## Evidence
**Scope note for Julius:** G3/G4 also remove context expansion from wholly added/deleted guide excerpts. That visible behavior extends beyond the stale working-tree crash named in the PR title.
Pinned comparison: [083c903950][baseline] → [d3e3620310][head]. **8/8 affected rows captured; 12/12 matrix rows accounted for**, including four explicit waivers. D1/D2 use the real app, each built and run from its own checkout. P1/P2/G1–G4 are visibly labelled **HARNESS**, mounting production components and the real shared PR contents loader with synthetic data and controlled successful RPC output.
Rendering environment for every primary capture: fs-11, Ubuntu Linux x86_64, Playwright headless Chromium 148.0.7778.96, no device emulation, DPR 1, 100% zoom and font scale, 1440 × 1000 CSS pixels, light theme, reduced motion, Date frozen at 2026-11-03T12:00:00Z. The four D2 supplemental pairs use the labelled 1440/390 × 1000 viewport and light/dark theme, with the same remaining settings. The supplemental 1440-wide app view is split; 390 uses stacked view so labels remain legible. The baseline root error has no diff-view control. These are browser viewport widths, not mobile device emulation.
Every selected still passed network idle, fonts ready, and two pixel-identical screenshots 500 ms apart. Full viewports are preserved. Diff0 amplifies **any** nonzero RGB change to solid magenta. Diff16 marks max-channel differences **>16/255** in orange; unmarked pixels show a faded after image. Counts are in the linked manifest and sheets. All sheets and their constituent frames/diffs were visually reviewed. Recordings were reviewed through evenly spaced frames and the corresponding interaction stills.
What else changed:
- D2a: the baseline root error card is replaced by the usable Diff panel and cached patch. The fixed build then refreshes into the latest truncated diff and expands numbered prefix context. The baseline has no Refresh control in its root error state; the refresh-outcome pair explicitly compares that retained error screen with the fixed build’s refreshed context.
- D2b/D2d: the initial settled cached-patch frames are identical, but the baseline emits trailing-context renderer errors and reaches its error boundary on refresh. The fixed build rejects the stale contents, retains the cached hunk and separators, and expands after refresh.
- D2c: the baseline hydrates the edited line 31 as unchanged context even though it disagrees with the cached patch. The fixed build retains the displayed patch until refresh, then shows the edit in the latest diff.
- G2: the baseline expands an excerpt using a stale full source. The fixed build retains its compact patch; after source/cache refresh both expand again. The recovered pair has twelve rounded-card edge pixels differing by one RGB level, with zero pixels above threshold 16.
- G3: the fixed build removes expansion separators/controls and the extra expanded context, reduces the card height, and keeps contents requests at zero instead of one. The added line, counts, caption and Open in Code remain visible.
- G4: the fixed build removes expansion controls and the baseline’s inline null-line renderer error, leaving the compact removed-code excerpt. Contents requests remain zero instead of one.
- D1, P1, P2 and G1: all selected settled pairs have zero changed pixels at threshold zero. No visible changes are detected in those pairs.
Console results: the fixed primary D2 clip logs four expected “This file changed since the diff was loaded” rejections, one per stale case, and no trailing-context or null-line renderer error. G2 logs the same expected rejection before hydration. Both pins also log the pre-existing unresolved `pierre-dark` worker theme error; transcripts retain it. This evidence does not claim zero console errors. The normal Collapse all / Expand all remount avoids a separate metadata-identity issue present on both pins, where full contents could be read without visible expansion. The scripted remount and post-crash restoration of sidebar, panel and diff scope are disclosed in the replay guide.
Hosted authentication, guide generation, persistence and routing are **unverified** by the harness rows. Exact hosted fixture shapes were not provisioned; the permitted controlled-RPC fallback isolates the changed components without a provider turn or hosted mutation. No affected matrix row is omitted.
