# Independent web proof audit, pass 2

Retained visual evidence: **SUFFICIENT for the requested web header change**. No missing matrix rows require recapture. Overall execution: **FRICTION, synthetic-only discovery requirement violated**. This verdict does not certify clean isolation, authorize merging, or erase the two reported native usage-summary requests.

Before [eb1510d249](https://git.cafe/pingdotgg/t3code/commit/eb1510d2493cbf28600432d5193cc22a1af8f3d8), After [37a7baecae](https://git.cafe/pingdotgg/t3code/commit/37a7baecae979f556b2218ba496a2ff2f6d725d1). The worktree stayed clean at After. Independent git diff confirms only `apps/web/src/components/usage/UsagePage.tsx`, 7 additions and 5 deletions. Dependency lock hashes match between revisions. No application changes, tests, commits, pushes, forge writes, ledger updates, other machines, or other worktrees were used.

## Scope and method

Read matrix.md, seeds.md, capture.md, report.md and public-manifest.json. Opened all 26 retained capture inspection sheets: each of the 196 primary Before/After images appears as a full-surface thumbnail with its native-size header. Opened all nine exact/threshold diff inspection sheets, covering every pair and both full/header diffs. Independently decoded and recomputed all 98 pairs, all stability partners, header crops and both diff algorithms; no cached metrics were trusted. Opened the four full-size compact-menu checkpoint images to inspect every option without relying on the clipped header crop.

Opened all eight actual MP4 recordings through 16 independently extracted contact sheets, including a video frame after every timestamped action (164 action states total). These are MP4-decoded frames, not the checkpoint PNGs. Inspected compact menu, metric/period changes, refresh settled states, disabled period, both sidebar crossings, old viewport edges and final state in every revision/theme recording. This is action-state visual inspection of every recording, not a claim to have watched every 30-fps frame. Audio/video streams, full final-state duration, narration source/timing, and non-silent decoded audio are checked separately; narration is post-produced TTS, not live human commentary.

The user explicitly requested the existing Xvfb/ffmpeg workflow. Rebuilt through its synthetic RPC adapter in headful Chromium on the retained Xvfb display. The audit copy rejects unexpected RPC method names; only explicit non-provider UI subscriptions/assets/preview reporting pass through. Usage summary, provider refresh, usage rates and config are handled synthetically. External HTTP requests are blocked. No native usage/provider files or quarantined native data were inspected. Existing isolated-session authentication is consumed only by the original browser harness, never printed or uploaded.

## Independent results

- 16/16 static pairs and 82/82 interaction checkpoint pairs present, with exact Before/After SHAs, matching viewport/sidebar/preferences/DPR/zoom/root-rem/crop geometry.
- All 196 second frames are pixel-identical to their primary frame, with manifest elapsed stability intervals at least 500 ms.
- All 392 full/header exact and threshold diff images independently match the algorithms. Threshold is maximum RGBA channel delta >16; exact display is absolute RGB delta multiplied by 8, clamped to 255. No image alignment or masks.
- Zero changed pixels outside the header rectangle in all 98 comparisons. Body/sidebar preservation was also visually inspected in the full-surface thumbnails.
- All 151 public files match public-manifest.json size and raw-byte SHA256. All 15 ZIPs have no private/runtime/auth/native-cache entry paths. Compared 1,502 current capture archive entries: all image/diff data match. The only 32 differences are static manifests' `originalHarnessHashes`, discussed below.
- Verified 69 usage replies against the deterministic master generator and contract, across 41 unique retained-session logs including rebuild; 41 config snapshots equal the synthetic config. Request IDs match in the final retained sessions. No fail-closed rejection occurred in the exact rebuild. Historical failed harness attempts remain in append-only logs; validation starts at the final subscribeServerConfig session boundary, as the capture report's corrected-attempt disclosure requires.

## Visual findings by required fixture

Both themes show the same intended behavior. H1 remains compact. H2 changes compact to full at C897 despite W1193. H3 changes full to compact at C895 despite W1415. H4 remains full, with filter truncation preserving controls. D1/D3 hide the old viewport-triggered date at C1279; D2/D4 retain their date at C1281. Limits never has a date and its period remains disabled. No After controls overlap, clip, disappear, or create horizontal page overflow in the inspected states.

I1 shows all compact metric options, Tokens and Limits, blocked pointer/period shortcut, limits refresh, return to Cost, 7/30/90/1 periods, usage refresh, T shortcut and repeated fixed-W sidebar crossings. Tokens/1 persists across the After controls-tree changes. Before stays compact at that viewport, correctly exposing the old viewport dependence.

I2 shows full controls, Limits/no date/disabled period, return Cost/date restored, all periods, T/C shortcuts, fixed-W date crossings and compact/full crossings, old W1279/1281/1535/1537 edges, and restored final state. Before visibly overlaps date/breadcrumb and controls at narrow header widths; After hides the date or switches controls based on available header width. Disabled period remains 1 in I1 and 7 in I2. The enabled day shortcut is pressed when day is already selected: this does not independently demonstrate a period transition by keyboard. The period buttons/popups prove all transitions, and keyboard dispatch logic is unchanged by this CSS-only change; this is a bounded weakness, not a missing changed-render-path row.

