format!("{stats}"),
+                    .iter()
-                    None => "no frames recorded".to_string(),
+                    .map(|p| format!("{p}\n"))
+                    .collect();
+                let per_phase = if phase_lines.is_empty() {
+                    String::new()
+                } else {
+                    format!("per phase:\n{phase_lines}\n")
                 };
+                let frames_block = match state.android_state().frame_report.report() {
+                    Some(stats) => {
+                        let (late, late_pct) =
+                            state.android_state().frame_report.late_at_hz(refresh_hz);
+                        format!(
+                            "frames:\n  {} frames over {:.1}s at {:.0}Hz ({:.1}ms budget)\n  \
+                             late: {late} ({late_pct:.1}%)\n  total  p50 {:.1}ms  p90 {:.1}ms  \
+                             p99 {:.1}ms\n  worst {:.1}ms\n  cpu_p50 {:.1}ms  gpu_wait_p50 {:.1}ms",
+                            stats.total_frames,
+                            total_seconds,
+                            refresh_hz,
+                            1000.0 / refresh_hz as f64,
+                            stats.p50.as_secs_f64() * 1000.0,
+                            stats.p90.as_secs_f64() * 1000.0,
+                            stats.p99.as_secs_f64() * 1000.0,
+                            stats.worst.as_secs_f64() * 1000.0,
+                            stats.cpu_p50.as_secs_f64() * 1000.0,
+                            stats.gpu_wait_p50.as_secs_f64() * 1000.0,
+                        )
+                    }
+                    None => "frames:\n  no frames recorded".to_string(),
+                };
+                let scroll_line = format!(
+                    "  scroll: {LEGACY_CYCLES} cycles ({} swipes, legacy tween), streamed \
+                     {sent}/{total} fixture events",
+                    LEGACY_CYCLES * 4
+                );
+                let fling_line = format!(
+                    "  fling: {FLING_COUNT} flings out + {FLING_COUNT} back at \
+                     {FLING_VELOCITY_PX_S}px/s, travel {travel}"
+                );
+                let type_line = format!(
+                    "  type: {} characters inserted then deleted, one per {TYPE_CHAR_MS}ms",
+                    TYPE_TEXT.chars().count()
+                );
                 let report = format!(
-                    "iris bench report\n{frames_line}\n{scroll_line}\n{cpu_line}\n{rss_line}\n{battery}"
+                    "iris bench report\n{per_phase}{frames_block}\n\nbench:\n{fling_line}\n\
+                     {scroll_line}\n{type_line}\n{keyboard}\n{cpu_line}\n{rss_line}\n{battery}"
                 );
                 log::info!("iris bench report: {report}");
                 state.report_display.edit(rsc).set(&report);
@@ -584,26 +714,287 @@ impl BenchClient {
     }
 }
 
-/// Moves `List::scroll` by `total_px` over `duration_ms`, in ~60Hz steps,
+/// Runs `f` against the real `BenchClient`/`Rsc` on the main thread (the
-/// so the swipe is many rendered frames rather than one jump -- the same
+/// same `ctx.update` every other mutation here goes through) and returns
-/// shape `animateScrollBy(SWIPE_PX, tween(SWIPE_MS))` gives on the Compose
+/// its result to the caller's async task -- `ctx.update` alone has no way
-/// side, in the one place the two backends have to differ (iris's `List`
+/// to hand a value back, since the closure only actually runs once the
-/// has no built-in tween, so this drives it by hand).
+/// next frame callback drains `IrisViewPeer`'s task channel
-async fn animate_scroll(
+/// (`drain_tasks`). **Must call `redraw.request_redraw()` itself, right
+/// after enqueueing** -- `ctx.update` only ever pushes onto a channel;
+/// nothing drains it until something schedules the frame callback that
+/// calls `drain_tasks`, and a caller relying on some *earlier*,
+/// already-in-flight `request_redraw()` to cover a *later* `ctx.update`
+/// deadlocks the moment that earlier callback has already fired and
+/// drained everything queued before this call existed. Cost a real hang
+/// in this file's first version of the fling phase: every loop iteration
+/// after the first sat forever with nothing scheduled to drain it.
+/// Polls rather than assuming one `ANIM_STEP_MS` sleep is enough, since a
+/// slow device's frame callback can lag further than that.
+async fn read_from_state<T, F>(
     ctx: &mut iris::task::TaskCtx<Rsc>,
-    redraw: &Arc<dyn iris::task::RequestRedraw>,
+    redraw: &Arc<dyn RequestRedraw>,
-    total_px: f32,
+    f: F,
-    duration_ms: u64,
+) -> T
-) {
+where
-    let steps = (duration_ms / ANIM_STEP_MS).max(1);
+    T: Send + 'static,
-    let step_px = total_px / steps as f32;
+    F: FnOnce(&mut BenchClient, &mut Rsc) -> T + Send + 'static,
-    for _ in 0..steps {
+{
+    let (tx, rx) = std::sync::mpsc::channel();
     ctx.update(move |state: &mut BenchClient, rsc| {
-            if let Some(screen) = &state.screen {
+        let _ = tx.send(f(state, rsc));
-                (screen.list)(rsc).scroll(step_px);
-            }
     });
     redraw.request_redraw();
+    loop {
+        if let Ok(value) = rx.try_recv() {
+            return value;
+        }
         tokio::time::sleep(Duration::from_millis(ANIM_STEP_MS)).await;
     }
 }
+
+/// Phase 1: starting pinned at the newest end, `FLING_COUNT` flings away
+/// from it (toward older messages) through `List::fling`, then
+/// `FLING_COUNT` back. Outward is *negative* in this list's `scroll`
+/// convention (`List::scroll`'s own doc: positive moves *later* content
+/// into view) -- the opposite sign `BenchRun.kt`'s `runFlingPhase` uses,
+/// since `TranscriptList`'s `LazyColumn` and this list define "positive"
+/// the other way around; the two apps' *travel* is still directly
+/// comparable because both report it as a row index + pixel offset, not a
+/// signed distance.
+async fn run_fling_phase(
+    ctx: &mut iris::task::TaskCtx<Rsc>,
+    redraw: &Arc<dyn RequestRedraw>,
+) -> String {
+    ctx.update(|state: &mut BenchClient, _rsc| {
+        state.android_state_mut().frame_report.mark_phase("fling");
+    });
+    ctx.update(|state: &mut BenchClient, rsc| {
+        if let Some(screen) = &state.screen {
+            (screen.list)(rsc).jump_to_end();
+        }
+    });
+    redraw.request_redraw();
+    // Lets the next frame's `repair_anchor` resolve `jump_to_end`'s
+    // `anchor = None` into a real slot before `start` is read.
+    tokio::time::sleep(Duration::from_millis(ANIM_STEP_MS * 2)).await;
+    let start = read_anchor_position(ctx, redraw).await;
+
+    for _ in 0..FLING_COUNT {
+        ctx.update(|state: &mut BenchClient, rsc| {
+            if let Some(screen) = &state.screen {
+                (screen.list)(rsc).fling(-FLING_VELOCITY_PX_S);
+            }
+        });
+        redraw.request_redraw();
+        wait_for_fling_settle(ctx, redraw).await;
+        tokio::time::sleep(Duration::from_millis(FLING_PAUSE_MS)).await;
+    }
+    let outward = read_anchor_position(ctx, redraw).await;
+
+    for _ in 0..FLING_COUNT {
+        ctx.update(|state: &mut BenchClient, rsc| {
+            if let Some(screen) = &state.screen {
+                (screen.list)(rsc).fling(FLING_VELOCITY_PX_S);
+            }
+        });
+        redraw.request_redraw();
+        wait_for_fling_settle(ctx, redraw).await;
+        tokio::time::sleep(Duration::from_millis(FLING_PAUSE_MS)).await;
+    }
+    let end = read_anchor_position(ctx, redraw).await;
+
+    format!("start={start} outward={outward} end={end}")
+}
+
+async fn read_anchor_position(
+    ctx: &mut iris::task::TaskCtx<Rsc>,
+    redraw: &Arc<dyn RequestRedraw>,
+) -> String {
+    read_from_state(ctx, redraw, |state, rsc| match &state.screen {
+        Some(screen) => (screen.list)(rsc).anchor_position_display(),
+        None => "idx=none".to_string(),
+    })
+    .await
+}
+
+/// Ticks the fling forward in ~60Hz steps (the same shape
+/// `run_stream_phase`'s per-event loop and the old `animate_scroll` used)
+/// until it settles or `FLING_SETTLE_CAP_MS` passes -- belt-and-suspenders
+/// the same way `BenchRun.kt`'s own `waitForSettle` is, since a fling's
+/// own spline-decided `duration()` already caps how long it can run.
+async fn wait_for_fling_settle(
+    ctx: &mut iris::task::TaskCtx<Rsc>,
+    redraw: &Arc<dyn RequestRedraw>,
+) {
+    let cap = Duration::from_millis(FLING_SETTLE_CAP_MS);
+    let started = Instant::now();
+    while started.elapsed() < cap {
+        let still_scrolling = read_from_state(ctx, redraw, |state, rsc| match &state.screen {
+            Some(screen) => (screen.list)(rsc).tick_fling(Instant::now()),
+            None => false,
+        })
+        .await;
+        if !still_scrolling {
+            return;
+        }
+        tokio::time::sleep(Duration::from_millis(ANIM_STEP_MS)).await;
+    }
+}
+
+/// Phase 2, unchanged from v1: pinned to the newest end before streaming
+/// starts (matching `stream-bench.sh`'s "Jump to latest" tap), then
+/// `STREAM_EVENTS_PER_SEC * STREAM_SECONDS` fixture events replayed
+/// through the real `fold_event`/`TranscriptScreen::apply` path. Returns
+/// `(sent, total)`.
+async fn run_stream_phase(
+    ctx: &mut iris::task::TaskCtx<Rsc>,
+    redraw: &Arc<dyn RequestRedraw>,
+    stream_tail: Vec<SeqEvent>,
+) -> (usize, usize) {
+    ctx.update(|state: &mut BenchClient, _rsc| {
+        state.android_state_mut().frame_report.mark_phase("stream");
+    });
+    ctx.update(|state: &mut BenchClient, rsc| {
+        if let Some(screen) = &state.screen {
+            (screen.list)(rsc).jump_to_end();
+        }
+    });
+    redraw.request_redraw();
+
+    let total = (STREAM_EVENTS_PER_SEC * STREAM_SECONDS) as usize;
+    let mut sent = 0usize;
+    for event in stream_tail.into_iter().take(total) {
+        ctx.update(move |state: &mut BenchClient, rsc| {
+            let old_items = state.items.clone();
+            state.items = fold_event(&state.items, &event);
+            match &state.screen {
+                Some(screen) => screen.apply(rsc, &old_items, &state.items),
+                None => state.rebuild_transcript(rsc),
+            }
+        });
+        redraw.request_redraw();
+        sent += 1;
+        tokio::time::sleep(Duration::from_millis(1000 / STREAM_EVENTS_PER_SEC)).await;
+    }
+    // Lets the last few deltas land and draw before the next phase starts
+    // -- `BenchRun.kt`'s own closing delay.
+    tokio::time::sleep(Duration::from_millis(300)).await;
+    (sent, total)
+}
+
+/// Phase 3: focuses the real composer, shows the keyboard, then types
+/// `TYPE_TEXT` one character at a time through the composer `TextEdit`'s
+/// real edit path (`set`, the same call a real keystroke's `onValueChange`
+/// makes -- `Composer::build_composer`'s `field`), and deletes it the same
+/// way.
+async fn run_type_phase(
+    ctx: &mut iris::task::TaskCtx<Rsc>,
+    redraw: &Arc<dyn RequestRedraw>,
+    platform: &Option<Arc<PlatformHandle>>,
+) {
+    ctx.update(|state: &mut BenchClient, _rsc| {
+        state.android_state_mut().frame_report.mark_phase("type");
+    });
+    ctx.update(|state: &mut BenchClient, rsc| {
+        if let Some(screen) = &state.screen {
+            (screen.list)(rsc).jump_to_end();
+            state.set_focus(Some(screen.composer.field));
+        }
+    });
+    redraw.request_redraw();
+    if let Some(p) = platform {
+        p.show_ime();
+    }
+    // Lets focus and the keyboard's opening animation land before typing
+    // starts, so the frames this phase records are the wrap/reflow it is
+    // measuring, not the keyboard opening -- `BenchRun.kt`'s own delay.
+    tokio::time::sleep(Duration::from_millis(300)).await;
+
+    let mut typed = String::new();
+    for ch in TYPE_TEXT.chars() {
+        typed.push(ch);
+        let text = typed.clone();
+        ctx.update(move |state: &mut BenchClient, rsc| {
+            if let Some(screen) = &state.screen {
+                screen.composer.field.edit(rsc).set(&text);
+            }
+        });
+        redraw.request_redraw();
+        tokio::time::sleep(Duration::from_millis(TYPE_CHAR_MS)).await;
+    }
+    tokio::time::sleep(Duration::from_millis(200)).await;
+    while !typed.is_empty() {
+        typed.pop();
+        let text = typed.clone();
+        ctx.update(move |state: &mut BenchClient, rsc| {
+            if let Some(screen) = &state.screen {
+                screen.composer.field.edit(rsc).set(&text);
+            }
+        });
+        redraw.request_redraw();
+        tokio::time::sleep(Duration::from_millis(TYPE_CHAR_MS)).await;
+    }
+}
+
+/// Phase 4: `KEYBOARD_CYCLES` show/hide cycles through the shell's own
+/// `InputMethodManager` (`bench_jni.rs`'s `show_ime`/`hide_ime`), each
+/// confirmed by `on_insets_changed`'s real `ime_bottom` transition rather
+/// than assumed from the JNI call having returned -- `ImeState`'s doc.
+/// "keyboard: could not be shown" if the platform never confirms it even
+/// once, per UI_RULES.md ("design the unknown/failed state before the
+/// answer's").
+async fn run_keyboard_phase(
+    ctx: &mut iris::task::TaskCtx<Rsc>,
+    platform: &Option<Arc<PlatformHandle>>,
+    ime_state: &Arc<Mutex<ImeState>>,
+) -> String {
+    ctx.update(|state: &mut BenchClient, _rsc| {
+        state
+            .android_state_mut()
+            .frame_report
+            .mark_phase("keyboard");
+    });
+    let mut shown = 0;
+    let mut hidden = 0;
+    for _ in 0..KEYBOARD_CYCLES {
+        let before_shown = ime_state.lock().unwrap().shown_events;
+        if let Some(p) = platform {
+            p.show_ime();
+        }
+        tokio::time::sleep(Duration::from_millis(KEYBOARD_WAIT_MS)).await;
+        if ime_state.lock().unwrap().shown_events > before_shown {
+            shown += 1;
+        }
+
+        let before_hidden = ime_state.lock().unwrap().hidden_events;
+        if let Some(p) = platform {
+            p.hide_ime();
+        }
+        tokio::time::sleep(Duration::from_millis(KEYBOARD_WAIT_MS)).await;
+        if ime_state.lock().unwrap().hidden_events > before_hidden {
+            hidden += 1;
+        }
+    }
+    if shown == 0 {
+        format!("  keyboard: could not be shown ({KEYBOARD_CYCLES} attempts, 0 confirmed visible)")
+    } else {
+        format!(
+            "  keyboard: shown {shown}/{KEYBOARD_CYCLES}, hidden {hidden}/{KEYBOARD_CYCLES} \
+             (confirmed via on_insets_changed)"
+        )
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::TYPE_TEXT;
+
+    /// `BenchRun.kt`'s own `TYPE_TEXT` is verified `.length == 600`; this
+    /// is the same string, so it has to match exactly or the two apps'
+    /// type phases stop typing the same content -- RUST.md's "Benchmark
+    /// v2" spec is one shared string for both.
+    #[test]
+    fn type_text_is_exactly_600_characters() {
+        assert_eq!(TYPE_TEXT.chars().count(), 600);
+    }
+}