Reproduction: [replay.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/90552dde2ebfb471/replay.md) · [seed.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/82ca309f5960240d/seed.py) · [capture-app.mjs](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bb92812849efa643/capture-app.mjs) · [capture-harness.mjs](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/39baa7cb103ae183/capture-harness.mjs) · [harness-entry.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a6c4f0a443cebbde/harness-entry.js) · [harness-p1-entry.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/fc23c12be4947609/harness-p1-entry.js) · [harness.html](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/eb64928a60d4a62f/harness.html) · [all-driver-scripts.tar.gz](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/73a596a6d271936a/all-driver-scripts.tar.gz) · [runs.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/eeb89386e7342af0/runs.json) · [manifest.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/89dae85e0910b458/manifest.json).
### D2 — stale working-tree contents and branch-range repeat — REAL APP
Environment: 1440 × 1000, light, Linux Chromium, DPR 1, 100% zoom/font scale, no emulation; fixed clock and reduced motion as above.
Scripted interaction: Keep each cached line-30 patch unhydrated, then externally truncate to 30 lines, append line 101, or replace context line 31 without changing the count. Attempt expansion before focus/refresh. Refresh and expand again. Repeat append in Changes against main. Baseline crash recovery requires reload and explicit restoration of panel layout and scope; the fixed build stays in the panel.
Times below are seconds from the driver’s interaction start. The trimmed MP4 uses the recorded raw-video offset; transcript timestamps are recorded after each action/wait.
**Before steps:** 0.00s recording interaction start; 4.45s external file mutation; no focus or refresh (D2a-truncate); 5.89s first stale expansion attempted (D2a-truncate); 6.85s switch to split after stale load; 7.57s app error screen; reload required (D2a-truncate); 13.04s restored panel layout and diff scope after error reload (Uncommitted); 18.98s external file mutation; no focus or refresh (D2b-append); 20.43s first stale expansion attempted (D2b-append); 22.32s refresh latest diff (D2b-append); 22.33s refresh reached baseline error boundary; reload latest patch (D2b-append); 28.60s restored panel layout and diff scope after error reload (Uncommitted); 35.37s fresh expansion after refresh (D2b-append); 40.59s external file mutation; no focus or refresh (D2c-inplace); 42.03s first stale expansion attempted (D2c-inplace); 43.89s refresh latest diff (D2c-inplace); 49.65s fresh expansion after refresh (D2c-inplace); 55.99s external file mutation; no focus or refresh (D2d-branch); 57.44s first stale expansion attempted (D2d-branch); 59.31s refresh latest diff (D2d-branch); 59.31s refresh reached baseline error boundary; reload latest patch (D2d-branch); 65.67s restored panel layout and diff scope after error reload (Changes); 72.46s fresh expansion after refresh (D2d-branch); 73.26s complete.
**After steps:** 0.00s recording interaction start; 4.52s external file mutation; no focus or refresh (D2a-truncate); 5.98s first stale expansion attempted (D2a-truncate); 6.89s switch to split after stale load; 8.68s refresh latest diff (D2a-truncate); 15.06s fresh expansion after refresh (D2a-truncate); 20.43s external file mutation; no focus or refresh (D2b-append); 21.86s first stale expansion attempted (D2b-append); 23.65s refresh latest diff (D2b-append); 29.62s fresh expansion after refresh (D2b-append); 34.81s external file mutation; no focus or refresh (D2c-inplace); 36.24s first stale expansion attempted (D2c-inplace); 38.03s refresh latest diff (D2c-inplace); 43.84s fresh expansion after refresh (D2c-inplace); 50.19s external file mutation; no focus or refresh (D2d-branch); 51.65s first stale expansion attempted (D2d-branch); 53.43s refresh latest diff (D2d-branch); 59.38s fresh expansion after refresh (D2d-branch); 60.13s complete.
