# Remaining-row evidence for [#24651](https://git.cafe/pingdotgg/t3code/pull/24651)
Unrelated visible differences: live recording clocks, row ages, generated prose, retained interrupted history and Working durations differ. The after iPad recording also has an iOS return-to-AgentDeviceRunner status-bar breadcrumb. Settled full-device diffs retain native chrome/battery differences, quantified below. No new product difference or speed/FPS conclusion is claimed from these recordings.
Captured **3/3 requested remaining matrix rows, 5/5 required new recordings**, plus two supplementary iPad column-control clips and four settled end-state stills. A03 before is retained from the prior index, as requested. All other matrix rows retain the lead's prior SUFFICIENT judgment and were not redone. This report does not re-audit those rows.
Before: [96fc7b9](https://git.cafe/pingdotgg/t3code/commit/96fc7b955196637605cbf3b415152ce4dd0e2294). After: [4859c3a](https://git.cafe/pingdotgg/t3code/commit/4859c3a7d372dba06a50adee68d157643a305a10). Source: [pingdotgg/t3code](https://git.cafe/pingdotgg/t3code), [GitHub mirror #24651](https://github.com/pingdotgg/t3code/pull/24651). The earlier instrumentation patch was not applied. Both real checkouts ran their own Metro bundle and fresh isolated server state with T3CODE_DEV_SEED=0; no live database was copied. Streams are **real Codex gpt-6.1-sol runs, Low reasoning**, with the identical 600-paragraph garden prompt, rather than deterministic synthetic ticks. Existing fixed-input harness counts remain the prior evidence.
## Rendering environment
Native clients installed through mobile:install and captured with agent-device 0.21.32; Hermes/Metro CDP drives existing preferences, pairing and occasional navigation, plus read-only diagnostics. No browser/device emulation through a browser and no component harness. Both native builds have React Compiler enabled. Dark system and app theme, fontScale 1.0; named pool simulators/emulators, not physical devices.
| Row | Pool device and OS | Native viewport | DPR and full capture | Orientation |
|---|---|---|---|---|
| A02 both | iPad Pro 13-inch (M5), iPadOS 27.2 simulator | 1376×1032pt | 2; 2752×2064 displayed px | Landscape |
| A04 both | iPhone 17 Pro, iOS 27.2 simulator | 402×874pt | 3; 1206×2622px | Portrait |
| A03 after | Pixel Tablet, Android 16/API36 emulator (reports sdk_gphone64_x86_64) | 1280×800dp | 2; 2560×1600px | Natural landscape |
The iPad recorder encodes 2064×2752 with rotation=90; players and contact sheets display 2752×2064 landscape. iOS native fingerprint f532a0c5902b69c7f69cc5fe9c807053700a07eb; Android ed36a9cc9856c84a322bb71e6d67744f0a00877d. Both sides prepared the matching cached native client; no native source changes.
VM fs-15.tail131df4.ts.net. Before server/Vite/Metro: **15881 / 7841 / 8081**. After: **16101 / 8061 / 8081**. Only one pair ran at a time. All owned process groups stopped and device leases closed; disposable state/worktree removed, after source remains clean.
Command-start timestamps below come from driver.json relative to record-start completion; native recorder startup and UI transitions can shift the corresponding visual change by about 1–2 seconds. Contact-sheet labels are exact media sample times.
## Evidence: A02 — iPad Pro 13-inch (M5), 1376×1032pt, landscape, dark
The actual native primary column is established by isPad=true, native support=true and a mounted NativePrimaryColumnContext with presented=true and the selected B key on both sides. [Native metrics and mounted-path proof](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0e4ecb55d99bafc6/native-primary-column-proof.json). No generic-sidebar override.
**Same scripted interaction, approximate command-start timestamps:** begin with streaming A and selected B. Before: 00:03 select A; 00:04.7→00:13.5 projection/text progress (324→1,494 chars); 00:19.7 select B; 00:21.1 RPC rename A; 00:24.6 stop; Done visible by 00:28.4. After: 00:03 select A; 00:04.4→00:13.2 text progress (7,016→8,165 chars); 00:16.5 select B; 00:18.3 RPC rename; 00:22.1 stop; Done by 00:25.9. The thread list stays in the native column throughout. Driver/network timing differences are not a speed comparison.