iris/android-app/src/bench_jni.rs

@@ -1,11 +1,14 @@
 //! JNI calls the `bench` feature needs that go through the shell's own
 //! Java side rather than anything `iris`/`android-view` already wraps:
 //! `BatteryManager.getIntProperty(BATTERY_PROPERTY_CURRENT_NOW)` for the
-//! per-second battery sample, and `ClipboardManager.setPrimaryClip` for
+//! per-second battery sample, `ClipboardManager.setPrimaryClip` for the
-//! the "Copy report" control (P0's iris half, docs/RUST.md). Neither is
+//! "Copy report" control (P0's iris half, docs/RUST.md), and -- added for
-//! part of `android_view::context`'s own `Context`/`Resources` wrappers
+//! RUST.md's "Benchmark v2" -- `Display.getRefreshRate()` for the phase
-//! (that file's own `// TODO: more methods?`), so this calls them
+//! report's real late-frame budget and `InputMethodManager.
-//! directly rather than growing that crate's wrapper for two one-off
+//! showSoftInput`/`hideSoftInputFromWindow` for the keyboard phase. None
+//! of these are part of `android_view::context`'s own `Context`/
+//! `Resources` wrappers (that file's own `// TODO: more methods?`), so
+//! this calls them directly rather than growing that crate's wrapper for
 //! calls this crate alone needs.
 //!
 //! Holds its own `JavaVM` + `GlobalRef` to the view (handed in through
@@ -132,6 +135,94 @@ impl PlatformHandle {
         Some(())
     }
 
+    /// The display's own refresh rate in Hz (`View::getDisplay()` ->
+    /// `Display::getRefreshRate()`), for RUST.md's "Benchmark v2": late
+    /// frames are judged against *this* device's real budget, not an
+    /// assumed 60Hz -- a 90Hz or 120Hz phone would otherwise call frames
+    /// "late" that met their own faster deadline. `None` if the view is
+    /// not yet attached to a window (`getDisplay` returns `null`) or the
+    /// platform reports a non-positive rate, which is not a real answer
+    /// either.
+    pub fn refresh_rate_hz(&self) -> Option<f32> {
+        let mut guard = self.vm.attach_current_thread().ok()?;
+        let env: &mut JNIEnv = &mut guard;
+        let display = env
+            .call_method(
+                self.view.as_obj(),
+                "getDisplay",
+                "()Landroid/view/Display;",
+                &[],
+            )
+            .ok()?
+            .l()
+            .ok()?;
+        if display.is_null() {
+            return None;
+        }
+        let rate = env
+            .call_method(&display, "getRefreshRate", "()F", &[])
+            .ok()?
+            .f()
+            .ok()?;
+        if rate > 0.0 { Some(rate) } else { None }
+    }
+
+    /// `InputMethodManager.showSoftInput(view, 0)` -- the keyboard phase's
+    /// own show, called directly rather than through the focus-driven
+    /// `pending_show_keyboard` path `android/view.rs` uses for a real tap,
+    /// since RUST.md's "Benchmark v2" spec asks for this "through the
+    /// shell's InputMethodManager" independent of focus state. `true` only
+    /// if the platform itself reports the request succeeded -- whether the
+    /// IME actually became visible is confirmed separately, from
+    /// `on_insets_changed`, per UI_RULES.md ("never present an inferred
+    /// value as a measured one").
+    pub fn show_ime(&self) -> bool {
+        self.try_toggle_ime(true).unwrap_or(false)
+    }
+
+    /// `InputMethodManager.hideSoftInputFromWindow(windowToken, 0)`.
+    pub fn hide_ime(&self) -> bool {
+        self.try_toggle_ime(false).unwrap_or(false)
+    }
+
+    fn try_toggle_ime(&self, show: bool) -> Option<bool> {
+        let mut guard = self.vm.attach_current_thread().ok()?;
+        let env: &mut JNIEnv = &mut guard;
+        let context = self.context(env)?;
+        let imm = self.system_service(env, &context, "input_method")?;
+        if show {
+            env.call_method(
+                &imm,
+                "showSoftInput",
+                "(Landroid/view/View;I)Z",
+                &[JValue::Object(self.view.as_obj()), JValue::Int(0)],
+            )
+            .ok()?
+            .z()
+            .ok()
+        } else {
+            let token = env
+                .call_method(
+                    self.view.as_obj(),
+                    "getWindowToken",
+                    "()Landroid/os/IBinder;",
+                    &[],
+                )
+                .ok()?
+                .l()
+                .ok()?;
+            env.call_method(
+                &imm,
+                "hideSoftInputFromWindow",
+                "(Landroid/os/IBinder;I)Z",
+                &[JValue::Object(&token), JValue::Int(0)],
+            )
+            .ok()?
+            .z()
+            .ok()
+        }
+    }
+
     /// Shows `report` in the shell's plain-view diagnostics overlay
     /// (`IrisView.showDiagnosticsOverlay`) -- a real `TextView` plus Copy
     /// and Close controls, added over whatever iris itself is drawing

iris/core/src/primitive/texture.rs

@@ -141,6 +141,27 @@ impl Textures {
         self.updates.push(Update::Patch(handle.slot, rect));
     }
 
+    /// Forget every image, page and pending update -- what a genuinely new
+    /// GPU device needs alongside [`crate::render::atlas::GlyphAtlas::
+    /// clear`], which this module's own doc references: every slot number
+    /// and every queued [`Update`] here describes the *old* device's
+    /// textures (an `Update::Push`/`Update::Patch` already drained into a
+    /// renderer that no longer exists is gone for good, and a fresh
+    /// `UiRenderNode`'s own texture manager starts with none of them
+    /// applied), so nothing is lost by starting this bookkeeping over too.
+    /// Any `TextureHandle` a caller still holds across the reset (none in
+    /// the transcript screen this reset is wired up for today -- confirmed
+    /// by grep, the only standalone (non-atlas) image anywhere in this
+    /// workspace is `iris/widget/image.rs`'s `Image`, used by the separate
+    /// `tabs-ui` example) is left pointing at a slot this instance no
+    /// longer recognises and needs reinserting via `add`/`add_page` again
+    /// -- the same pre-existing gap a renderer restart already left for
+    /// such a handle before this method existed, just named rather than
+    /// silent now.
+    pub fn reset(&mut self) {
+        *self = Self::new();
+    }
+
     pub fn free(&mut self) {
         for (kind, idx) in self.recv.try_iter() {
             self.images[idx as usize] = None;

iris/core/src/render/atlas.rs

@@ -173,6 +173,25 @@ impl GlyphAtlas {
     pub fn glyph_count(&self) -> usize {
         self.entries.len()
     }
+
+    /// Forget every page and every rasterised entry -- what a genuinely new
+    /// GPU device needs (`android::view::IrisViewPeer::surface_changed`'s
+    /// "not already live" branch, e.g. after backgrounding): the pages this
+    /// atlas remembers are `TextureHandle`s into the *old* device's
+    /// textures, which no longer exist, and every `GlyphEntry`'s `uv_min`/
+    /// `uv_max`/`layer` point into them. Without this, a glyph already
+    /// cached here is treated as "already placed" and never re-inserted
+    /// into the fresh (empty) atlas the new renderer actually has --
+    /// exactly the "rectangles stay, glyphs disappear" bug the resize path
+    /// (`AndroidRenderer::resize`) was built to avoid for the reuse case;
+    /// this is its counterpart for the case where the renderer really is
+    /// new. Dropping `pages` also drops its `TextureHandle`s, which send a
+    /// free message back through their `Textures`; see `Textures::reset`'s
+    /// doc for why that is harmless here.
+    pub fn clear(&mut self) {
+        self.pages.clear();
+        self.entries.clear();
+    }
 }
 
 fn fits(page: &Page, need_w: u32, need_h: u32) -> bool {

iris/core/src/render/frame_report.rs

@@ -1,15 +1,87 @@
-use std::time::Duration;
+use std::time::{Duration, Instant};
 
 /// The frame budget `dumpsys gfxinfo` also uses to call a frame "janky": the
 /// 60Hz vsync period. Kept as the same threshold so a percentage from this
-/// report and a percentage from `gfxinfo` mean the same thing.
+/// report and a percentage from `gfxinfo` mean the same thing. Only a
+/// fallback now that a caller can read the display's real refresh rate
+/// (`report_at_hz`/`mark_phase`'s callers) -- most devices are 60Hz, but a
+/// 90Hz or 120Hz phone judged against this constant would call every frame
+/// "late" that merely met its own, faster budget.
 pub const JANK_THRESHOLD: Duration = Duration::from_nanos(16_666_667);
 
