# Row N evidence recapture — pass 2

For [#32905](https://git.cafe/pingdotgg/t3code/pull/32905), baseline [bcee0f5d28](https://github.com/pingdotgg/t3code/commit/bcee0f5d2872ba193c66ed9ba979bbe424f89ff9) → head [3d4b2607cd](https://github.com/pingdotgg/t3code/commit/3d4b2607cdb18e74f7db35a9059d0315606932fe). This is new N evidence, not a replacement for the already accepted [F/E before/after transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/db8d6ad0fc085944/F-E-before-after.txt). F/E were not rerun.

Captured 2026-11-10 UTC. Environment: web, Linux x86_64 (fs-54), Chromium 157, 1440×1000 CSS px, DPR 2 (2880×2000 files), dark, interface font 16px, normal zoom, no device emulation. Date remained native throughout; no clock freezing or jumps. Both sides use the same synthetic local-only repository and initially fresh isolated databases, with T3CODE_DEV_SEED=0 and explicit separate home directories. No live data was copied or migrated, no provider turn was sent, and no tracked source was changed.

## Result and loaded Changes proof

Both real-app recordings show the project **Local-only diagnostics**, empty thread **Local feature changes**, and loaded **Changes: feature → main**, with feature.txt expanded: +1/−0, `One local commit ahead of main.` The feature branch is one committed change ahead of local main, has no remote/upstream, and has a clean working tree. Two real authenticated status refreshes succeed while this view is visible. The actual Refresh diff control is then clicked; the local diff remains loaded before Settings opens. This fixes the first pass's missing client-flow proof.

Before's refreshed Diagnostics snapshot shows **11 failures**, a Latest Failures table of resolver errors (capped at ten), and Most Common Failures groups **6 + 4 + 1**. After shows **0 failures** and both failure-table empty states. The two explicit refresh calls alone produce three resolver spans each: before 0 → 3 → 6 failed; after 0 → 3 → 6 successful and zero failed. Status outputs agree on feature/main, ahead=1, behind=0, no remote/upstream, branch insertions=1, deletions=0.

Normal app status/review/background work also invokes the resolver. The final raw window, collected later than the visible snapshot, contains **15 failed / 0 successful** resolver spans before and **0 failed / 13 successful** after. Both explicit refresh RPC spans are successful on both sides. Final diagnostics reads are 1379 records/15 failures before and 1181/0 after. These differing background counts reflect different capture dwell times; they are not a performance comparison.

## Actual MP4 timeline

[Before recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6f78f3b4dde535ae/before-N.mp4), 118.485933 seconds, 67 encoded frames. [After recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a1f23fe5b447d3bb/after-N.mp4), 87.037833 seconds, 59 encoded frames. Original shared-browser MP4 files are unchanged. Every encoded frame was extracted with ffmpeg and visually inspected through complete labelled contact sheets, with full-resolution checks of the evidence frames. ffprobe presentation timestamps, not estimates from tool-call timing, determine visual offsets below.

| Step | Before PTS / frame | After PTS / frame |
| --- | --- | --- |
| Loaded local Changes already visible | 00:00.000 / 1 | 00:00.000 / 1 |
| Refresh diff click begins | 00:26.914433 / 4 | 00:32.992900 / 3 |
| Loaded local diff after refresh; click indicator settled | 00:28.736800 / 17 | 00:34.454500 / 17 |
| Settings General visible | 00:33.827567 / 33 | 00:37.808367 / 32 |
| Diagnostics route/resources visible | 01:22.671933 / 39 | 01:03.861267 / 34 |
| Final Trace Diagnostics heading, Failures and Latest Failures | 01:24.537933 / 58 | 01:06.644233 / 52 |
| Most Common Failures scrolled to top | 01:53.645733 / 65 | 01:23.451533 / 57 |
| Top Span Names (context only; resolver below ten-row cap) | 01:57.087033 / 66 | 01:26.263067 / 58 |

The status RPCs are offscreen commands while the loaded client view remains visible. Transcript UTC start times relative to recording-start metadata are 5.318s and 13.090s before, 2.715s and 16.913s after; these establish execution, not a visible click or video-frame boundary. The video provides loaded-client proof before and after them. Recording start metadata: before 10:50:31.797Z; after 10:55:18.539Z.

Initial resource-monitor loading resolves before accepted Diagnostics frames. Baseline video has visible lossy compression patches on dark backgrounds; text remains legible. The PNG stills are pristine and are the pixel-comparison sources. No interpolated frames, retiming, or fabricated video was used.

## Settled whole-viewport stills

Each accepted position has two snapshots separated by an explicit 600ms plus tool time. All four pairs are exactly pixel-identical at 2880×2000. The verification archive includes both copies, their hashes and settled-verification.json. Preliminary Changes stills had a changing refresh icon and baseline Top Span Names stills were unsettled during scrolling, so none of those preliminary PNGs are published as accepted stills.