| Before recording | After recording |
| --- | --- |
| [D2-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/828de19131bdc309/D2-before.mp4)
[083c903950][baseline] | [D2-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a9fb74c3ad262f81/D2-after.mp4)
[d3e3620310][head] |
Raw recordings and transcripts: [D2-before-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a9870c9de1a5a21b/D2-before-raw.webm) · [D2-after-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/54e6a32e33fe3464/D2-after-raw.webm) · [D2-before-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a9be19cd6537839d/D2-before-transcript.json) · [D2-after-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0a3a2d5c760e069d/D2-after-transcript.json).
| Condition / viewport / theme | Before
[083c903950][baseline] | After
[d3e3620310][head] | Diff0
[083c903950][baseline] → [d3e3620310][head] | Diff16
[083c903950][baseline] → [d3e3620310][head] |
| --- | --- | --- | --- | --- |
| D2 / D2a-truncate-stale-1440-light / 1440×1000 / light / REAL APP |  |  |  |  |
| D2 / D2b-append-recovered / 1440×1000 / light / REAL APP |  |  |  |  |
| D2 / D2b-append-stale-1440-light / 1440×1000 / light / REAL APP |  |  |  |  |
| D2 / D2c-inplace-recovered / 1440×1000 / light / REAL APP |  |  |  |  |
| D2 / D2c-inplace-stale-1440-light / 1440×1000 / light / REAL APP |  |  |  |  |
| D2 / D2d-branch-recovered / 1440×1000 / light / REAL APP |  |  |  |  |
| D2 / D2d-branch-stale-1440-light / 1440×1000 / light / REAL APP |  |  |  |  |
| D2 / D2-end-1440-dark / 1440×1000 / dark / REAL APP |  |  |  |  |
| D2 / D2-end-1440-light / 1440×1000 / light / REAL APP |  |  |  |  |
| D2 / D2-end-390-dark / 390×1000 / dark / REAL APP |  |  |  |  |
| D2 / D2-end-390-light / 390×1000 / light / REAL APP |  |  |  |  |
| D2 / D2a refresh outcome (baseline retained root error; head refreshed) / 1440×1000 / light / REAL APP |  |  |  |  |
| D2 / D2b-append-refresh-outcome (baseline refresh error; head refreshed) / 1440×1000 / light / REAL APP |  |  |  |  |
| D2 / D2d-branch-refresh-outcome (baseline refresh error; head refreshed) / 1440×1000 / light / REAL APP |  |  |  |  |
Supplemental viewport/theme recording and transcripts: [D2-before-supplemental.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e2ffc9e62927c96a/D2-before-supplemental.webm) · [D2-before-supplemental-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d3be5dc9ab94717d/D2-before-supplemental-transcript.json) · [D2-after-supplemental.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/85305fd5c486aa61/D2-after-supplemental.webm) · [D2-after-supplemental-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/13a336dbda9467b5/D2-after-supplemental-transcript.json).
### D1 — matching full contents / whitespace / immutable commit — REAL APP
Environment: 1440 × 1000, light, Linux Chromium, DPR 1, 100% zoom/font scale, no emulation; fixed clock and reduced motion as above.
Scripted interaction: Expand prefix and tail; switch unified → split; toggle wrapping on the long line; show then hide whitespace, refresh, and expand the no-final-newline fixture; select the immutable commit, make an unrelated dirty edit, and expand it.
Times below are seconds from the driver’s interaction start. The trimmed MP4 uses the recorded raw-video offset; transcript timestamps are recorded after each action/wait.
**Before steps:** 0.00s recording interaction start; 4.45s D1 matching context expansion; 8.41s split diff view; 9.60s word wrap off on visible long line; 11.12s whitespace shown; 18.73s whitespace-insensitive expansion succeeded; 24.15s unrelated working-tree edit after immutable commit patch; 26.33s immutable commit expansion succeeded; 26.33s complete.