 /// Enough frames for several minutes of scrolling before the oldest ones
 /// start being overwritten -- the same "diagnostic, not a log" sizing
 /// `FrameStats.kt`'s `CAP` uses on the Compose side, chosen independently
 /// here since a `Duration` is smaller than the six `Long` arrays it keeps.
-const RING_CAPACITY: usize = 4096;
+/// Bumped from 4096 for RUST.md's "Benchmark v2": a fling+stream+type+
+/// keyboard run is ~6,500+ frames on the Compose side, comfortably under
+/// this so `phase_stats` never has to report a phase as partially evicted.
+const RING_CAPACITY: usize = 16384;
+
+/// One `mark_phase` call: the wall-clock instant and the (0-based,
+/// never-reset-by-`reset`-except-at-`reset`-time) absolute frame index at
+/// which a phase began -- `phase_stats` slices `index_ring` against this to
+/// find which recorded samples belong to which phase, since the ring
+/// itself only keeps the most recent `RING_CAPACITY` samples' *values*,
+/// not which phase they were in.
+struct PhaseMark {
+    name: String,
+    start_index: u64,
+    start_at: Instant,
+}
+
+/// One phase's own slice of a report -- RUST.md's "Benchmark v2" spec's
+/// "per-phase blocks in `FrameReport`... frames, late count/percent...
+/// p50/p90/p99, worst, duration". `Display` matches the shape
+/// `docs/bench/compose-phone-v2-2026-09-06.md`'s report already uses, so
+/// the two apps' reports read the same way side by side.
+pub struct PhaseStats {
+    pub name: String,
+    /// How many frames were recorded during this phase in total -- may
+    /// exceed `late + (samples counted)` if some of this phase's frames
+    /// have since been evicted from the ring by a very long run; that
+    /// case is named in the `Display` rather than silently under-counted.
+    pub frames: u64,
+    pub duration: Duration,
+    pub late: u64,
+    pub late_percent: f64,
+    pub p50: Duration,
+    pub p90: Duration,
+    pub p99: Duration,
+    pub worst: Duration,
+    /// `false` if this phase's frame count exceeds how many samples of it
+    /// are still in the ring -- the percentiles above are then computed
+    /// over whatever survived, not the whole phase. UI_RULES.md: this is
+    /// the "we don't fully know" state, named rather than folded silently
+    /// into a number that looks exact.
+    pub complete: bool,
+}
+
+impl std::fmt::Display for PhaseStats {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        writeln!(
+            f,
+            "  {}: {} frames over {:.1}s{}",
+            self.name,
+            self.frames,
+            self.duration.as_secs_f64(),
+            if self.complete {
+                ""
+            } else {
+                " (ring evicted some of this phase)"
+            },
+        )?;
+        writeln!(f, "    late: {} ({:.1}%)", self.late, self.late_percent)?;
+        writeln!(
+            f,
+            "      total  p50 {:.1}ms  p90 {:.1}ms  p99 {:.1}ms",
+            self.p50.as_secs_f64() * 1000.0,
+            self.p90.as_secs_f64() * 1000.0,
+            self.p99.as_secs_f64() * 1000.0,
+        )?;
+        write!(f, "    worst {:.1}ms", self.worst.as_secs_f64() * 1000.0)
+    }
+}
 
 /// A per-frame wall-time report iris keeps of itself, because `dumpsys
 /// gfxinfo` cannot see a `SurfaceView`'s own GPU-drawn frames at all
@@ -41,6 +113,11 @@ pub struct FrameReport {
     /// "Where iris's frame time goes" CPU/GPU split, added 2026-09-05).
     /// `ring[i] - submit_ring[i]` is that frame's `redraw_to_submit` half.
     submit_ring: Box<[Duration; RING_CAPACITY]>,
+    /// The absolute (0-based, since the last `reset`) frame index each
+    /// `ring`/`submit_ring` slot's sample belongs to -- what `phase_stats`
+    /// slices against `PhaseMark::start_index` to tell which recorded
+    /// frames fall in which phase.
+    index_ring: Box<[u64; RING_CAPACITY]>,
     /// How many of `ring`'s slots hold a real sample -- saturates at
     /// `RING_CAPACITY`, unlike `total_frames` below which keeps counting.
     len: usize,
@@ -50,6 +127,12 @@ pub struct FrameReport {
     /// correct even once the ring itself only holds the most recent frames.
     total_frames: u64,
     janky_frames: u64,
+    /// `mark_phase` calls since the last `reset`, oldest first -- see
+    /// `phase_stats`. Empty on an ordinary run that never calls
+    /// `mark_phase`, so `phase_stats` returns an empty `Vec` and a caller
+    /// prints no "per phase:" section at all, matching RUST.md's "empty/
+    /// absent on an ordinary 'Copy' press, which never marks a phase."
+    phases: Vec<PhaseMark>,
 }
 
 /// One resolved reading. `Display` is the log line both the "Frame report"
@@ -107,10 +190,12 @@ impl FrameReport {
         Self {
             ring: Box::new([Duration::ZERO; RING_CAPACITY]),
             submit_ring: Box::new([Duration::ZERO; RING_CAPACITY]),
+            index_ring: Box::new([0; RING_CAPACITY]),
             len: 0,
             pos: 0,
             total_frames: 0,
             janky_frames: 0,
+            phases: Vec::new(),
         }
     }
 
@@ -131,6 +216,7 @@ impl FrameReport {
     pub fn record_split(&mut self, total: Duration, submit_to_present: Duration) {
         self.ring[self.pos] = total;
         self.submit_ring[self.pos] = submit_to_present;
+        self.index_ring[self.pos] = self.total_frames;
         self.pos = (self.pos + 1) % RING_CAPACITY;
         self.len = (self.len + 1).min(RING_CAPACITY);
         self.total_frames += 1;
@@ -142,12 +228,89 @@ impl FrameReport {
     /// Clears every counter and every sample -- what the "Reset frame
     /// report" control calls, so a report covers only what was scrolled
     /// after the button was pressed (the same reason `FrameStats.kt`'s
-    /// `reset()` exists on the Compose side).
+    /// `reset()` exists on the Compose side). Also clears every phase
+    /// mark, so a fresh run starts with no "per phase:" section until it
+    /// marks one of its own.
     pub fn reset(&mut self) {
         self.len = 0;
         self.pos = 0;
         self.total_frames = 0;
         self.janky_frames = 0;
+        self.phases.clear();
+    }
+
+    /// Marks the start of a named phase at the current moment -- every
+    /// frame recorded from here until the next `mark_phase` (or `reset`)
+    /// belongs to it. RUST.md's "Benchmark v2": a scripted bench run calls
+    /// this once per phase (fling/stream/type/keyboard) so `phase_stats`
+    /// can slice one whole run's frames by what was happening during each.
+    pub fn mark_phase(&mut self, name: &str) {
+        self.phases.push(PhaseMark {
+            name: name.to_string(),
+            start_index: self.total_frames,
+            start_at: Instant::now(),
+        });
+    }
+
+    /// One [`PhaseStats`] per `mark_phase` call since the last `reset`,
+    /// oldest first. `now` closes the last phase's wall-clock span (there
+    /// is no "next phase" instant to use for it); `refresh_hz` is what
+    /// each phase's own `late`/`late_percent` is judged against, read from
+    /// the display rather than assumed -- RUST.md's "Benchmark v2": "late
+    /// count/% against the display's refresh rate."
+    pub fn phase_stats(&self, now: Instant, refresh_hz: f32) -> Vec<PhaseStats> {
+        if self.phases.is_empty() || refresh_hz <= 0.0 {
+            return Vec::new();
+        }
+        let budget = Duration::from_secs_f64(1.0 / refresh_hz as f64);
+        self.phases
+            .iter()
+            .enumerate()
+            .map(|(i, phase)| {
+                let (end_index, end_at) = match self.phases.get(i + 1) {
+                    Some(next) => (next.start_index, next.start_at),
+                    None => (self.total_frames, now),
+                };
+                let frames = end_index.saturating_sub(phase.start_index);
+                let mut samples: Vec<Duration> = (0..self.len)
+                    .filter(|&j| {
+                        let idx = self.index_ring[j];
+                        idx >= phase.start_index && idx < end_index
+                    })
+                    .map(|j| self.ring[j])
+                    .collect();
+                let complete = samples.len() as u64 >= frames;
+                if samples.is_empty() {
+                    return PhaseStats {
+                        name: phase.name.clone(),
+                        frames,
+                        duration: end_at.saturating_duration_since(phase.start_at),
+                        late: 0,
+                        late_percent: 0.0,
+                        p50: Duration::ZERO,
+                        p90: Duration::ZERO,
+                        p99: Duration::ZERO,
+                        worst: Duration::ZERO,
+                        complete,
+                    };
+                }
+                samples.sort_unstable();
+                let pct = |p: usize| samples[(samples.len() * p / 100).min(samples.len() - 1)];
+                let late = samples.iter().filter(|&&d| d > budget).count() as u64;
+                PhaseStats {
+                    name: phase.name.clone(),
+                    frames,
+                    duration: end_at.saturating_duration_since(phase.start_at),
+                    late,
+                    late_percent: 100.0 * late as f64 / samples.len() as f64,
+                    p50: pct(50),
+                    p90: pct(90),
+                    p99: pct(99),
+                    worst: *samples.last().expect("checked not empty above"),
+                    complete,
+                }
+            })
+            .collect()
     }
 
     /// `None` if nothing has been recorded since the last reset -- the
@@ -186,6 +349,28 @@ impl FrameReport {
             gpu_wait_p50: median(submit_samples),
         })
     }
+
+    /// `(late count, late percent)` over every sample still in the ring,
+    /// judged against `refresh_hz`'s own frame budget rather than the
+    /// fixed 60Hz `JANK_THRESHOLD` -- RUST.md's "Benchmark v2": "late
+    /// count/% against the display's refresh rate... print 'at N Hz (X ms
+    /// budget)' like Compose does." A separate method from `report()`
+    /// rather than a parameter on it, so `report()`'s own `janky_percent`
+    /// (and the exact-boundary test pinned to `JANK_THRESHOLD`) is
+    /// unaffected for every existing caller that never measured a real
+    /// refresh rate. `(0, 0.0)` with nothing recorded or a non-positive
+    /// `refresh_hz`.
+    pub fn late_at_hz(&self, refresh_hz: f32) -> (u64, f64) {
+        if self.len == 0 || refresh_hz <= 0.0 {
+            return (0, 0.0);
+        }
+        let budget = Duration::from_secs_f64(1.0 / refresh_hz as f64);
+        let late = self.ring[..self.len]
+            .iter()
+            .filter(|&&d| d > budget)
+            .count() as u64;
+        (late, 100.0 * late as f64 / self.len as f64)
+    }
 }
 
 impl Default for FrameReport {
@@ -296,4 +481,68 @@ mod tests {
         // same pattern here.
         assert!(stats.worst <= Duration::from_millis(5));
     }
+
+    #[test]
+    fn no_marks_means_no_phases() {
+        let mut r = FrameReport::new();
+        r.record(Duration::from_millis(5));
+        assert!(r.phase_stats(Instant::now(), 60.0).is_empty());
+    }
+
+    #[test]
+    fn phases_slice_frames_by_when_they_were_marked() {
+        let mut r = FrameReport::new();
+        r.mark_phase("a");
+        for _ in 0..5 {
+            r.record(Duration::from_millis(10)); // 10ms: late at 60Hz (16.7ms budget)... no, 10<16.7, not late
+        }
+        r.mark_phase("b");
+        for _ in 0..3 {
+            r.record(Duration::from_millis(20)); // 20ms: late at 60Hz
+        }
+        let now = Instant::now();
+        let phases = r.phase_stats(now, 60.0);
+        assert_eq!(phases.len(), 2);
+        assert_eq!(phases[0].name, "a");
+        assert_eq!(phases[0].frames, 5);
+        assert_eq!(phases[0].late, 0);
+        assert_eq!(phases[0].worst, Duration::from_millis(10));
+        assert_eq!(phases[1].name, "b");
+        assert_eq!(phases[1].frames, 3);
+        assert_eq!(phases[1].late, 3);
+        assert_eq!(phases[1].late_percent, 100.0);
+        assert_eq!(phases[1].worst, Duration::from_millis(20));
+        assert!(phases[0].complete);
+        assert!(phases[1].complete);
+    }
+
+    #[test]
+    fn the_last_phase_runs_until_now() {
+        let mut r = FrameReport::new();
+        r.mark_phase("only");
+        r.record(Duration::from_millis(1));
+        std::thread::sleep(Duration::from_millis(20));
+        let now = Instant::now();
+        let phases = r.phase_stats(now, 60.0);
+        assert_eq!(phases.len(), 1);
+        assert!(phases[0].duration >= Duration::from_millis(20));
+    }
+
+    #[test]
+    fn reset_clears_phase_marks() {
+        let mut r = FrameReport::new();
+        r.mark_phase("a");
+        r.record(Duration::from_millis(1));
+        r.reset();
+        assert!(r.phase_stats(Instant::now(), 60.0).is_empty());
+    }
+
+    #[test]
+    fn late_at_hz_uses_the_given_refresh_rate_not_the_fixed_60hz_constant() {
+        let mut r = FrameReport::new();
+        // 10ms is under 60Hz's 16.7ms budget but over 120Hz's 8.3ms one.
+        r.record(Duration::from_millis(10));
+        assert_eq!(r.late_at_hz(60.0), (0, 0.0));
+        assert_eq!(r.late_at_hz(120.0), (1, 100.0));
+    }
 }

iris/core/src/ui/render_state.rs

@@ -20,6 +20,21 @@ pub struct UiRenderState {
     resized: bool,
     draw_started: HashSet<WidgetId>,
 