| Before | After | Diff0, 255× | DiffT, >8/255, 255× |
|---|---|---|---|
| [A02-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d523261e519a0588/A02-before.mp4)
[96fc7b9](https://git.cafe/pingdotgg/t3code/commit/96fc7b955196637605cbf3b415152ce4dd0e2294) | [A02-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/18914bb20006c4d6/A02-after.mp4)
[4859c3a](https://git.cafe/pingdotgg/t3code/commit/4859c3a7d372dba06a50adee68d157643a305a10) | 
[96fc7b9](https://git.cafe/pingdotgg/t3code/commit/96fc7b955196637605cbf3b415152ce4dd0e2294) → [4859c3a](https://git.cafe/pingdotgg/t3code/commit/4859c3a7d372dba06a50adee68d157643a305a10) | 
[96fc7b9](https://git.cafe/pingdotgg/t3code/commit/96fc7b955196637605cbf3b415152ce4dd0e2294) → [4859c3a](https://git.cafe/pingdotgg/t3code/commit/4859c3a7d372dba06a50adee68d157643a305a10) |
Supplementary end-state controls: both press **Maximize content**, then **Show thread sidebar**. Before ~00:05 hide / ~00:09 reveal; after ~00:01 hide / ~00:10 reveal. These occur after the stream stopped.
| Before | After |
|---|---|
| [A02-before-columns.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bd941afe41c469ac/A02-before-columns.mp4)
[96fc7b9](https://git.cafe/pingdotgg/t3code/commit/96fc7b955196637605cbf3b415152ce4dd0e2294) | [A02-after-columns.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d19d546b78ff6442/A02-after-columns.mp4)
[4859c3a](https://git.cafe/pingdotgg/t3code/commit/4859c3a7d372dba06a50adee68d157643a305a10) |
## Evidence: A04 — iPhone 17 Pro, 402×874pt, portrait, dark
The real Arrange threads sheet is opened through the row menu and scrolled to A before recording. It remains mounted throughout the RPC rename and subsequent real stream updates. No close/reopen or snapshot overlay supplies the renamed title.
**Same scripted interaction, approximate command-start timestamps:** before: 00:03.2 RPC rename; updated title visible by ~00:05.9; later stream projections at 00:06.5→00:19.4 grow 1,056→2,202 chars; stop at 00:20.9. After: 00:03.0 RPC rename; title visible by ~00:05.8; projections at 00:06.5→00:19.4 grow 9,535→11,294 chars; stop at 00:21.0. The sheet stays open with A visible. Before's extra read-only probes show sheetOpen=true before and after the interval. The same after-side continuity is visible in its recording. Both close normally after their end-state captures.
| Before | After | Diff0, 255× | DiffT, >8/255, 255× |
|---|---|---|---|
| [A04-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/908e9debfca8fd55/A04-before.mp4)
[96fc7b9](https://git.cafe/pingdotgg/t3code/commit/96fc7b955196637605cbf3b415152ce4dd0e2294) | [A04-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f994379365af6017/A04-after.mp4)
[4859c3a](https://git.cafe/pingdotgg/t3code/commit/4859c3a7d372dba06a50adee68d157643a305a10) | 
[96fc7b9](https://git.cafe/pingdotgg/t3code/commit/96fc7b955196637605cbf3b415152ce4dd0e2294) → [4859c3a](https://git.cafe/pingdotgg/t3code/commit/4859c3a7d372dba06a50adee68d157643a305a10) | 
[96fc7b9](https://git.cafe/pingdotgg/t3code/commit/96fc7b955196637605cbf3b415152ce4dd0e2294) → [4859c3a](https://git.cafe/pingdotgg/t3code/commit/4859c3a7d372dba06a50adee68d157643a305a10) |
## Evidence: A03-after — Pixel Tablet, 1280×800dp, landscape, dark
**Existing before semantic chapter (approximate media timestamps):** ~00:57 sidebar scroll; ~00:59 RPC rename; ~01:03 open A; ~01:12 real text; ~01:23 B; ~01:24 stop; terminal by ~01:27. Its earlier synthetic harness prefix and shorter rename title are retained, not regenerated. **New after (approximate command-start timestamps):** starts connected with Working A and B selected; 00:03.0 / 00:04.3 sidebar scrolling; 00:05.7 RPC long-title rename; 00:10.7 open A; 00:11.6→00:20.4 real text grows 1,769→2,841 chars; 00:24.2 return B; 00:24.7 stop; Done by 00:28.6. No reconnecting/syncing banner appears in the curated take. The connected brand region remains unchanged across all 176 encoded frames: [A03-connected-frame-check.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8b305f0827690991/A03-connected-frame-check.json).