**After steps:** 0.00s recording interaction start; 4.45s D1 matching context expansion; 8.40s split diff view; 9.65s word wrap off on visible long line; 11.15s whitespace shown; 18.23s whitespace-insensitive expansion succeeded; 23.53s unrelated working-tree edit after immutable commit patch; 25.70s immutable commit expansion succeeded; 25.70s complete.
| Before recording | After recording |
| --- | --- |
| [D1-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c71a9475490da35d/D1-before.mp4)
[083c903950][baseline] | [D1-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/fa25cabd772e1234/D1-after.mp4)
[d3e3620310][head] |
Raw recordings and transcripts: [D1-before-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/cec9896d7184e842/D1-before-raw.webm) · [D1-after-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f66062307e148f14/D1-after-raw.webm) · [D1-before-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4eed4f94d6d6f47b/D1-before-transcript.json) · [D1-after-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a28c6103e399917e/D1-after-transcript.json).
| Condition / viewport / theme | Before
[083c903950][baseline] | After
[d3e3620310][head] | Diff0
[083c903950][baseline] → [d3e3620310][head] | Diff16
[083c903950][baseline] → [d3e3620310][head] |
| --- | --- | --- | --- | --- |
| D1 / D1-immutable-commit / 1440×1000 / light / REAL APP |  |  |  |  |
| D1 / D1-long-line-scroll / 1440×1000 / light / REAL APP |  |  |  |  |
| D1 / D1-matching-split / 1440×1000 / light / REAL APP |  |  |  |  |
| D1 / D1-matching-unified / 1440×1000 / light / REAL APP |  |  |  |  |
| D1 / D1-whitespace-no-final-newline / 1440×1000 / light / REAL APP |  |  |  |  |
### G1 — modified-source excerpts: new / old / addition-only / deletion-only — HARNESS
Environment: 1440 × 1000, light, Linux Chromium, DPR 1, 100% zoom/font scale, no emulation; fixed clock and reduced motion as above.
Scripted interaction: Expand before and after each excerpt. Visit new-side, old-side, addition-only and deletion-only variants in a modified source with another length-changing hunk. Use the same guide scroll position on both pins: top for line-30 excerpts, 1400 px (clamped by the container) for the line-81 addition so the changed line stays visible. Matching expanded context renders on both pins; no baseline renderer crash was observed in this particular matching fixture.
Times below are seconds from the driver’s interaction start. The trimmed MP4 uses the recorded raw-video offset; transcript timestamps are recorded after each action/wait.
**Before steps:** 0.00s start; 0.91s attempt context expansion (new); 3.52s expand context after excerpt (new); 5.10s select variant (old); 5.10s attempt context expansion (old); 7.69s expand context after excerpt (old); 9.27s select variant (addition); 9.27s attempt context expansion (addition); 11.88s expand context after excerpt (addition); 13.55s select variant (deletion); 13.55s attempt context expansion (deletion); 16.15s expand context after excerpt (deletion); 16.97s complete.
**After steps:** 0.00s start; 0.89s attempt context expansion (new); 3.53s expand context after excerpt (new); 5.20s select variant (old); 5.20s attempt context expansion (old); 7.81s expand context after excerpt (old); 9.46s select variant (addition); 9.47s attempt context expansion (addition); 12.16s expand context after excerpt (addition); 13.84s select variant (deletion); 13.84s attempt context expansion (deletion); 16.45s expand context after excerpt (deletion); 17.25s complete.
| Before recording | After recording |
| --- | --- |
| [G1-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f5defa7c8a38c8eb/G1-before.mp4)
[083c903950][baseline] | [G1-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/32050e32d9b0b33d/G1-after.mp4)
[d3e3620310][head] |
Raw recordings and transcripts: [G1-before-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ab9bdd15802a5aea/G1-before-raw.webm) · [G1-after-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0dc26f486a54488d/G1-after-raw.webm) · [G1-before-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b756b0df817784a2/G1-before-transcript.json) · [G1-after-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/823a69e39a3cf848/G1-after-transcript.json).