+    /// The widget currently holding exclusive pointer input, if any --
+    /// `iris::sense::SensorUi::run_sensors` reads and clears this every
+    /// call. Interior mutability (a `Mutex`, not a bare `Cell`, since a
+    /// `CursorData` reaching this through an async `task_on` handler needs
+    /// `Send`/`Sync`) because `run_sensors` takes `&self` (widgets are
+    /// dispatched to, not owned, at that layer) and this render state is
+    /// the one structure both backends (winit, android-view) already hold
+    /// across frames, the same way `old_root`/`resized` are -- see
+    /// `iris::sense`'s pointer-capture doc for why a drag needs this: once
+    /// a gesture has committed to panning or selecting, every later sample
+    /// of it must reach the same widget even if the finger has moved off
+    /// whatever hit region first noticed the press. Never held across an
+    /// await or another lock -- every access here is a single get/set.
+    captured: std::sync::Mutex<Option<WidgetId>>,
+
     /// `Widget::draw` calls and `Primitives::region_mut` rewrites since the
     /// last `take_counters`. LAYOUT.md section 8's pass conditions are
     /// stated in terms of these two: an unchanged frame must cost 0 of
@@ -45,6 +60,7 @@ impl UiRenderState {
             old_root: None,
             resized: false,
             draw_started: Default::default(),
+            captured: Default::default(),
             draw_count: 0,
             region_mut_count: 0,
             mov_count: 0,
@@ -357,6 +373,13 @@ impl UiRenderState {
             active.textures.clear();
             rsc.ui_mut().textures.free();
             if undraw {
+                // A captured widget that goes away mid-gesture (List's
+                // virtualisation retiring a row, a rebuild) must not leave
+                // the pointer permanently captured by an id nothing will
+                // ever draw again -- `captured`'s own path out.
+                if *self.captured.lock().unwrap() == Some(id) {
+                    *self.captured.lock().unwrap() = None;
+                }
                 // Permanent removal: retire this widget's own move slot
                 // (the self-ownership ref taken when it was allocated) and
                 // the up-link ref it held on its parent's slot -- read from
@@ -429,6 +452,27 @@ impl UiRenderState {
         self.active.len()
     }
 
+    /// Give `id` exclusive pointer input from the next `run_sensors` call
+    /// on -- see `captured`'s field doc. Overwrites any previous capture
+    /// (a gesture that starts a new one has already decided the old one
+    /// is over).
+    pub fn capture_pointer(&self, id: WidgetId) {
+        *self.captured.lock().unwrap() = Some(id);
+    }
+
+    /// Release exclusive pointer input, if any is held -- called once
+    /// `run_sensors` has delivered the terminal `Drop` to the capturing
+    /// widget, or by that widget itself if it decides the gesture is over
+    /// some other way.
+    pub fn release_pointer(&self) {
+        *self.captured.lock().unwrap() = None;
+    }
+
+    /// The widget currently holding exclusive pointer input, if any.
+    pub fn captured_pointer(&self) -> Option<WidgetId> {
+        *self.captured.lock().unwrap()
+    }
+
     pub fn debug(&self, widgets: &Widgets, label: &str) -> impl Iterator<Item = &ActiveData> {
         self.active.iter().filter_map(move |(&id, inst)| {
             let l = widgets.label(id);

iris/src/android/attr.rs

@@ -12,6 +12,10 @@ impl<T: HasAndroidUiState> FocusHost for T {
         self.android_state_mut().focus = id;
     }
 
+    fn is_focused(&self, id: WeakWidget<TextEdit>) -> bool {
+        self.android_state().focus == Some(id)
+    }
+
     fn focus_gained(&mut self, region: Option<PixelRegion>) {
         // Showing the keyboard is a JNI call (`InputMethodManager.showSoftInput`),
         // and this runs deep inside the platform-agnostic sensor dispatch

iris/src/android/ime.rs

@@ -49,6 +49,52 @@ impl<State: AndroidAppState> IrisViewPeer<State> {
     fn focus(&self) -> Option<WeakWidget<TextEdit>> {
         self.state.android_state().focus
     }
+
+    /// Tell Gboard where the caret/selection and the composing region
+    /// actually are, via `InputMethodManager.updateSelection` -- every one
+    /// of android-view's own demo's `set_composing_text_internal`/`render`
+    /// calls this, and this bridge never did, which is what left Gboard's
+    /// own model of the field diverging from `TextEdit`'s real one after
+    /// the very first edit (RUST.md's P0 box, "doesn't enter it until I
+    /// hit space, and also doesn't move cursor forward" -- Gboard holds
+    /// its composing keystrokes back until it believes the app has caught
+    /// up, and without this call it never does). Called from
+    /// [`IrisViewPeer::after_input`], the one tail every touch/key/IME
+    /// callback already runs through, rather than duplicated at each of
+    /// this file's mutating methods.
+    ///
+    /// `candidates_start`/`candidates_end` report the composing region;
+    /// `-1, -1` when nothing is composing, matching `EditorInfo`'s own
+    /// convention. `compose_len` is tracked in `char`s (this module's doc
+    /// comment), so this reports it as that many UTF-16 units back from the
+    /// caret -- exact for the common BMP case, the same approximation
+    /// `set_composing_text` already makes.
+    pub(super) fn update_ime_selection(&mut self, ctx: &mut CallbackCtx) {
+        let Some(focus) = self.focus() else { return };
+        let text = &self.rsc[focus];
+        let Some(sel) = text.selection_range() else {
+            return;
+        };
+        let content = text.text();
+        let sel_start = byte_to_utf16(content, sel.start) as i32;
+        let sel_end = byte_to_utf16(content, sel.end) as i32;
+        let compose_len = self.state.android_state().compose_len;
+        let (comp_start, comp_end) = if compose_len > 0 {
+            let caret = byte_to_utf16(content, text.caret().unwrap_or(sel.end)) as i32;
+            (caret - compose_len as i32, caret)
+        } else {
+            (-1, -1)
+        };
+        let imm = ctx.view.input_method_manager(&mut ctx.env);
+        imm.update_selection(
+            &mut ctx.env,
+            &ctx.view,
+            sel_start,
+            sel_end,
+            comp_start,
+            comp_end,
+        );
+    }
 }
 
 impl<State: AndroidAppState> InputConnection for IrisViewPeer<State> {

iris/src/android/view.rs

@@ -325,6 +325,13 @@ impl<State: AndroidAppState> IrisViewPeer<State> {
             show_soft_input(&mut ctx.env, &ctx.view);
         }
 
+        // RUST.md's P0 box, "doesn't enter it until I hit space, and also
+        // doesn't move cursor forward": Gboard needs `updateSelection`
+        // after every edit to keep its own model of the field in sync, or
+        // it holds keystrokes back rather than trusting a screen it
+        // believes is stale. See `update_ime_selection`'s own doc.
+        self.update_ime_selection(ctx);
+
         let ui_state = self.state.android_state_mut();
         ui_state.cursor.end_frame();
         if self.render.needs_redraw(&ui_state.root, self.rsc.widgets()) {
@@ -650,6 +657,22 @@ impl<State: AndroidAppState> ViewPeer for IrisViewPeer<State> {
         let content_scale = self.state.android_state().content_scale;
         match AndroidRenderer::new(window, width as u32, height as u32, content_scale) {
             Ok(renderer) => {
+                // A genuinely new renderer means a genuinely new GPU device
+                // and a fresh, empty glyph atlas -- the CPU-side glyph
+                // cache (`TextData::atlas`) and the texture bookkeeping it
+                // is built on (`UiData::textures`) both outlive `renderer`
+                // itself (they live on `self.rsc`, not on `AndroidRenderer`),
+                // so without this they would keep pointing at the *old*
+                // device's now-gone textures -- the app-switch counterpart
+                // to the keyboard-resize glyph wipe this same function's
+                // `already_live` branch above already fixed by reusing the
+                // renderer instead of rebuilding it. One mechanism either
+                // way: this call only runs on the branch that actually
+                // builds a new renderer, exactly where invalidation is
+                // needed, never on the reuse branch, where it would throw
+                // away perfectly valid GPU state for nothing.
+                self.rsc.ui.text.atlas.clear();
+                self.rsc.ui.textures.reset();
                 self.state.android_state_mut().renderer = Some(renderer);
                 self.render(ctx);
             }

iris/src/attr.rs

@@ -22,6 +22,13 @@ pub trait FocusHost {
     /// it was hit in (`None` when the widget could not be located, which
     /// happens for one it was just deselected from).
     fn focus_gained(&mut self, region: Option<PixelRegion>);
+    /// Whether `id` is the current focus target -- what [`select`] uses to
+    /// tell a fresh press (which must wait to see whether it becomes a tap
+    /// or a drag before focusing/showing the IME, Iris 2026-09-06: "if I
+    /// swipe over the input bar it brings up the keyboard") from a drag
+    /// continuing inside a field that was already focused (an ordinary
+    /// drag-to-select, unaffected).
+    fn is_focused(&self, id: WeakWidget<TextEdit>) -> bool;
 }
 
 /// Helper shared by every `FocusHost` impl, so the double-click window is
@@ -33,6 +40,17 @@ pub fn recent_click(last_click: &mut Instant) -> bool {
     recent
 }
 
+/// `PressStart`/`Pressing`/`PressEnd`, all for the left button -- what
+/// [`Selector`]/[`Selectable`] register instead of [`CursorSense::
+/// click_or_drag`], so their shared handler (`on_press`, below) sees every
+/// frame of a gesture and can tell a completed tap from a drag itself,
+/// rather than reacting to `PressStart` alone the way `click_or_drag`'s
+/// consumer used to (Iris, 2026-09-06: "if I swipe over the input bar it
+/// brings up the keyboard").
+fn press_track() -> CursorSenses {
+    CursorSense::click() | CursorSense::Pressing(CursorButton::Left) | CursorSense::unclick()
+}
+
 pub struct Selector;
 
 impl<Rsc: HasEvents, W: Widget + 'static> WidgetAttr<Rsc, W> for Selector
@@ -42,7 +60,7 @@ where
     type Input = WeakWidget<TextEdit>;
 
     fn run(rsc: &mut Rsc, container: WeakWidget<W>, id: Self::Input) {
-        rsc.register_event(container, CursorSense::click_or_drag(), move |ctx, rsc| {
+        rsc.register_event(container, press_track(), move |ctx, rsc| {
             let region = ctx.data.render.window_region(&id, &*rsc).unwrap();
             let id_pos = region.top_left;
             let container_pos = ctx
@@ -53,14 +71,14 @@ where
                 .top_left;
             let pos = ctx.data.pos + container_pos - id_pos;
             let size = region.size();
-            select(
+            on_press(
                 rsc,
                 ctx.data.render,
                 ctx.state,
                 id,
                 pos,
                 size,
-                ctx.data.sense.is_dragging(),
+                ctx.data.sense,
             );
         });
     }
@@ -75,31 +93,71 @@ where
     type Input = ();
 
     fn run(rsc: &mut Rsc, id: WeakWidget<TextEdit>, _: Self::Input) {
-        rsc.register_event(id, CursorSense::click_or_drag(), move |ctx, rsc| {
+        rsc.register_event(id, press_track(), move |ctx, rsc| {
-            select(
+            on_press(
                 rsc,
                 ctx.data.render,
                 ctx.state,
                 id,
                 ctx.data.pos,
                 ctx.data.size,
-                ctx.data.sense.is_dragging(),
+                ctx.data.sense,
             );
         });
     }
 }
 