All rows below were visually inspected. Exact/threshold counts are identical for full/header scope because all changed pixels lie inside the header. Stable=0 changed pixels for both revisions; outside=0 for every row.

| Row/theme | W / sidebar / content width | Metric / days | Before → After controls | Date before → after | Exact / >16 pixels | Stable / outside |
|---|---|---|---|---|---|---|
| D1/dark | 1575 / 256 / 1279 | cost / 30 | full → full | True → False | 571 / 450 | 0 / 0 |
| D1/light | 1575 / 256 / 1279 | cost / 30 | full → full | True → False | 571 / 477 | 0 / 0 |
| D2/dark | 1577 / 256 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| D2/light | 1577 / 256 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| D3/dark | 1799 / 480 / 1279 | tokens / 90 | full → full | True → False | 538 / 445 | 0 / 0 |
| D3/light | 1799 / 480 / 1279 | tokens / 90 | full → full | True → False | 538 / 461 | 0 / 0 |
| D4/dark | 1801 / 480 / 1281 | tokens / 7 | full → full | True → True | 0 / 0 | 0 / 0 |
| D4/light | 1801 / 480 / 1281 | tokens / 7 | full → full | True → True | 0 / 0 | 0 / 0 |
| H1/dark | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| H1/light | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| H2/dark | 1193 / 256 / 897 | tokens / 7 | compact → full | False → False | 15307 / 5828 | 0 / 0 |
| H2/light | 1193 / 256 / 897 | tokens / 7 | compact → full | False → False | 12779 / 3876 | 0 / 0 |
| H3/dark | 1415 / 480 / 895 | limits / 30 | full → compact | False → False | 15238 / 4353 | 0 / 0 |
| H3/light | 1415 / 480 / 895 | limits / 30 | full → compact | False → False | 14071 / 3854 | 0 / 0 |
| H4/dark | 1417 / 480 / 897 | limits / 30 | full → full | False → False | 711 / 694 | 0 / 0 |
| H4/light | 1417 / 480 / 897 | limits / 30 | full → full | False → False | 707 / 691 | 0 / 0 |
| I1-00/dark | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-00/light | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-01/dark | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-01/light | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-02/dark | 1191 / 256 / 895 | tokens / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-02/light | 1191 / 256 / 895 | tokens / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-03/dark | 1191 / 256 / 895 | limits / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-03/light | 1191 / 256 / 895 | limits / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-04/dark | 1191 / 256 / 895 | limits / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-04/light | 1191 / 256 / 895 | limits / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-05/dark | 1191 / 256 / 895 | limits / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-05/light | 1191 / 256 / 895 | limits / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-06/dark | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-06/light | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-07/dark | 1191 / 256 / 895 | cost / 7 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-07/light | 1191 / 256 / 895 | cost / 7 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-08/dark | 1191 / 256 / 895 | cost / 30 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-08/light | 1191 / 256 / 895 | cost / 30 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-09/dark | 1191 / 256 / 895 | cost / 90 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-09/light | 1191 / 256 / 895 | cost / 90 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-10/dark | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-10/light | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-11/dark | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-11/light | 1191 / 256 / 895 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-12/dark | 1191 / 256 / 895 | tokens / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-12/light | 1191 / 256 / 895 | tokens / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-13/dark | 1191 / 248 / 903 | tokens / 1 | compact → full | False → False | 15166 / 6017 | 0 / 0 |
| I1-13/light | 1191 / 248 / 903 | tokens / 1 | compact → full | False → False | 12440 / 3852 | 0 / 0 |
| I1-14/dark | 1191 / 256 / 895 | tokens / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-14/light | 1191 / 256 / 895 | tokens / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I1-15/dark | 1191 / 248 / 903 | tokens / 1 | compact → full | False → False | 15166 / 6017 | 0 / 0 |
| I1-15/light | 1191 / 248 / 903 | tokens / 1 | compact → full | False → False | 12440 / 3852 | 0 / 0 |
| I1-16/dark | 1191 / 248 / 903 | tokens / 1 | compact → full | False → False | 15166 / 6017 | 0 / 0 |
| I1-16/light | 1191 / 248 / 903 | tokens / 1 | compact → full | False → False | 12440 / 3852 | 0 / 0 |
| I2-00/dark | 1801 / 480 / 1281 | tokens / 7 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-00/light | 1801 / 480 / 1281 | tokens / 7 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-01/dark | 1801 / 480 / 1281 | limits / 7 | full → full | False → False | 0 / 0 | 0 / 0 |
| I2-01/light | 1801 / 480 / 1281 | limits / 7 | full → full | False → False | 0 / 0 | 0 / 0 |
| I2-02/dark | 1801 / 480 / 1281 | limits / 7 | full → full | False → False | 0 / 0 | 0 / 0 |
| I2-02/light | 1801 / 480 / 1281 | limits / 7 | full → full | False → False | 0 / 0 | 0 / 0 |
| I2-03/dark | 1801 / 480 / 1281 | cost / 7 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-03/light | 1801 / 480 / 1281 | cost / 7 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-04/dark | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 5 / 0 | 0 / 0 |
| I2-04/light | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 5 / 0 | 0 / 0 |
| I2-05/dark | 1801 / 480 / 1281 | cost / 7 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-05/light | 1801 / 480 / 1281 | cost / 7 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-06/dark | 1801 / 480 / 1281 | cost / 30 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-06/light | 1801 / 480 / 1281 | cost / 30 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-07/dark | 1801 / 480 / 1281 | cost / 90 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-07/light | 1801 / 480 / 1281 | cost / 90 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-08/dark | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-08/light | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-09/dark | 1801 / 480 / 1281 | tokens / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-09/light | 1801 / 480 / 1281 | tokens / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-10/dark | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-10/light | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-11/dark | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-11/light | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-12/dark | 1801 / 484 / 1277 | cost / 1 | full → full | True → False | 995 / 815 | 0 / 0 |
| I2-12/light | 1801 / 484 / 1277 | cost / 1 | full → full | True → False | 995 / 858 | 0 / 0 |
| I2-13/dark | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-13/light | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-14/dark | 1801 / 864 / 897 | cost / 1 | full → full | True → False | 1720 / 1489 | 0 / 0 |
| I2-14/light | 1801 / 864 / 897 | cost / 1 | full → full | True → False | 1361 / 1229 | 0 / 0 |
| I2-15/dark | 1801 / 868 / 893 | cost / 1 | full → compact | True → False | 15717 / 6244 | 0 / 0 |
| I2-15/light | 1801 / 868 / 893 | cost / 1 | full → compact | True → False | 13227 / 4235 | 0 / 0 |
| I2-16/dark | 1801 / 864 / 897 | cost / 1 | full → full | True → False | 1720 / 1489 | 0 / 0 |
| I2-16/light | 1801 / 864 / 897 | cost / 1 | full → full | True → False | 1361 / 1229 | 0 / 0 |
| I2-17/dark | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-17/light | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-18/dark | 1279 / 480 / 759 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I2-18/light | 1279 / 480 / 759 | cost / 1 | compact → compact | False → False | 0 / 0 | 0 / 0 |
| I2-19/dark | 1281 / 480 / 761 | cost / 1 | full → compact | False → False | 14656 / 5382 | 0 / 0 |
| I2-19/light | 1281 / 480 / 761 | cost / 1 | full → compact | False → False | 12306 / 3473 | 0 / 0 |
| I2-20/dark | 1535 / 480 / 1015 | cost / 1 | full → full | False → False | 0 / 0 | 0 / 0 |
| I2-20/light | 1535 / 480 / 1015 | cost / 1 | full → full | False → False | 0 / 0 | 0 / 0 |
| I2-21/dark | 1537 / 480 / 1017 | cost / 1 | full → full | True → False | 1974 / 1714 | 0 / 0 |
| I2-21/light | 1537 / 480 / 1017 | cost / 1 | full → full | True → False | 1867 / 1653 | 0 / 0 |
| I2-22/dark | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-22/light | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-23/dark | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |
| I2-23/light | 1801 / 480 / 1281 | cost / 1 | full → full | True → True | 0 / 0 | 0 / 0 |