| Condition / viewport / theme | Before
[083c903950][baseline] | After
[d3e3620310][head] | Diff0
[083c903950][baseline] → [d3e3620310][head] | Diff16
[083c903950][baseline] → [d3e3620310][head] |
| --- | --- | --- | --- | --- |
| G1 / G1-addition-expanded / 1440×1000 / light / HARNESS |  |  |  |  |
| G1 / G1-deletion-expanded / 1440×1000 / light / HARNESS |  |  |  |  |
| G1 / G1-new-expanded / 1440×1000 / light / HARNESS |  |  |  |  |
| G1 / G1-old-expanded / 1440×1000 / light / HARNESS |  |  |  |  |
### G2 — stale source outside excerpt, then refresh — HARNESS
Environment: 1440 × 1000, light, Linux Chromium, DPR 1, 100% zoom/font scale, no emulation; fixed clock and reduced motion as above.
Scripted interaction: Expand with a successful RPC response whose later insertion differs from the cached source patch, although line 30 is unchanged. Refresh source metadata and loader cache key, then expand again. The real shared loader throws its own validation error on the fixed pin; no hand-thrown rejection is substituted.
Times below are seconds from the driver’s interaction start. The trimmed MP4 uses the recorded raw-video offset; transcript timestamps are recorded after each action/wait.
**Before steps:** 0.00s start; 1.22s attempt context expansion (stale source outside excerpt); 4.44s refresh source patch and loader cache; 4.45s attempt context expansion (matching refreshed source); 6.81s complete.
**After steps:** 0.00s start; 0.97s attempt context expansion (stale source outside excerpt); 4.33s refresh source patch and loader cache; 4.33s attempt context expansion (matching refreshed source); 6.75s complete.
| Before recording | After recording |
| --- | --- |
| [G2-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/338bd864441bc04f/G2-before.mp4)
[083c903950][baseline] | [G2-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c34e70950127e729/G2-after.mp4)
[d3e3620310][head] |
Raw recordings and transcripts: [G2-before-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/01f96cf452dde2b8/G2-before-raw.webm) · [G2-after-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e87473e0b7b654eb/G2-after-raw.webm) · [G2-before-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/cb2ffe469f096ba9/G2-before-transcript.json) · [G2-after-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9a983e32e6e491f2/G2-after-transcript.json).
| Condition / viewport / theme | Before
[083c903950][baseline] | After
[d3e3620310][head] | Diff0
[083c903950][baseline] → [d3e3620310][head] | Diff16
[083c903950][baseline] → [d3e3620310][head] |
| --- | --- | --- | --- | --- |
| G2 / G2-recovered / 1440×1000 / light / HARNESS |  |  |  |  |
| G2 / G2-stale / 1440×1000 / light / HARNESS |  |  |  |  |
### G3 — wholly added source excerpt — HARNESS
Environment: 1440 × 1000, light, Linux Chromium, DPR 1, 100% zoom/font scale, no emulation; fixed clock and reduced motion as above.
Scripted interaction: Use genuine new-file metadata and /dev/null old path. Attempt expansion if offered, or scan/click the excerpt when no control exists. Then invoke Open in Code. The callback is logged; real route navigation is outside the harness.
Times below are seconds from the driver’s interaction start. The trimmed MP4 uses the recorded raw-video offset; transcript timestamps are recorded after each action/wait.
**Before steps:** 0.00s start; 1.22s attempt context expansion (wholly added/deleted source); 3.84s Open in Code remains available; 3.84s complete.
**After steps:** 0.00s start; 0.92s attempt context expansion (wholly added/deleted source); 3.07s Open in Code remains available; 3.07s complete.
| Before recording | After recording |
| --- | --- |
| [G3-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e8ee94c2df133e2d/G3-before.mp4)
[083c903950][baseline] | [G3-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c0438f2331863a75/G3-after.mp4)
[d3e3620310][head] |
Raw recordings and transcripts: [G3-before-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/1548c895a9f6813b/G3-before-raw.webm) · [G3-after-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b6015836d20e393b/G3-after-raw.webm) · [G3-before-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e0c63c3c398cf1af/G3-before-transcript.json) · [G3-after-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0efd2f2dce146336/G3-after-transcript.json).
