# Evidence capture 4 — three requested rows

PR [#33042](https://git.cafe/pingdotgg/t3code/pull/33042), GitHub mirror [#33042](https://github.com/pingdotgg/t3code/pull/33042). After [20ef152](https://git.cafe/pingdotgg/t3code/commit/20ef1526935674e44d09707db50699444a767728), before [d313b2a](https://git.cafe/pingdotgg/t3code/commit/d313b2aff31991ab94c8ac80cdc38a54dd575604).

AI-assisted; verified with the tests and tools listed above.

| Requested row | Result |
|---|---|
| A19-completed | Running and completed stills, both recordings, Diff0 and DiffT16; zero changed pixels outside sidebar. |
| A14-release | Both recordings release a real held Replay queue and show the same run starting on the server and normal Working status in the UI. |
| A02-A Android | Both recordings show Un-settle for both imported threads and collapse/reopen Working; stable endpoint stills and both diffs included. |

All sixteen accepted source stills are pixel-identical to their second capture, taken at least 500 ms later. Web viewport: 1440×900 CSS pixels, dark, DPR 2 (2880×1800 captures). Android: Pixel 9, 1080×2424, dark, verified font scale 1.2999999523162842. UI relative clock: 2026-11-10 09:00 UTC. Reduced motion enabled. Synthetic imports only; every server used T3CODE_DEV_SEED=0 and its own fresh home. No live user database or paid provider turn was used.

## A19-completed — HARNESS

The mounted full projection contains synthetic silent-compaction activity and the shell omits that run. This is a disclosed projection harness, not an executed compaction. Before remains falsely Waiting in Working; after shows the imported thread with its 3h age outside Working. Running and completed captures differ only in the sidebar across revisions.

Running stills: ![Before and after — shell-hidden silent compaction running; HARNESS](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9e642c1160d0e6b5/A19-running-HARNESS-pair.png)

Completed stills: ![Before and after — shell-hidden silent compaction completed; HARNESS](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9b35d0a85ac8b4b7/A19-completed-HARNESS-pair.png)

Running diffs: [Diff0 amplified](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/635ccaeaaf249794/A19-running-HARNESS-diff0.png) · [DiffT16](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b63bc0f6364302c4/A19-running-HARNESS-diffT16.png). Completed diffs: [Diff0 amplified](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/de7cf25cd9aaf03b/A19-completed-HARNESS-diff0.png) · [DiffT16](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f6de00b8e6f3b7d2/A19-completed-HARNESS-diffT16.png).

Diff0: 115,562 pixels; DiffT16 (>16 max RGB channel delta): 115,433 pixels, for each state. Bounding box [16,204,494,1626); outside sidebar (x≥512): **0 pixels**. No masks, registration or image smoothing.

Recording timestamps (UTC): before 2026-11-10T15:32:53.332Z → 2026-11-10T15:32:58.382Z; after 2026-11-10T15:33:29.609Z → 2026-11-10T15:33:35.234Z.

[Before recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e6f2d96b321b4d27/before-A19-HARNESS.mp4) · [After recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b8f3142638d0b3d5/after-A19-HARNESS.mp4)

## A14-release — HARNESS boundary, real queue release

The imported thread is bound (historyOrigin=v1_import). The sole nonterminal run is queued and held. The visible HARNESS restricts the initial mounted full projection to that real held run and omits its shell run. Before falsely shows Waiting; after has no Working shelf entry for the import. At +1.8s both recordings call the normal queue.resume command through the app. The free Replay adapter emits actual tool/text events and both sidebars become normal blue Working 0s.

![Before and after — bound imported thread with only a held queued run; HARNESS](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/dcecff789aa067b7/A14-held-HARNESS-pair.png)

[Diff0 amplified](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3d0469169d88e782/A14-held-HARNESS-diff0.png) · [DiffT16](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ba31a22d2b8dbb7d/A14-held-HARNESS-diffT16.png). Diff0: 115,562 pixels; DiffT16: 115,433 pixels; bounding box [16,204,494,1626), no changes outside sidebar.

The bundle contains before/after held and started server snapshots. Each proves that the same queued run becomes running, queueHeld=false, and startedAt is set: before ordinal 7 started at 15:51:55.802Z; after ordinal 5 started at 15:50:11.214Z. These are server observations, not a changed status flag in the harness. The composer model picker is synthetic presentation; the executing adapter is Replay.