-fn select(
+/// One press-track frame (`PressStart`, `Pressing` or `PressEnd`) over a
+/// selectable field. A field that is *already* focused behaves exactly as
+/// `click_or_drag` always did -- every frame updates the selection, which
+/// is what lets a finger already inside a focused field drag out a
+/// selection. A field that is **not** focused withholds `select`'s
+/// focus-granting side effects (and so the platform-specific `focus_gained`
+/// that shows the keyboard) until the press resolves as a tap: `PressEnd`
+/// with no frame in between having moved past [`DRAG_SLOP`] from where the
+/// press began. A drag recognised before release simply cancels the
+/// pending tap and does nothing further here -- it is not consumed, so
+/// whatever is behind the field (a list to pan) still sees every frame of
+/// it, the same as a drag that never touched a selectable field at all.
+fn on_press(
     rsc: &mut impl UiRsc,
     render: &UiRenderState,
     state: &mut impl FocusHost,
     id: WeakWidget<TextEdit>,
     pos: Vec2,
     size: Vec2,
-    dragging: bool,
+    sense: CursorSense,
 ) {
+    if state.is_focused(id) {
+        let recent = matches!(sense, CursorSense::PressStart(_)) && state.recent_click();
+        id.edit(rsc).select(pos, size, sense.is_dragging(), recent);
+        return;
+    }
+
+    match sense {
+        CursorSense::PressStart(_) => {
+            id.edit(rsc).text.press_origin = Some(pos);
+        }
+        CursorSense::Pressing(_) => {
+            let ctx = id.edit(rsc);
+            if let Some(origin) = ctx.text.press_origin
+                && ((pos.x - origin.x).abs() > DRAG_SLOP || (pos.y - origin.y).abs() > DRAG_SLOP)
+            {
+                // Past the slop before release: this is a drag, not a tap
+                // -- give up the pending focus rather than granting it once
+                // the finger lifts wherever it happens to be by then.
+                ctx.text.press_origin = None;
+            }
+        }
+        CursorSense::PressEnd(_) => {
+            let was_tap = id.edit(rsc).text.press_origin.take().is_some();
+            if was_tap {
                 let recent = state.recent_click();
-    id.edit(rsc).select(pos, size, dragging, recent);
+                id.edit(rsc).select(pos, size, false, recent);
                 state.set_focus(Some(id));
                 state.focus_gained(render.window_region(&id, &*rsc));
+            }
+        }
+        _ => {}
+    }
 }

iris/src/default/attr.rs

@@ -10,6 +10,10 @@ impl<T: HasDefaultUiState> FocusHost for T {
         self.default_state_mut().focus = id;
     }
 
+    fn is_focused(&self, id: WeakWidget<TextEdit>) -> bool {
+        self.default_state().focus == Some(id)
+    }
+
     fn focus_gained(&mut self, region: Option<PixelRegion>) {
         let state = self.default_state_mut();
         let Some(region) = region else { return };

iris/src/layout_tests.rs

@@ -183,3 +183,80 @@ fn a_mask_stays_put_while_its_scrolled_content_moves() {
     assert_eq!(mask_delta_before, [0.0, 0.0]);
     assert_eq!(mask_delta_after, [0.0, 0.0]);
 }
+
+/// Reproduces `transcript_ui::composer::build_composer`'s exact tree shape
+/// (a `Rect` background stacked behind a `Span::RIGHT`-wrapped, padded,
+/// `rest`-width `TextEdit`, itself the second child of an outer
+/// `Span::DOWN` beside a `rest(1)`-height sibling) without the event/
+/// resource plumbing `composer.rs`'s builders need, to isolate whether the
+/// bug Iris reported on 2026-09-06 ("text seems to not appear in box")
+/// is this crate's layout engine or something specific to the real
+/// composer/screen. `TextEditable::edit` only needs `UiRsc`, so a plain
+/// insert exercises the exact redraw path a keystroke does.
+fn composer_like_tree(rsc: &mut TestRsc) -> (WeakWidget<TextEdit>, StrongWidget) {
+    let field = wtext("")
+        .editable(EditMode::MultiLine)
+        .text_align(Align::LEFT)
+        .wrap(true)
+        .size(18)
+        .color(UiColor::WHITE)
+        .add(rsc);
+    let bar = (field.pad(dp(12)).width(rest(1)),)
+        .span(Dir::RIGHT)
+        .background(rect(UiColor::new(40, 40, 46, 255)))
+        .add(rsc);
+    let list_stand_in = rect(UiColor::BLACK).height(rest(1)).add(rsc);
+    let tree = (list_stand_in, bar).span(Dir::DOWN).add_strong(rsc).any();
+    (field, tree)
+}
+
+/// The reproduction itself. A window this tall stands in for the keyboard
+/// closed; the second, shorter `resize` stands in for `adjustResize`
+/// shrinking the surface when the IME opens -- exactly the sequence
+/// `IrisViewPeer::surface_changed` drives on a real keyboard open. Typing
+/// happens both before and after, since Iris's report was specifically
+/// that text typed *after* the keyboard was already up did not appear.
+#[test]
+fn composing_text_after_a_keyboard_resize_lands_in_the_bars_own_region() {
+    let mut rsc = TestRsc {
+        ui: UiData::default(),
+    };
+    let (field, root) = composer_like_tree(&mut rsc);
+    let mut render = UiRenderState::new();
+
+    render.resize((1080.0, 2298.0));
+    render.update(&root, &mut rsc);
+    field.edit(&mut rsc).insert("a");
+    render.update(&root, &mut rsc);
+
+    let before_px = render.window_region(&field, &rsc).unwrap();
+    // The field is one line plus 12dp of padding on a 2298-tall window --
+    // nowhere near the whole window's height, and anchored at the bottom.
+    assert!(
+        before_px.bot_right.y - before_px.top_left.y < 200.0,
+        "before a resize: {before_px:?}"
+    );
+    assert!(
+        before_px.top_left.y > 1800.0,
+        "expected the bar near the bottom before a resize: {before_px:?}"
+    );
+
+    // The keyboard opens: a real `surface_changed`/`resize` to a shorter
+    // window, then a further keystroke -- the redraw that must land in the
+    // bar's new (also short) region, not whatever region a provisional
+    // measurement pass used along the way.
+    render.resize((1080.0, 1478.0));
+    render.update(&root, &mut rsc);
+    field.edit(&mut rsc).insert("b");
+    render.update(&root, &mut rsc);
+
+    let after_px = render.window_region(&field, &rsc).unwrap();
+    assert!(
+        after_px.bot_right.y - after_px.top_left.y < 200.0,
+        "after a resize + keystroke: {after_px:?}"
+    );
+    assert!(
+        after_px.top_left.y > 1200.0,
+        "expected the bar near the bottom of the shorter window: {after_px:?}"
+    );
+}

iris/src/sense.rs

@@ -22,6 +22,14 @@ pub enum CursorSense {
     Hovering,
     HoverEnd,
     Scroll,
+    /// Delivered exactly once, in place of `PressEnd`, to whichever widget
+    /// currently holds pointer capture (`UiRenderState::capture_pointer`)
+    /// when the button lifts -- see `iris::sense`'s pointer-capture doc
+    /// and `DragGesture`. A widget must register this explicitly (it is
+    /// never bundled into `click_or_drag`/`unclick`, since most widgets
+    /// never call `capture_pointer` and have no use for it) to receive it
+    /// at all; ordinary hit-tested widgets keep seeing `PressEnd`.
+    Drop,
 }
 
 #[derive(Clone)]
@@ -31,6 +39,21 @@ impl Event for CursorSenses {
     type Data<'a> = CursorData<'a>;
     type State = SensorState;
     fn should_run<'a>(&self, data: &Self::Data<'a>) -> Option<Self::Data<'a>> {
+        // `Drop` is never derived from raw cursor/hover state below (the
+        // free `should_run`'s own arm for it is only ever asked here,
+        // never independently true or false against the button) -- it is
+        // set exclusively by `run_sensors`' pointer-capture branch, which
+        // has already decided this exact frame is the captured widget's
+        // terminal event. Matching it by identity, ahead of the general
+        // derivation, matters because a captured widget's registration
+        // list very likely also carries `PressEnd` (`unclick()`, for the
+        // ordinary un-captured case) -- the same button-lift condition
+        // `PressEnd` matches on, so falling through to the loop below
+        // would let whichever of the two happens to be registered first
+        // win, silently swallowing the `Drop` a caller relied on.
+        if data.sense == CursorSense::Drop {
+            return self.contains(&CursorSense::Drop).then(|| data.clone());
+        }
         if let Some(sense) = should_run(self, &data.cursor, data.hover) {
             let mut data = data.clone();
             data.sense = sense;
@@ -177,6 +200,48 @@ impl SensorUi for UiRenderState {
         cursor: CursorState,
         window_size: Vec2,
     ) {
+        // Exclusive pointer capture (`UiRenderState::capture_pointer`,
+        // `DragGesture`): once some widget has committed to a drag, every
+        // other widget sees nothing from this pointer at all -- no hover,
+        // no click, no press -- until it releases. This is what lets a
+        // fast pan or a selection keep going once the finger has moved
+        // off whatever hit region first noticed the press (including
+        // right off the end of the gesture, at `PressEnd`/`Cancel`): a
+        // per-widget hit test would otherwise silently stop delivering to
+        // *anyone* the moment the pointer left every registered region,
+        // which is exactly what used to leave a fling never started (no
+        // widget ever saw the release). The captured widget keeps getting
+        // ordinary `Pressing` frames while the button is down and gets
+        // exactly one `Drop` -- not `PressEnd` -- the frame it lifts,
+        // which also releases the capture.
+        if let Some(id) = self.captured_pointer() {
+            let Some(shape) = self.resolved_region(&id, rsc) else {
+                self.release_pointer();
+                return;
+            };
+            let region = shape.to_px(window_size);
+            let button_down = cursor.buttons.select(&CursorButton::Left).is_on();
+            let sense = if button_down {
+                CursorSense::Pressing(CursorButton::Left)
+            } else {
+                CursorSense::Drop
+            };
+            let data = CursorData {
+                pos: cursor.pos - region.top_left,
+                size: region.bot_right - region.top_left,
+                scroll_delta: cursor.scroll_delta,
+                hover: ActivationState::On,
+                cursor: cursor.clone(),
+                sense,
+                render: self,
+            };
+            rsc.run_event::<CursorSense>(id, data, state);
+            if !button_down {
+                self.release_pointer();
+            }
+            return;
+        }
+
         // in order to remove this take, need to store active list in UiRenderState somehow
         // this would probably be done through a generic parameter that adds yet another rsc /
         // state like thing, but local to render state, and is passed to UiRsc events so you can
@@ -266,6 +331,15 @@ pub fn should_run(
             CursorSense::Hovering => hover.is_on(),
             CursorSense::HoverEnd => hover.is_end(),
             CursorSense::Scroll => cursor.scroll_delta != Vec2::ZERO,
+            // Never derived here -- `Drop` only ever fires through
+            // `CursorSenses::should_run`'s own special case, ahead of this
+            // loop, for the one widget `run_sensors`' capture branch is
+            // delivering it to this frame. If this arm answered from raw
+            // button state instead, an ordinary hit-tested widget that
+            // happened to register `Drop` (with no capture involved at
+            // all) would see it fire on every plain button-up under the
+            // cursor.
+            CursorSense::Drop => false,
         } {
             return Some(*sense);
         }
@@ -541,6 +615,138 @@ impl DragArbiter {
     }
 }
 