| Before | After | Diff0, 255× | DiffT, >8/255, 255× |
|---|---|---|---|
| [before-android-sidebar.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d4d6d7d214670fb6/before-android-sidebar.mp4)
[96fc7b9](https://git.cafe/pingdotgg/t3code/commit/96fc7b955196637605cbf3b415152ce4dd0e2294) | [A03-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/69d80f88cf144d9a/A03-after.mp4)
[4859c3a](https://git.cafe/pingdotgg/t3code/commit/4859c3a7d372dba06a50adee68d157643a305a10) | After-only replacement; no new still pair requested | After-only replacement; prior static diffs retained |
## Timestamped contact sheets
These show six full-surface frames, including the final state, preserving native display aspect ratio. They are review aids; recordings remain the interaction evidence.
| Row | Before sheet | After sheet |
|---|---|---|
| A02 |  |  |
| A04 |  |  |
| A03-after | Existing before retained |  |
| A02 column controls |  |  |
## Streaming proof and what else changed
[Progress summary](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7bf8a0d92bf59fd1/progress-summary.json) links each authoritative take to its active runs and projection growth; complete driver/shell/projection JSON is in the sanitized raw archive. In the captured pre-stop intervals: A02 before 32 active samples / 28 unique updatedAt / counts≥11, after 29 / 29 / counts≥21; A04 before 27 / 26 / counts≥22, after 28 / 28 / counts≥11; A03 after 32 / 32 / counts≥16. All have non-null activeRunId and activityRunStartedAt. These numbers establish live streams, not comparative performance.
- **A02 still:** 14,983 / 5,680,128 pixels differ at threshold zero, max RGB delta **1**; **0** pixels exceed 8/255. Differences are concentrated at native glass/chrome edges, including the composer. Do not claim whole-device identity.
- **A04 still:** 2,896 / 3,162,132 pixels differ at zero, 421 exceed 8/255, max delta **57**. Every difference is inside the native charging-battery rectangle x1017..1099/y79..117. Every app pixel below the native status bar is identical at zero. [Inspection report](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9112fb7c5efa064f/inspection-report.json).
- **Recordings:** native clocks, relative age labels, generated prose, prior interrupted transcript history, run ordinals/counts, stream age/Working durations and action completion times differ. A02 after has the native return-to-AgentDeviceRunner breadcrumb; it is absent from the paired stills. Screenshot normalization briefly changes native clock/battery chrome during portions of the videos, including overlap during original after-side iOS captures. A03's retained before also has its older harness prefix and shorter renamed title. These are evidence-condition differences outside the PR's selector change.
- The native platform/theme/font scale/viewport and same 12-thread seed plus default New thread row match. A04's sheet scroll is at the same bottom position; A02's end-state column shows the same full list and B selection. No app-pixel semantic difference is attributed to the PR by these stills.
Diff0 and DiffT8 are original full-resolution RGB absolute differences amplified **255×**, so a one-level difference is visible; DiffT8 retains a pixel if any channel differs by **more than 8/255**. No crop or edited status-bar pixels replaces a full-surface original. Each side's authoritative end-state still has an exact pixel-identical pair captured ≥500ms apart. All curated stills/diffs and full-surface video keyframes/contact sheets were opened for inspection.
## Unverified conditions
No requested interaction row is missing a recording. **Frozen-clock video matching is not fulfilled:** app/native clocks and provider timing remain live in the recordings; only settled stills freeze app time and normalize native status bars. Native network-idle/font-ready and reduced-motion APIs are not exposed by the pool tooling. Repeated identical stills establish observed stability. No FPS, elapsed-speed, or whole-device zero-diff conclusion is made. These limitations must remain with the evidence.
## Rebuild and raw files
[REBUILD-r3.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ec8b1bc20d7b61cd/REBUILD-r3.md)
Main scripts: [rpc-driver.ts](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c45156e231681d25/rpc-driver.ts); [capture-run.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a3aaafa76ede02e4/capture-run.py); [cdp.mjs](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/978052868cefc368/cdp.mjs); [pair-via-cdp.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5167e202f5f7d5c2/pair-via-cdp.py); [freeze-clock.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c0858733c3c7af97/freeze-clock.js); [settled-capture.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/cfc30c8034bea309/settled-capture.py); [diff-pair.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6700792db5f2d886/diff-pair.py); [timestamped-sheets.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/445fb60df0ded8b4/timestamped-sheets.py). Additional script links and manifests are in the verbatim upload block below. The RPC driver is byte-identical on both sides. Private bearer files and full server configuration are excluded.