Recording timestamps (UTC): before 2026-11-10T15:51:53.943Z → 2026-11-10T15:52:12.208Z; after 2026-11-10T15:50:09.361Z → 2026-11-10T15:50:29.638Z.

[Before recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8a5418172e37356b/before-A14-release-HARNESS.mp4) · [After recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/397b25e976e1b530/after-A14-release-HARNESS.mp4)

## A02-A — Android Pixel 9, font scale 1.3

The same driver long-presses the primary import, selects Un-settle, repeats for the long-title import, collapses Working, then reopens it. Before both imports incorrectly join Working (3 threads including the watch). After both imports remain outside Working, ordered by their real Un-settle activity; the surviving watch remains in Working (1 thread). The watch and synthetic branch are explicitly injected shell fixtures; Un-settle is performed through the real native UI and server RPC.

![Before and after — Android imported threads un-settled, with surviving Working shelf expanded](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b1697d35c67ac53c/A02-A-android-pair.png)

[Diff0 amplified](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/dd09402efec9ce8f/A02-A-android-diff0.png) · [DiffT16](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/109216542305ec94/A02-A-android-diffT16.png). Diff0: 120,599 pixels, bbox [46,304,1032,999); DiffT16: 95,851 pixels, bbox [46,304,1031,999). Changes are confined to thread rows and their shelf. Header, composer action and navigation remain identical.

Recording timestamps (UTC): before 2026-11-10T15:55:02.117473+00:00 → 2026-11-10T15:55:16.228192+00:00; after 2026-11-10T15:49:46.474663+00:00 → 2026-11-10T15:50:10.922960+00:00.

[Before recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3b9b4170d18a6976/before-A02-A-android.mp4) · [After recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/920d7449d8f62e2c/after-A02-A-android.mp4)

Android encoded durations: before 14.946s, after 25.067s. The action ledgers contain UTC start/end timestamps for each step. The final encoded frame of each recording was inspected and shows Working expanded. Recordings retain the real OS status-bar clock; stills normalize it to 9:41.

## What else changed

- The web provider readiness/model picker is the same synthetic presentation on both sides. Only Replay executes the queue-release turn. The supported pill environment-identification setting is identical on both sides; it avoids unrelated low-amplitude artwork rendering deltas.
- A14 releases restore terminal setup history in the chat. Retakes left different counts of terminal fixture runs (tested ordinals 7 before, 5 after); their live timestamps and minor streamed-content timing differ. This does not change the one held nonterminal run at the boundary. Live keyframes in the bundle are documentary frames, not accepted stable stills.
- Android row order, title wrapping, dimming, Working membership and age labels change as part of the corrected status presentation. The long title uses the app’s intended two-line ellipsis at font scale 1.3. The orange provider marks and erroneous before Waiting indicators are the behavior being demonstrated, not stale loading overlays.
- Android system clock and radio indicators vary in the recordings. Normalized stills have identical system chrome. Both revisions share the same native fingerprint and load JavaScript from the matching checkout’s Metro server.
- No product source change was made. Both temporary relative-clock patches were restored before removing the worktrees.

## Files, tools and cleanup

[Complete evidence bundle (captures, diffs, recordings, scripts, metrics, SHA-256 manifest and reproduction recipe)](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/036b9073365e79cc/evidence-bundle.tar.gz). Every raw a/b still, video contact sheet, server snapshot, used driver and cleanup verification is included. `superseded-*` and `android-current.png` are exploratory/rejected captures retained for provenance and are not proof. `reproduction.md` documents the precise harness boundaries and normal RPC actions.

Tools: vp install, vp run dev, free Replay (dev), T3 Browser preview/snapshot/recording, server typed RPC, device-pool/agent-device, mobile:install, Metro inspector, Python/Pillow/NumPy exact pixel comparison, ffmpeg/ffprobe, and ghdrop. No repository code check or CI/review audit was requested in this capture pass.

Cleanup verified: all nine owned ports closed; all owned server/Metro process groups stopped; pool session closed; preview tab closed; /home/user/wt/pr-33042-before4 and /home/user/wt/pr-33042-after4 removed. The protected /home/user/wt/pr-33042 was untouched. Synthetic scratch and uploaded evidence remain under the authorized scratch directory.

No requested row is missing.