| Condition / viewport / theme | Before
[083c903950][baseline] | After
[d3e3620310][head] | Diff0
[083c903950][baseline] → [d3e3620310][head] | Diff16
[083c903950][baseline] → [d3e3620310][head] |
| --- | --- | --- | --- | --- |
| G3 / G3-attempted / 1440×1000 / light / HARNESS |  |  |  |  |
| G3 / G3-initial / 1440×1000 / light / HARNESS |  |  |  |  |
### G4 — wholly deleted source excerpt — HARNESS
Environment: 1440 × 1000, light, Linux Chromium, DPR 1, 100% zoom/font scale, no emulation; fixed clock and reduced motion as above.
Scripted interaction: Use genuine deleted-file metadata and /dev/null new path. Attempt expansion if offered, or scan/click the excerpt when no control exists. Then invoke Open in Code. The callback is logged; real route navigation is outside the harness.
Times below are seconds from the driver’s interaction start. The trimmed MP4 uses the recorded raw-video offset; transcript timestamps are recorded after each action/wait.
**Before steps:** 0.00s start; 1.09s attempt context expansion (wholly added/deleted source); 3.82s Open in Code remains available; 3.82s complete.
**After steps:** 0.00s start; 0.87s attempt context expansion (wholly added/deleted source); 2.58s Open in Code remains available; 2.58s complete.
| Before recording | After recording |
| --- | --- |
| [G4-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/854af40caa475d5a/G4-before.mp4)
[083c903950][baseline] | [G4-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8efe04984f639893/G4-after.mp4)
[d3e3620310][head] |
Raw recordings and transcripts: [G4-before-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a5410f793daf1357/G4-before-raw.webm) · [G4-after-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2dbd2c2fad0c156b/G4-after-raw.webm) · [G4-before-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a29b307c1f8959ab/G4-before-transcript.json) · [G4-after-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/22eab63e58282e40/G4-after-transcript.json).
| Condition / viewport / theme | Before
[083c903950][baseline] | After
[d3e3620310][head] | Diff0
[083c903950][baseline] → [d3e3620310][head] | Diff16
[083c903950][baseline] → [d3e3620310][head] |
| --- | --- | --- | --- | --- |
| G4 / G4-attempted / 1440×1000 / light / HARNESS |  |  |  |  |
| G4 / G4-initial / 1440×1000 / light / HARNESS |  |  |  |  |
### P1 — PR Code view, whole-PR commit:null — HARNESS
Environment: 1440 × 1000, light, Linux Chromium, DPR 1, 100% zoom/font scale, no emulation; fixed clock and reduced motion as above.
Scripted interaction: Open the real PullRequestCodeTab file, remount through Collapse all / Expand all, expand context using the actual PR loader, then switch to split view. The query and RPC hook outputs are controlled; the renderer and shared contents validator are real.
Times below are seconds from the driver’s interaction start. The trimmed MP4 uses the recorded raw-video offset; transcript timestamps are recorded after each action/wait.
**Before steps:** 0.00s start; 3.55s attempt context expansion (whole PR, commit null); 6.91s switch split/unified; 7.79s complete.
**After steps:** 0.00s start; 2.63s attempt context expansion (whole PR, commit null); 5.80s switch split/unified; 6.70s complete.
| Before recording | After recording |
| --- | --- |
| [P1-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f326e69c20700fc6/P1-before.mp4)
[083c903950][baseline] | [P1-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6c667d3a8dbe2729/P1-after.mp4)
[d3e3620310][head] |
Raw recordings and transcripts: [P1-before-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/1b2e50f061df01fb/P1-before-raw.webm) · [P1-after-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e425fb2ab77203f6/P1-after-raw.webm) · [P1-before-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/95220641310d366b/P1-before-transcript.json) · [P1-after-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a23e0faa7548e0f4/P1-after-transcript.json).