[raw-evidence-r3.tar.gz](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c52b41314be8ff1a/raw-evidence-r3.tar.gz)
[raw-archive-audit.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/42d4697ab7fba586/raw-archive-audit.json)
[curated-manifest.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6aa8d4d63761f091/curated-manifest.json)
[cleanup-report.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8450379da1d3900b/cleanup-report.json)
Raw archive: 293 files, 68,380,571 compressed bytes, 30 credential/session-identifier redactions; known bearer values checked absent. Non-curated takes are rejected diagnostics, not alternate comparisons. The published manifest names the authoritative source directories. All seed/driver scripts, raw capture media and logs have been uploaded; none is committed.
## Friction
```text
FRICTION:
infra | Intermittent DNS and remote-daemon failures interrupted early takes and required retries | Reliable pool/DNS service and a pre-record connection health check.
tool | Natural tablet orientation, rotated iPad recording metadata and iOS breadcrumb/battery normalization needed capture corrections | Report display orientation and normalize native status chrome consistently.
runbook | Installing the baseline from its own cwd conflicted with the lease's after-worktree ownership | Document using the baseline installer entry point from the lease-owning cwd.
model | A finished run and an install-reset detail selection invalidated early takes; both were caught and retaken | Preflight the active run, connected status and actual selected detail immediately before recording.
```
No source edits, commits, push or PR-body edit. Both leases and owned server/Metro groups are stopped; the disposable baseline worktree and after dev state are removed. Scratch is removed after this index is uploaded. Capture agent: gpt-6.1-sol, Codex harness. Checks/reviews not assessed.
## Verbatim ghdrop output
The final REBUILD-r3.md link in the correction block supersedes its first upload; all other curated links are unchanged.
[A02-after-columns.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d19d546b78ff6442/A02-after-columns.mp4)
[A02-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/18914bb20006c4d6/A02-after.mp4)
[A02-before-columns.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bd941afe41c469ac/A02-before-columns.mp4)
[A02-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d523261e519a0588/A02-before.mp4)
[A02-diff.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2618e4af97671a20/A02-diff.json)
[A03-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/69d80f88cf144d9a/A03-after.mp4)
[A03-connected-frame-check.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8b305f0827690991/A03-connected-frame-check.json)
[A04-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f994379365af6017/A04-after.mp4)
[A04-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/908e9debfca8fd55/A04-before.mp4)
[A04-diff.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b9d6f448dc43510e/A04-diff.json)
[REBUILD-r3.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/28d9990709b9d3a7/REBUILD-r3.md)
[capture-run.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a3aaafa76ede02e4/capture-run.py)
[cdp.mjs](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/978052868cefc368/cdp.mjs)
[column-toggle.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/77aa0bce743b75fa/column-toggle.py)
[contact-sheets.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/08f113bfd04115f7/contact-sheets.py)
[curated-manifest.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6aa8d4d63761f091/curated-manifest.json)
[diff-pair.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6700792db5f2d886/diff-pair.py)
[freeze-clock.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c0858733c3c7af97/freeze-clock.js)
[native-path.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/282c733a83542f97/native-path.js)
[native-primary-column-proof.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0e4ecb55d99bafc6/native-primary-column-proof.json)
[pair-via-cdp.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5167e202f5f7d5c2/pair-via-cdp.py)
[progress-summary.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7bf8a0d92bf59fd1/progress-summary.json)
[rpc-driver.ts](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c45156e231681d25/rpc-driver.ts)
[select-environment.mjs](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/cec22436d4f5f179/select-environment.mjs)
[settled-capture.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/cfc30c8034bea309/settled-capture.py)
[start-process.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d6046de1951cf328/start-process.py)
[stop-process.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/acbd271dc24320f0/stop-process.py)
[summarize-progress.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/edbfb602064aa88f/summarize-progress.py)
[timestamped-sheets.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/445fb60df0ded8b4/timestamped-sheets.py)
[verify-connected.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/1fabf968a5e079ba/verify-connected.py)
[raw-evidence-r3.tar.gz](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c52b41314be8ff1a/raw-evidence-r3.tar.gz)
[raw-archive-audit.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/42d4697ab7fba586/raw-archive-audit.json)
[REBUILD-r3.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ec8b1bc20d7b61cd/REBUILD-r3.md)
[inspection-report.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9112fb7c5efa064f/inspection-report.json)
[cleanup-report.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8450379da1d3900b/cleanup-report.json)