+/// What a [`DragGesture`] decided this frame -- [`DragOutcome`] plus the
+/// one further state a shared gesture needs: the drag ending, with the
+/// released velocity if (and only if) it had committed to panning.
+#[derive(Debug, Clone, Copy, PartialEq)]
+pub enum GestureOutcome {
+    Undecided,
+    /// Same units and sign as [`DragOutcome::Pan`] -- the caller's own
+    /// convention (`List::scroll`'s, for a transcript) to apply.
+    Pan(f32),
+    SelectStart,
+    SelectExtend,
+    /// The drag ended -- `PressEnd` or the capture's own terminal `Drop`.
+    /// `Some(velocity)` only if the gesture had committed to panning
+    /// (never a tap, a long-press selection, or one still `Undecided`);
+    /// same units as `Pan`, so a caller hands it to `List::fling` with
+    /// whatever sign flip it already applies to `Pan`.
+    Released(Option<f32>),
+}
+
+/// Bundles a [`DragArbiter`] and a [`VelocityTracker`] into the one thing
+/// most drag-driven widgets need: arbitrate pan-vs-hold, track the pan's
+/// velocity, and take pointer capture (`UiRenderState::capture_pointer`)
+/// the moment the gesture commits so the rest of it -- including the
+/// terminal release -- keeps reaching the same widget even after the
+/// finger has moved off whatever hit region first noticed the press. Iris
+/// asked for this to live here rather than in `transcript-ui::Selection`
+/// (2026-09-06, recorded in `IRIS.md`): "dragging should be part of the
+/// default input system ... anything that provides good performance and
+/// can be generalized well is part of iris rather than the app." A caller
+/// still decides what a committed pan or a completed selection *means*
+/// (transcript-ui's pan-vs-select is one call site; a slider or a plain
+/// scroll area is another) -- this only owns the *mechanics* every one of
+/// them would otherwise duplicate.
+pub struct DragGesture {
+    arbiter: DragArbiter,
+    velocity: VelocityTracker,
+}
+
+impl Default for DragGesture {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl DragGesture {
+    pub fn new() -> Self {
+        Self {
+            arbiter: DragArbiter::new(),
+            velocity: VelocityTracker::new(),
+        }
+    }
+
+    /// Whether this gesture has no press in flight -- a thin passthrough
+    /// to the underlying `DragArbiter::is_idle`, for a caller (a test, a
+    /// diagnostic) that wants to observe the recovery behaviour `handle`'s
+    /// idle-recovery branch documents without reaching into a private
+    /// field.
+    pub fn is_idle(&self) -> bool {
+        self.arbiter.is_idle()
+    }
+
+    /// Feed one frame of a gesture through. `id` is the widget iris should
+    /// give exclusive pointer input to once this gesture commits to
+    /// panning or selecting -- a stable widget that outlives the gesture
+    /// (a `List`'s own id, not one of its virtualised rows, which can be
+    /// retired mid-drag as content scrolls). `render` is `CursorData`'s
+    /// own field, already in hand at every call site. `already_selected`
+    /// only matters for the first frame of a gesture (`PressStart`, or the
+    /// recovery branch below) -- see `DragArbiter::press_start`'s doc.
+    pub fn handle(
+        &mut self,
+        render: &UiRenderState,
+        id: WidgetId,
+        sense: CursorSense,
+        pos_window: Vec2,
+        now: Instant,
+        already_selected: bool,
+    ) -> GestureOutcome {
+        match sense {
+            CursorSense::PressStart(_) => {
+                self.velocity.reset();
+                self.arbiter.press_start(pos_window, now, already_selected);
+                self.dispatch(render, id, pos_window, now)
+            }
+            CursorSense::Drop | CursorSense::PressEnd(_) => {
+                let released = if self.arbiter.is_panning() {
+                    Some(self.velocity.velocity())
+                } else {
+                    None
+                };
+                self.arbiter.release();
+                render.release_pointer();
+                GestureOutcome::Released(released)
+            }
+            // See `DragArbiter::update`'s own doc: a `Pressing` frame can
+            // arrive with no matching `PressStart` if the touch-down
+            // landed outside whichever hit region first noticed it.
+            _ if self.arbiter.is_idle() => {
+                self.velocity.reset();
+                self.arbiter.press_start(pos_window, now, already_selected);
+                self.dispatch(render, id, pos_window, now)
+            }
+            _ => self.dispatch(render, id, pos_window, now),
+        }
+    }
+
+    fn dispatch(
+        &mut self,
+        render: &UiRenderState,
+        id: WidgetId,
+        pos: Vec2,
+        now: Instant,
+    ) -> GestureOutcome {
+        match self.arbiter.update(pos, now) {
+            DragOutcome::Undecided => GestureOutcome::Undecided,
+            DragOutcome::Pan(dy) => {
+                render.capture_pointer(id);
+                self.velocity.add_sample(dy, now);
+                GestureOutcome::Pan(dy)
+            }
+            DragOutcome::SelectStart => {
+                render.capture_pointer(id);
+                GestureOutcome::SelectStart
+            }
+            DragOutcome::SelectExtend => {
+                render.capture_pointer(id);
+                GestureOutcome::SelectExtend
+            }
+        }
+    }
+}
+
 /// How far back a [`VelocityTracker`] looks when estimating a fling's
 /// initial speed -- Android's own `VelocityTracker` defaults to a similar
 /// short window so a gesture's last flick dominates over its slower start.

iris/src/sense_tests.rs

@@ -122,3 +122,127 @@ fn a_button_over_a_list_scrolls_the_list_and_still_clicks() {
         "the button on top must still receive an actual click"
     );
 }
+
+/// The bug behind "finger flings do nothing" (RUST.md's P0 phone report,
+/// defect 2): a fast gesture's `PressEnd` can land at a screen position
+/// nothing is registered at -- past the edge of whatever widget noticed
+/// the press, in a gap, or off the loaded content entirely. Before pointer
+/// capture, `run_sensors`' hit test simply delivered nothing that frame,
+/// so a widget mid-drag never saw its release and never got a chance to
+/// start a fling. `UiRenderState::capture_pointer`/`DragGesture` fix this
+/// by giving the drag's widget every frame regardless of where the
+/// pointer is, including the terminal `Drop` in place of `PressEnd`.
+#[test]
+fn a_release_outside_every_hit_region_still_reaches_the_captured_widget() {
+    let mut rsc = SenseRsc {
+        ui: UiData::default(),
+        events: EventManager::default(),
+    };
+
+    // A small draggable widget in the corner -- the release below lands
+    // far outside it, exactly the "moved off the hit region" case.
+    let draggable = rsc.ui.widgets.add_strong(Rect::new(UiColor::WHITE)).any();
+    let draggable_weak = draggable.weak();
+
+    let dropped = Rc::new(Cell::new(false));
+    {
+        let dropped = dropped.clone();
+        rsc.register_event(
+            draggable_weak,
+            CursorSense::click_or_drag() | CursorSense::unclick() | CursorSense::Drop,
+            move |ctx, rsc| match ctx.data.sense {
+                CursorSense::PressStart(_) | CursorSense::Pressing(_) => {
+                    // Any committed drag takes capture -- a real caller
+                    // would gate this on a `DragArbiter`/`DragGesture`
+                    // decision, but this test only needs to exercise the
+                    // capture-and-release mechanics themselves.
+                    ctx.data.render.capture_pointer(draggable_weak.id());
+                    let _ = rsc;
+                }
+                CursorSense::Drop => dropped.set(true),
+                _ => {}
+            },
+        );
+    }
+
+    let mut render = UiRenderState::new();
+    render.resize((100.0, 100.0));
+    render.update(&draggable, &mut rsc);
+
+    let mut state = ();
+    let mut press = cursor_at((5.0, 5.0).into());
+    press.buttons.left = ActivationState::Start;
+    render.run_sensors(&mut rsc, &mut state, press, (100.0, 100.0).into());
+    assert_eq!(
+        render.captured_pointer(),
+        Some(draggable.id()),
+        "the press should have taken capture"
+    );
+
+    // The release lands nowhere near the widget's own region -- the exact
+    // shape of a fast fling's `ACTION_UP`.
+    let mut release = cursor_at((95.0, 95.0).into());
+    release.buttons.left = ActivationState::End;
+    render.run_sensors(&mut rsc, &mut state, release, (100.0, 100.0).into());
+
+    assert!(
+        dropped.get(),
+        "a release outside every widget's hit region must still reach \
+         the widget holding pointer capture"
+    );
+    assert_eq!(
+        render.captured_pointer(),
+        None,
+        "Drop must release the capture"
+    );
+}
+
+/// A widget that never registers `CursorSense::Drop` at all must not be
+/// affected by someone else's capture -- capture is per-gesture, not
+/// global suppression of the whole input system for widgets that were
+/// never party to it. (Practically this matters because a captured
+/// widget's registration list still has to include `Drop` for `should_run`
+/// to ever match it; this pins that half of the contract.)
+#[test]
+fn capturing_one_widget_starves_every_other_widget_of_events() {
+    let mut rsc = SenseRsc {
+        ui: UiData::default(),
+        events: EventManager::default(),
+    };
+
+    let a = rsc.ui.widgets.add_strong(Rect::new(UiColor::WHITE));
+    let a_weak = a.weak();
+    let b = rsc.ui.widgets.add_strong(Rect::new(UiColor::RED));
+    let b_weak = b.weak();
+
+    let b_hovered = Rc::new(Cell::new(false));
+    {
+        let b_hovered = b_hovered.clone();
+        rsc.register_event(b_weak, CursorSense::Hovering, move |_ctx, _rsc| {
+            b_hovered.set(true);
+        });
+    }
+
+    let root = rsc
+        .ui
+        .widgets
+        .add_strong(Stack {
+            children: vec![a.any(), b.any()],
+            size: StackSize::default(),
+        })
+        .any();
+
+    let mut render = UiRenderState::new();
+    render.resize((100.0, 100.0));
+    render.update(&root, &mut rsc);
+    render.capture_pointer(a_weak.id());
+
+    let mut state = ();
+    let cursor = cursor_at((50.0, 50.0).into());
+    render.run_sensors(&mut rsc, &mut state, cursor, (100.0, 100.0).into());
+
+    assert!(
+        !b_hovered.get(),
+        "while a's drag holds capture, b must see no hover at all"
+    );
+}

iris/src/widget/list.rs

@@ -489,6 +489,24 @@ impl List {
         true
     }
 
+    /// The anchor's own row index and pixel offset, formatted the same
+    /// shape Compose's `firstVisibleItemIndex`/`firstVisibleItemScrollOffset`
+    /// report (`idx=N/off=Mpx`) -- what RUST.md's "Benchmark v2" fling
+    /// phase reads before/after/between its fling runs so the two apps'
+    /// travel can be compared directly. `more_before`/`more_after`
+    /// sentinels print as `idx=more-before`/`idx=more-after` rather than
+    /// leaking their internal `isize` representation; `idx=none` if the
+    /// list has never drawn (no anchor yet -- e.g. right after
+    /// `jump_to_end` and before the next frame runs `repair_anchor`).
+    pub fn anchor_position_display(&self) -> String {
+        match self.anchor {
+            None => "idx=none".to_string(),
+            Some(a) if a.slot == BEFORE_SLOT => "idx=more-before".to_string(),
+            Some(a) if a.slot == AFTER_SLOT => "idx=more-after".to_string(),
+            Some(a) => format!("idx={}/off={}px", a.slot, a.offset.round() as i64),
+        }
+    }
+
     /// Snap to the newest content (last item, or the `more_after`
     /// sentinel if set), bottom-aligned to the viewport. O(1).
     pub fn jump_to_end(&mut self) {
@@ -534,6 +552,20 @@ impl List {
         self.extents.get(&key).map(|e| (e.top, e.bottom))
     }
 