| Condition / viewport / theme | Before
[083c903950][baseline] | After
[d3e3620310][head] | Diff0
[083c903950][baseline] → [d3e3620310][head] | Diff16
[083c903950][baseline] → [d3e3620310][head] |
| --- | --- | --- | --- | --- |
| P1 / P1-expanded / 1440×1000 / light / HARNESS |  |  |  |  |
| P1 / P1-split / 1440×1000 / light / HARNESS |  |  |  |  |
### P2 — GuideStop “Also in this stop” full file — HARNESS
Environment: 1440 × 1000, light, Linux Chromium, DPR 1, 100% zoom/font scale, no emulation; fixed clock and reduced motion as above.
Scripted interaction: Open other.ts under Also in this stop, expand that full file, then expand the excerpt. GuideStop, GuideFileDiff and the shared loader are real. Both remain usable.
Times below are seconds from the driver’s interaction start. The trimmed MP4 uses the recorded raw-video offset; transcript timestamps are recorded after each action/wait.
**Before steps:** 0.00s start; 1.00s open Also in this stop other.ts; 5.06s excerpt expanded after other file; 5.92s complete.
**After steps:** 0.00s start; 1.57s open Also in this stop other.ts; 5.75s excerpt expanded after other file; 6.67s complete.
| Before recording | After recording |
| --- | --- |
| [P2-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a686a4a4ca38ecbb/P2-before.mp4)
[083c903950][baseline] | [P2-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2a650d08495ef3e5/P2-after.mp4)
[d3e3620310][head] |
Raw recordings and transcripts: [P2-before-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/744244e12821134b/P2-before-raw.webm) · [P2-after-raw.webm](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8a4b0055560c0ed8/P2-after-raw.webm) · [P2-before-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f96c27b67d1f2f52/P2-before-transcript.json) · [P2-after-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/fcdaa8d5d42564f8/P2-after-transcript.json).
| Condition / viewport / theme | Before
[083c903950][baseline] | After
[d3e3620310][head] | Diff0
[083c903950][baseline] → [d3e3620310][head] | Diff16
[083c903950][baseline] → [d3e3620310][head] |
| --- | --- | --- | --- | --- |
| P2 / P2-both-expanded / 1440×1000 / light / HARNESS |  |  |  |  |
| P2 / P2-other-expanded / 1440×1000 / light / HARNESS |  |  |  |  |
### Coverage and limits
| Matrix row | Status |
| --- | --- |
| D1 | Captured — real app |
| D2 | Captured — real app |
| P1 | Captured — labelled harness |
| P2 | Captured — labelled harness |
| G1 | Captured — labelled harness |
| G2 | Captured — labelled harness |
| G3 | Captured — labelled harness |
| G4 | Captured — labelled harness |
| U1 | Waived by supplied matrix — unchanged multi-hunk excerpt guard |
| U2 | Waived by supplied matrix — unchanged loading branch |
| U3 | Waived by supplied matrix — unchanged missing-range fallback |
| U4 | Waived by supplied matrix — unchanged annotations/commenting |
Unverified affected rows: **none** under the permitted harness fallback. Hosted end-to-end prerequisites remain unverified as disclosed above. Native mobile, Electron shell, relay and tunnel were not independently exercised; the supplied matrix scopes this change to the web loaders and renderer.
All data is synthetic. No source changes, commits, pushes, PR-body edits, hosted comments or provider turns were made. The temporary baseline checkout and all servers/browser processes started for capture are removed/stopped before handoff. The lead’s head checkout remains in place.
All exploratory seed/driver scripts are uploaded individually and included in the archive. Private auth files and startup logs are excluded.
[baseline]: https://github.com/pingdotgg/t3code/commit/083c9039505fc9207b88343e19d0a3de4a5e622d
[head]: https://github.com/pingdotgg/t3code/commit/d3e3620310e8489e1cee5776e79d64ad7173bc63
Evidence captured with gpt-6.1-sol through the Codex harness.