Before — failures summary and Latest Failures:

![Before — 11 failures and resolver error rows](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8e3d35c6fa89459b/before-failures.png)

After — zero failures and both empty states:

![After — zero failures and empty failure tables](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/76b4d6b0f481f8fa/after-failures.png)

Before — Most Common Failures:

![Before — resolver failures grouped 6, 4 and 1](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3d6d196acb3546ac/before-common.png)

After — Most Common Failures:

![After — no repeated failures](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b714314c6dc81556/after-common.png)

## Scoped comparisons and algorithms

The clock ran normally. Comparisons are cropped to the affected Failures tile and failure tables; whole-viewport originals above retain context. Unrelated live counts, slow-span durations, trace IDs, ports and telemetry are outside these diffs. Coordinates below are CSS pixels (multiply by two for source pixels), [left, top, right, bottom]. No image scaling or registration is applied. Shorter after empty-state panels are placed at the same top-left on a black canvas; removed panel area remains visible as a change, without including the unrelated following Slowest Spans section.

| Region | Before source/crop | After source/crop | Pixels changed >0 / >16 |
| --- | --- | --- | --- |
| Failures tile | failures [596,128,840,200] | same | 633 / 606 of 70272 |
| Latest Failures, heading/header and first two representative rows | failures [351,241,1329,505] | failures [351,241,1329,329] | 701574 / 54586 of 1032768 |
| Most Common Failures, complete three groups | common [351,89,1329,445] | common [351,89,1329,179] | 1057757 / 78752 of 1392672 |

Each region has two exact before/after difference images: absolute RGB difference ×32, saturating at 255, with zero threshold; and a max-channel RGB difference **strictly greater than 16/255**, shown magenta on a 35%-brightness baseline. Zero-threshold images emphasize dark panel-background removal as well as text. Both algorithms show only the intended removal of failures, their rows, and panel compaction. This is not a full-page visual equivalence claim. Full crop PNGs and machine-readable statistics are in the verification archive.

Failures tile, zero threshold amplified / threshold 16:

![Failures tile — zero threshold amplified ×32](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/41a0904460528540/failure-tile-diff0-amplified.png)

![Failures tile — threshold greater than 16/255](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7e10fb0d5169138b/failure-tile-diff16.png)

Latest Failures, zero threshold amplified / threshold 16:

![Latest Failures — zero threshold amplified ×32](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5257748e26f5f426/latest-failures-diff0-amplified.png)

![Latest Failures — threshold greater than 16/255](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/018de2e3dea418bd/latest-failures-diff16.png)

Most Common Failures, zero threshold amplified / threshold 16:

![Most Common Failures — zero threshold amplified ×32](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/63c67e5d913db7ea/common-failures-diff0-amplified.png)

![Most Common Failures — threshold greater than 16/255](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7ce0f92f38c2bd34/common-failures-diff16.png)

## What else changed

- Expected: Failures 11 red → 0 neutral; resolver error rows and common groups disappear; empty states replace them and compact the page.
- Unrelated whole-page values: total spans 1016 → 903, slow spans 7 → 6, parse errors stay zero. Later raw totals differ as disclosed above. Slow-span names, durations, ordering, trace IDs and relative-age labels vary with real execution time. No performance claim is made.
- Whole-page port labels differ (web 8484 → 6408); native CPU/memory/resource counters and sidebar relative ages also vary. These are excluded from cropped comparisons.
- Recording dwell times differ, including lazy Settings loading and the after route retry. The browser click indicator animates. No clock jump creates artificial multi-thousand-second spans.
- Settings navigation labels this route Maintenance while the breadcrumb says Diagnostics; this is existing app behavior.

## Downloads, reproducibility and limits

- [Before runtime transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/fc1ed516c8e8e829/before-N-runtime.txt), [after runtime transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0e12e70ec8c9a65b/after-N-runtime.txt).
- [Updated scripts and runbook](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e5b74111d5105daf/N-scripts.zip): seed.py, rpc-driver.mjs, trace-window.py, trace-summary.py, assemble-evidence.py, inspect-recordings.py, capture-runbook.md, with hashes. The public empty-thread driver takes fixture provider/model values from environment variables rather than publishing operational identifiers.
- [Verification archive](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c0ba734652aea474/N-verification.zip): all eight accepted whole-viewport PNGs, six crop PNGs, four-pair pixel verification, diff statistics, both frame-PTS JSON files, all 11 frame sheets covering 126/126 encoded frames, recording metadata, exact resolver/RPC NDJSON subsets, final span summaries, fixture transcript, and hashes.
- [Previously accepted F/E transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/db8d6ad0fc085944/F-E-before-after.txt). These controls were not rerun for this N-only task.

Unverified: the resolver's visible Top Span Names row remains below the production ten-row cap; exact raw records provide the previously accepted substitution. Native prefers-reduced-motion emulation is unavailable; capture-only CSS suppressed animations/transitions/caret/smooth scroll, but the media query stayed false. Optional controls, desktop/mobile surfaces, retained-history behavior and performance were outside this N recapture. No new audit verdict is claimed.

Cleanup verified: both tracked server terminals stopped with Ctrl-C and exited 130; own server/web ports have no listeners; both owned tabs closed; baseline worktree removed; after checkout clean at the requested head. Root disk has 108 GB available. No commits, pushes or forge writes. Uploads contain only synthetic data and evidence/scripts, without pairing credentials, private homes/logs or runtime model identifiers.

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

FRICTION:
- tool | Native reduced-motion media emulation is not exposed; CSS motion suppression was used and disclosed | Expose browser media emulation.
- navigation | After's first Diagnostics navigation raced General initialization and timed out; retry succeeded | Wait for General to finish mounting before route navigation.
- capture | Preliminary Changes still pairs had a changing refresh icon; baseline Top Span Names pairs were unsettled and excluded | Capture after refresh/scroll completion, and verify each pair.
- codec | Original baseline MP4 has lossy dark-background patches; clean PNGs remain the pixel sources | Expose higher-quality recording settings.