+    /// The row whose on-screen box (as of the last layout) contains
+    /// `viewport_pos`, or `None` if it falls outside every row currently
+    /// drawn (a gap, a header, or off the loaded content entirely). O
+    /// (visible rows), same as `reanchor_at_tap`. What a caller resolves a
+    /// pointer-captured gesture's row-under-the-finger against once the
+    /// gesture is no longer being delivered through any one row's own hit
+    /// region -- see `iris::sense`'s pointer-capture doc.
+    pub fn key_at(&self, viewport_pos: f32) -> Option<RowKey> {
+        self.extents
+            .iter()
+            .find(|(_, ext)| viewport_pos >= ext.top && viewport_pos <= ext.bottom)
+            .map(|(&key, _)| key)
+    }
+
     fn slot_exists(&self, slot: isize) -> bool {
         match slot {
             BEFORE_SLOT => self.more_before.is_some(),
@@ -1314,6 +1346,63 @@ mod tests {
         }
     }
 
+    /// RUST.md's P0 phone report (Iris's screenshot, 2026-09-06): a
+    /// replaced row's primitives drawn a second time, overlapping the
+    /// replacement. Reproduces the exact path `TranscriptScreen::apply`'s
+    /// `ReplaceLast` case drives up to 400 times during a streamed reply
+    /// (`bench_client.rs`'s stream phase): the last slot's widget is
+    /// swapped for a brand-new one, same key, and (since a fresh widget
+    /// has no cached height) placed via `place`'s `draw_twice` path every
+    /// time -- the provisional-then-real two-draw sequence LAYOUT.md
+    /// documents as the one place in this crate that deliberately draws a
+    /// widget twice. If `draw_inner`'s old-children diffing or
+    /// `UiRenderState::remove`'s primitive freeing ever failed to retire
+    /// the evicted widget (or the provisional draw's own primitives), it
+    /// would show up here as `active_widgets` growing without bound.
+    /// **Passes as written** -- this pins the widget-arena layer as
+    /// correct in isolation; see the P0 box for where the duplicate was
+    /// actually chased to instead (`Span`'s two-phase draw and the
+    /// `redraw_all`-vs-`redraw_updates` split, still open).
+    #[test]
+    fn replacing_the_last_row_many_times_does_not_leak_primitives() {
+        let mut rsc = TestRsc {
+            ui: UiData::default(),
+        };
+        let mut list = List::new(Axis::Y);
+        for key in 0..5u64 {
+            let (_bg_id, row) = background_styled_row(&mut rsc, 20.0);
+            list.push_back(ListRow::new(key, row));
+        }
+        let (list_weak, root) = add_list(&mut rsc, list);
+
+        let mut render = UiRenderState::new();
+        render.resize((100.0, 100.0));
+        render.update(&root, &mut rsc);
+
+        let before = render.active_widgets();
+        for i in 0..400u32 {
+            // A varying height keeps every replace on the `draw_twice`
+            // (cache-miss) path rather than settling into the O(1)
+            // same-size `mov` fast path once the height happens to repeat.
+            let (_bg_id, new_row) = background_styled_row(&mut rsc, 20.0 + (i % 3) as f32);
+            rsc.ui
+                .widgets
+                .get_mut(&list_weak)
+                .unwrap()
+                .replace_back(ListRow::new(4, new_row));
+            render.update(&root, &mut rsc);
+        }
+        let after = render.active_widgets();
+
+        assert_eq!(
+            before, after,
+            "400 replaces of the last row must leave exactly the same \
+             number of active widgets as before a leaked id (and the \
+             primitives that live as long as its ActiveData does) would \
+             show up here as growth"
+        );
+    }
+
     /// Enough rows, tall enough, that a fling toward the start has real
     /// room to travel before `at_start` clamps it -- shared by the fling
     /// tests below.
@@ -1456,4 +1545,22 @@ mod tests {
             first.top
         );
     }
+
+    #[test]
+    fn anchor_position_display_before_any_draw_is_none() {
+        let list = List::new(Axis::Y);
+        assert_eq!(list.anchor_position_display(), "idx=none");
+    }
+
+    #[test]
+    fn anchor_position_display_reports_slot_and_offset() {
+        let mut rsc = TestRsc {
+            ui: UiData::default(),
+        };
+        let (list_weak, root, mut render) = build_flingable_list(&mut rsc);
+        let _ = (&root, &mut render);
+        let list_ref = rsc.ui.widgets.get(&list_weak).unwrap();
+        assert!(list_ref.anchor_position_display().starts_with("idx="));
+        assert!(!list_ref.anchor_position_display().contains("none"));
+    }
 }

iris/src/widget/text/edit.rs

@@ -32,6 +32,15 @@ pub struct TextEdit {
     #[cfg_attr(target_os = "android", allow(dead_code))]
     history: Vec<(String, Option<Selection>)>,
     double_hit: Option<usize>,
+    /// Where an in-flight press over this field began, while it is still
+    /// undecided whether the gesture is a tap (focus/show the IME) or a
+    /// drag (attr.rs's `Selector`/`Selectable`, Iris 2026-09-06: a swipe
+    /// over the composer must not summon the keyboard). `None` both before
+    /// any press and once the gesture has been decided either way --
+    /// `attr.rs` is the only reader/writer, kept `pub(crate)` rather than
+    /// behind an accessor since it is pure bookkeeping with no invariant
+    /// beyond "some press is undecided," same shape as `double_hit` above.
+    pub(crate) press_origin: Option<Vec2>,
     pub mode: EditMode,
 }
 
@@ -48,6 +57,7 @@ impl TextEdit {
             selection: None,
             history: Default::default(),
             double_hit: None,
+            press_origin: None,
             mode,
         }
     }
@@ -698,6 +708,67 @@ mod tests {
         assert_eq!(content(&t), "に");
     }
 
+    /// `android/ime.rs`'s `set_composing_text` calls `replace` and expects
+    /// the caret to land right after the inserted text, growing with it on
+    /// every re-send -- the buffer-level half of RUST.md's P0 box ("doesn't
+    /// enter it until I hit space, and also doesn't move cursor forward").
+    #[test]
+    fn composing_advances_the_caret_with_the_growing_text() {
+        let (mut t, mut d) = edit("", EditMode::SingleLine);
+        ctx(&mut t, &mut d).set_caret(0);
+        ctx(&mut t, &mut d).replace(0, "h");
+        assert_eq!(t.caret(), Some(1));
+        ctx(&mut t, &mut d).replace(1, "hi");
+        assert_eq!(content(&t), "hi");
+        assert_eq!(t.caret(), Some(2));
+        ctx(&mut t, &mut d).replace(2, "hit");
+        assert_eq!(content(&t), "hit");
+        assert_eq!(t.caret(), Some(3));
+    }
+
+    /// The IME's `commitText` (`android_view::InputConnection::commit_text`'s
+    /// default body): finish a composition in place, same as a real word
+    /// boundary (a space) landing after Gboard's composing span.
+    #[test]
+    fn committing_composed_text_leaves_it_in_place_with_the_caret_after_it() {
+        let (mut t, mut d) = edit("say ", EditMode::SingleLine);
+        ctx(&mut t, &mut d).set_caret(4);
+        ctx(&mut t, &mut d).replace(0, "hi");
+        assert_eq!(content(&t), "say hi");
+        // `finish_composing_text`/`commit_text` do not themselves touch the
+        // buffer -- only the IME's own `compose_len` bookkeeping resets, in
+        // `android/ime.rs`. Confirms the buffer already holds committed
+        // text as plain, uncomposed content: a further `replace(0, " ")`
+        // (the space that ends the word) appends rather than overwriting.
+        ctx(&mut t, &mut d).replace(0, " ");
+        assert_eq!(content(&t), "say hi ");
+        assert_eq!(t.caret(), Some(7));
+    }
+
+    /// `TextEditCtx::delete_byte_range` is `deleteSurroundingText`'s entry
+    /// point once `android/ime.rs` has converted UTF-16 code units to
+    /// bytes -- exercised directly here in bytes, since the UTF-16 math
+    /// itself is `android/ime.rs`'s own `byte_to_utf16`/`utf16_to_byte`,
+    /// outside this widget-only test module.
+    #[test]
+    fn delete_byte_range_removes_exactly_that_range() {
+        let (mut t, mut d) = edit("hello world", EditMode::SingleLine);
+        ctx(&mut t, &mut d).delete_byte_range(5, 11);
+        assert_eq!(content(&t), "hello");
+        assert_eq!(t.caret(), Some(5));
+    }
+
+    /// `set_cursor_byte` is `setSelection`'s entry point -- collapses to a
+    /// caret at the given byte offset regardless of any span that was there.
+    #[test]
+    fn set_cursor_byte_collapses_to_a_caret_there() {
+        let (mut t, mut d) = edit("hello world", EditMode::SingleLine);
+        ctx(&mut t, &mut d).select_all();
+        ctx(&mut t, &mut d).set_cursor_byte(5);
+        assert_eq!(t.selected_text(), None);
+        assert_eq!(t.caret(), Some(5));
+    }
+
     #[test]
     fn motion_moves_the_caret_and_shift_extends_a_span() {
         let (mut t, mut d) = edit("abc", EditMode::SingleLine);

iris/transcript-ui/src/composer.rs

@@ -8,14 +8,58 @@
 //! measures whatever vertical space is left each frame -- nothing here
 //! computes a height by hand, and growing this field is exactly the
 //! O(1)-move-chain case LAYOUT.md and I3's benchmark already measured.
+//!
+//! **Rebuilt 2026-09-06** (Iris's phone report on the dc01f88 build: the
+//! grey bar drawn as a short, fixed strip with the typed text ~150px below
+//! it on black, and empty black between the bar and the keyboard). One
+//! widget now, top to bottom: an opaque background sized to its content
+//! (`.background`, the same `Stack` idiom the header row's `HEADER_SURFACE`
+//! already uses), the field inside `dp` padding and capped at
+//! [`MAX_LINES`] before it scrolls instead of growing forever, and an
+//! outer [`Pad`] whose `bottom` [`TranscriptScreen::set_bottom_inset`]
+//! rewrites in place whenever the keyboard opens/closes -- never rebuilt,
+//! since `field` is strongly owned inside this tree and this crate's
+//! widgets cannot be re-parented once added (this module's own comment
+//! below on why `build_composer` hands back a **weak** id).
 
 use iris::prelude::*;
 
+/// Caps the field's growth at roughly six lines of its own 18px text
+/// before it scrolls instead of consuming the whole screen -- an
+/// approximation (line-height and padding folded into one round `dp`
+/// number) rather than a value derived from the font's real metrics,
+/// which nothing in this crate exposes to a caller today.
+const MAX_LINES: f32 = 6.0;
+const APPROX_LINE_HEIGHT_DP: f32 = 24.0;
+const FIELD_PAD_DP: f32 = 12.0;
+
 /// `field` is exposed so the caller can read its content on submit
 /// (`field.edit(rsc).text()`) and clear it afterward
 /// (`field.edit(rsc).set("")`).
 pub struct Composer {
     pub field: WeakWidget<TextEdit>,
+    /// The bar's own outer padding -- only `bottom` is ever changed, by
+    /// [`Self::set_bottom_inset`]. A `Pad` around the whole bar rather than
+    /// a rebuilt tree, because `field` lives inside it and cannot be
+    /// re-added to a new wrapper once it is strongly owned here.
+    outer_pad: WeakWidget<Pad>,
+}
+
+impl Composer {
+    /// Called by the platform shell (Android's `on_insets_changed`, e.g.)
+    /// whenever the space below the bar changes: the IME's own inset while
+    /// it is open, the navigation-bar inset otherwise. Takes a plain
+    /// `f32` in the caller's own physical-pixel units rather than an
+    /// Android-specific insets type, so this crate stays usable from the
+    /// winit backend too, which has no navigation bar to report.
+    /// Rewrites the existing `Pad` in place (marking it dirty through the
+    /// ordinary `Widgets::get_mut` path) instead of swapping in a new one,
+    /// so the field's focus, selection and in-progress text are untouched.
+    pub fn set_bottom_inset(&self, rsc: &mut impl UiRsc, inset: f32) {
+        if let Some(pad) = rsc.ui_mut().widgets.get_mut(&self.outer_pad) {
+            pad.padding.bottom = Len::abs(inset);
+        }
+    }
 }
 
 /// Returns the composer plus its own bar as a **weak** id -- the caller
@@ -39,10 +83,20 @@ where
         .label("Message")
         .add(rsc);
 
