diff --git a/RUST.md b/RUST.md
index 0ccc49d..ff0493d 100644
--- a/RUST.md
+++ b/RUST.md
@@ -436,6 +436,52 @@ accepted.
       builds with `cargo-ndk` + Gradle, renders on the GPU, and the phone's
       own keyboard types into its editor with autocorrect and suggestions.
       Note which `wgpu` backend it took and any environment flags needed.
+      **Part-done 2026-09-04; the keyboard half is untested, so the box
+      stays open.**
+
+      *Builds and renders.* `~/src/android-view` at `bec6c62`, built with
+      `cargo ndk -t x86_64 -P 26 -o masonry-app/src/main/jniLibs/ build -p
+      android-view-masonry-demo` and `./gradlew :masonry-app:assembleDebug`.
+      The demo draws: its text field and "Add task" button appear, laid out
+      correctly. Note the demo's README says `arm64-v8a`; this emulator is
+      x86_64. **`libmain.so` is 181 MB in debug and 11 MB in release**,
+      which is the single loudest number about Vello's dependency graph.
+
+      *Accessibility works, which E2's condition 6 depends on.* `ui-trace`
+      reads Masonry's AccessKit tree: "Add task" comes through as a named
+      `Button` and the editor as an `EditText` node. So the bench rig's
+      tap-by-name would work against a Masonry screen — for controls that
+      carry a name. The demo's editor does not, so a coordinate was needed
+      for it, which is a demo omission rather than a framework one.
+
+      *The keyboard was not reached.* Tapping the editor never brought the
+      IME up (`mInputShown=false` throughout) before the run ended. Nothing
+      here says android-view's `InputConnection` does not work — only that
+      it has not been shown to. This is the half of E1 that matters most,
+      since it is the constraint the whole framework decision turns on, and
+      it is what to do first when this is picked up.
+
+      *A crash worth knowing about, seen once.* With an accessibility
+      client attached the app aborted, and the stack attributes it
+      precisely: `android_view::view::do_frame` →
+      `CallbackCtx::finish` → `accesskit_android::event::QueuedEvents::raise`
+      → `send_completed_event` → `unwrap()` on `Err(JavaException)`.
+      android-view builds with `panic = "abort"`, so a JNI call that throws
+      takes the process. It did **not** reproduce: two plain
+      `ui-trace record` runs afterwards left the app alive, so the trigger
+      is narrower than "an accessibility client is connected" and is not
+      yet known. Recorded rather than chased because it is upstream and
+      one unwrap wide.
+
+      *Two changes were made at once and cannot be separated.* The first
+      launches crashed in `Surface::configure` ("Invalid surface") on GLES,
+      then, with Vulkan enabled, segfaulted inside the emulator's own
+      driver (`vulkan.ranchu.so`, `SetDebugUtilsObjectNameEXT`) while wgpu
+      labelled an image view — a debug-build-only call. It then ran after
+      **both** a switch to a release native library and a switch to a
+      software-rendered emulator. Which of the two fixed it is unmeasured;
+      whoever picks this up should vary one at a time rather than inherit
+      the assumption.
 - [ ] **E2 — a transcript in Masonry.** One screen: open a sandbox session,
       page 800 events into `VirtualScroll` bottom-anchored, draw markdown
       from `pulldown-cmark` into Parley rich text with links and code
diff --git a/iris/src/default/render.rs b/iris/src/default/render.rs
index 77829cc..b317498 100644
--- a/iris/src/default/render.rs
+++ b/iris/src/default/render.rs
@@ -115,12 +115,10 @@ impl UiRenderer {
             format: surface_format,
             width: size.width,
             height: size.height,
-            // Vsync, because a UI toolkit aiming at battery life must not
-            // present frames a display will never show. AutoNoVsync accepts
-            // them as fast as the GPU will take them, so a redraw burst costs
-            // whatever the hardware can be made to do rather than one frame --
-            // and on this machine the GPU is the host's real one, reached
-            // through virtio-gpu, so that cost lands on somebody's desktop.
+            // Vsync, because a toolkit aiming at battery life must not present
+            // frames a display will never show: AutoNoVsync accepts them as
+            // fast as the GPU will take them, so a redraw burst costs whatever
+            // the hardware can be made to do rather than one frame.
             // AutoVsync picks Fifo, which every backend supports.
             present_mode: PresentMode::AutoVsync,
             alpha_mode: surface_caps.alpha_modes[0],