-    let bar: WeakWidget = (field.pad(dp(12)).width(rest(1)),)
+    // One widget: an opaque bar sized to its own content (`.background`'s
-        .span(Dir::RIGHT)
+    // `Stack{child: 1}`, the header row's own idiom) wrapping the padded,
+    // height-capped field -- not a background rect and a field drawn as
+    // two independent siblings, which is what let the two disagree on
+    // where the bar actually was.
+    let content = field
+        .pad(dp(FIELD_PAD_DP))
+        .max_height(dp(APPROX_LINE_HEIGHT_DP * MAX_LINES + FIELD_PAD_DP * 2.0))
+        .scrollable()
+        .width(rest(1))
         .background(rect(UiColor::new(40, 40, 46, 255)))
         .add(rsc);
 
-    (Composer { field }, bar)
+    let outer_pad: WeakWidget<Pad> = content.pad(Padding::ZERO).add(rsc);
+
+    (Composer { field, outer_pad }, outer_pad)
 }

iris/transcript-ui/src/lib.rs

@@ -51,7 +51,7 @@ pub mod selection;
 use client_core::transcript_fold::TranscriptRow as FoldedRow;
 use iris::prelude::*;
 use selection::Selection;
-use std::{cell::RefCell, rc::Rc};
+use std::{cell::RefCell, rc::Rc, time::Instant};
 
 pub struct TranscriptScreen {
     /// The transcript's own `List` -- exposed so a caller can read
@@ -220,6 +220,43 @@ where
     })
     .add(rsc);
 
+    // The continuation of a row-started drag once it has committed and
+    // taken pointer capture on `list`'s own id (`row.rs`'s registration is
+    // only ever the gesture's first frame) -- registered once here, not
+    // once per row, since `DragGesture`'s single shared instance must see
+    // each frame of one gesture exactly once. `ctx.data.pos`/`size` are
+    // already relative to `list`'s own on-screen box (this is what it was
+    // registered against), which is exactly the viewport-pixel space
+    // `List::key_at`/`extent` work in, so the row-under-the-pointer is
+    // resolved from those instead of a per-row hit test.
+    {
+        let selection = selection.clone();
+        list.on(
+            CursorSense::Pressing(CursorButton::Left) | CursorSense::Drop,
+            move |ctx, rsc| {
+                let pos = ctx.data.pos;
+                let row = list(rsc).key_at(pos.y).and_then(|key| {
+                    let (top, bottom) = list(rsc).extent(key)?;
+                    Some((
+                        key,
+                        Vec2::new(pos.x, pos.y - top),
+                        Vec2::new(ctx.data.size.x, bottom - top),
+                    ))
+                });
+                selection.borrow_mut().drag(
+                    rsc,
+                    list,
+                    row,
+                    ctx.data.cursor.pos,
+                    ctx.data.sense,
+                    Instant::now(),
+                    ctx.data.render,
+                );
+            },
+        )
+        .add(rsc);
+    }
+
     let (composer, composer_bar) = composer::build_composer(rsc);
 
     let tree = (list.width(rest(1)).height(rest(1)), composer_bar)

iris/transcript-ui/src/row.rs

@@ -137,20 +137,26 @@ where
 
     field
         // `| CursorSense::unclick()` on top of the usual click-or-drag set
-        // -- the arbiter inside `Selection::drag` needs the release too,
+        // -- this row's own registration only ever needs to see a
-        // to go back to idle for the next press (`DragArbiter::release`).
+        // gesture's *first* frame (`PressStart`, or a `Pressing` that
+        // missed it -- `DragGesture::handle`'s idle-recovery branch); once
+        // it commits, `DragGesture` takes pointer capture on `list`'s own
+        // id and every further frame, including the terminal `Drop`,
+        // reaches `lib.rs`'s list-level registration instead -- see
+        // `iris::sense`'s pointer-capture doc for why that has to be a
+        // stable id rather than this row's, which `List` can retire mid-
+        // drag as content scrolls.
         .on(
             CursorSense::click_or_drag() | CursorSense::unclick(),
             move |ctx, rsc| {
                 selection.borrow_mut().drag(
                     rsc,
                     list,
-                    key,
+                    Some((key, ctx.data.pos, ctx.data.size)),
-                    ctx.data.pos,
-                    ctx.data.size,
                     ctx.data.cursor.pos,
                     ctx.data.sense,
                     Instant::now(),
+                    ctx.data.render,
                 );
             },
         )

iris/transcript-ui/src/selection.rs

@@ -36,17 +36,15 @@ use std::{collections::BTreeMap, time::Instant};
 pub struct Selection {
     rows: BTreeMap<RowKey, WeakWidget<TextEdit>>,
     anchor: Option<(RowKey, Vec2)>,
-    /// One arbiter shared by every row's drag handler -- RUST.md's I5
+    /// One gesture shared by every row's drag handler -- RUST.md's I5
     /// gesture conflict (a row's own `click_or_drag()` and a list-level
     /// pan wanting the same touch gesture). See `drag` below, and
-    /// `iris::sense::DragArbiter`'s own doc for the decision itself.
+    /// `iris::sense::DragGesture`'s own doc for the arbitration, velocity
-    arbiter: DragArbiter,
+    /// tracking and pointer-capture mechanics this no longer owns itself
-    /// Tracks the last ~100ms of this gesture's pan deltas (in the same
+    /// -- Iris's 2026-09-06 ask (`IRIS.md`) that a drag's *mechanics* live
-    /// signed units `list.scroll` takes), so a release that turns out to
+    /// in iris's default input layer, with only the pan-vs-select
-    /// have been panning can hand `List::fling` a realistic initial
+    /// *decision* staying here.
-    /// velocity instead of one frame's noisy last delta --
+    gesture: DragGesture,
-    /// IRIS_TODO.md's "swiping has no momentum."
-    velocity: VelocityTracker,
 }
 
 impl Default for Selection {
@@ -60,8 +58,7 @@ impl Selection {
         Self {
             rows: BTreeMap::new(),
             anchor: None,
-            arbiter: DragArbiter::new(),
+            gesture: DragGesture::new(),
-            velocity: VelocityTracker::new(),
         }
     }
 
@@ -165,85 +162,66 @@ impl Selection {
     /// row, is what makes that consistent as a drag crosses row
     /// boundaries).
     ///
-    /// `pos_row`/`size` are row-local, as `begin`/`extend` want;
+    /// `row`, if given, is `(key, pos_row, size)` for whichever row the
+    /// pointer is currently over -- row-local, as `begin`/`extend` want.
+    /// `None` once the gesture is pointer-captured (`iris::sense`'s
+    /// pointer-capture doc) and the current position falls outside every
+    /// row `List` has loaded (a gap, or off the end of the content); a
+    /// `Pan` outcome never needs it, so this only actually matters mid-
+    /// selection, where it is rare and the frame is simply dropped.
     /// `pos_window` is in window space, since a pan's delta has to stay
     /// meaningful even when this frame's event landed on a different row
-    /// than the last one.
+    /// than the last one. `render` is `CursorData`'s own field -- what
+    /// `DragGesture` needs to take pointer capture.
     #[allow(clippy::too_many_arguments)]
     pub fn drag(
         &mut self,
         ui: &mut impl UiRsc,
         list: WeakWidget<List>,
-        key: RowKey,
+        row: Option<(RowKey, Vec2, Vec2)>,
-        pos_row: Vec2,
-        size: Vec2,
         pos_window: Vec2,
         sense: CursorSense,
         now: Instant,
+        render: &UiRenderState,
     ) {
-        let outcome = match sense {
+        if matches!(sense, CursorSense::PressStart(_)) {
-            CursorSense::PressStart(_) => {
-                let already_selected = self.has_selection(ui);
-                self.arbiter.press_start(pos_window, now, already_selected);
-                self.velocity.reset();
             // A fresh touch-down cancels any fling still coasting from
             // the previous gesture -- `List::fling`'s own doc, and
             // Android's `Scroller::abortAnimation` for the same reason.
             list(ui).cancel_fling();
-                self.arbiter.update(pos_window, now)
         }
-            CursorSense::PressEnd(_) => {
-                // A fling only ever follows a pan -- never a selection
-                // that happened to end with the finger still moving, and
-                // never a tap/long-press that never left `Undecided`.
-                if self.arbiter.is_panning() {
-                    let v = self.velocity.velocity();
-                    list(ui).fling(v);
-                }
-                self.arbiter.release();
-                return;
-            }
-            // A `Pressing` frame with the arbiter still `Idle` means this
-            // gesture's `ACTION_DOWN` landed somewhere no row's sensor
-            // covers (a row's own padding/gap, or a header with no
-            // selection handler) and this row is only now getting the
-            // touch as it moves across it -- the touch is definitely still
-            // down (that's what `Pressing` means), so without this the
-            // arbiter would sit in `Idle` answering `Undecided` for the
-            // rest of the gesture (`DragArbiter::update`'s own doc).
-            // Recovered by starting the press here instead of where it
-            // was missed -- RUST.md's I5 intermittent-touch-scroll-dropout
-            // finding, 2026-09-05.
-            _ if self.arbiter.is_idle() => {
         let already_selected = self.has_selection(ui);
-                self.arbiter.press_start(pos_window, now, already_selected);
+        let outcome =
-                self.velocity.reset();
+            self.gesture
-                list(ui).cancel_fling();
+                .handle(render, list.id(), sense, pos_window, now, already_selected);
-                self.arbiter.update(pos_window, now)
-            }
-            _ => self.arbiter.update(pos_window, now),
-        };
         match outcome {
-            DragOutcome::Undecided => {}
+            GestureOutcome::Undecided => {}
-            DragOutcome::Pan(dy) => {
+            GestureOutcome::Pan(dy) => list(ui).scroll(-dy),
-                let amt = -dy;
+            GestureOutcome::SelectStart => {
-                self.velocity.add_sample(amt, now);
+                if let Some((key, pos_row, size)) = row {
-                list(ui).scroll(amt);
+                    // Grep-able on "iris selection" the way the frame
-            }
+                    // report is on "iris frame report" -- selection has no
-            DragOutcome::SelectStart => {
+                    // accessibility label of its own yet, so this is the
-                // Grep-able on "iris selection" the way the frame report is
+                    // smallest way to confirm a real on-device long-
-                // on "iris frame report" -- selection has no accessibility
+                    // press-then-drag actually reached here (RUST.md's I5
-                // label of its own yet, so this is the smallest way to
+                    // box, "Measurements taken" (c)).
-                // confirm a real on-device long-press-then-drag actually
-                // reached here (RUST.md's I5 box, "Measurements taken" (c)).
                     log::info!("iris selection: begin at row {key:?}");
                     self.begin(ui, key, pos_row, size);
                 }
-            DragOutcome::SelectExtend => {
+            }
+            GestureOutcome::SelectExtend => {
+                if let Some((key, pos_row, size)) = row {
                     log::info!("iris selection: extend to row {key:?}");
                     self.extend(ui, key, pos_row, size);
                 }
             }
+            // A fling only ever follows a pan -- never a selection that
+            // happened to end with the finger still moving, and never a
+            // tap/long-press that never left `Undecided` -- exactly what
+            // `DragGesture`'s `Some(v)` already encodes.
+            GestureOutcome::Released(Some(v)) => list(ui).fling(-v),
+            GestureOutcome::Released(None) => {}
+        }
     }
 
     /// The concatenated selected text, in row order, `None` if nothing is
@@ -344,8 +322,9 @@ mod tests {
 
         let mut sel = Selection::new();
         sel.register(1, field);
-        assert!(sel.arbiter.is_idle());
+        assert!(sel.gesture.is_idle());
 
+        let render = UiRenderState::new();
         let now = Instant::now();
         let size = Vec2::new(100.0, 20.0);
         // No `PressStart` is ever sent -- only the `Pressing` frames a
@@ -353,15 +332,14 @@ mod tests {
         sel.drag(
             &mut rsc,
             list,
-            1,
+            Some((1, Vec2::ZERO, size)),
-            Vec2::ZERO,
-            size,
             Vec2::new(540.0, 700.0),
             CursorSense::Pressing(CursorButton::Left),
             now,
+            &render,
         );
         assert!(
-            !sel.arbiter.is_idle(),
+            !sel.gesture.is_idle(),
             "a Pressing frame with the arbiter still Idle must recover \
              the press rather than leaving it stuck"
         );