## Recordings

Each row's two contact sheets were opened and visually inspected. Narration describes actual measured geometry per revision, including Before staying compact at C903 and retaining the date at C897. Labels in the action transcript describe intended actions; `narration` and the WAV source carry the corrected observations.

| Recording | Action states inspected | Duration (seconds) | Video / audio |
|---|---|---|---|
| I1-dark-after | 17 | 102.066000 | H.264 / AAC |
| I1-dark-before | 17 | 101.933000 | H.264 / AAC |
| I1-light-after | 17 | 101.300000 | H.264 / AAC |
| I1-light-before | 17 | 102.066000 | H.264 / AAC |
| I2-dark-after | 24 | 144.066000 | H.264 / AAC |
| I2-dark-before | 24 | 144.000000 | H.264 / AAC |
| I2-light-after | 24 | 143.933000 | H.264 / AAC |
| I2-light-before | 24 | 144.000000 | H.264 / AAC |

## Independent capture rebuild

Rebuilt H2/light/After with fresh browser context, W1193×1000, S256, C897, Tokens/7, UTC/en-US/24-hour, frozen time, DPR1, zoom1, root rem16, same browser/Xvfb/fonts and deterministic synthetic replies. No source checkout or server restart was necessary.

First attempt opened directly at final W1193. Its stable image differed by 5,442 exact pixels (5,438 >16, maximum channel delta35), exclusively in the existing body scrollbar at x1186–1191, y56–962. The header was already pixel-identical. This was a setup-history mismatch: the original After static harness opens at C+S+48=1201 and then resizes to measured W1193. Replaying that exact setup produced **0 exact changed pixels, 0 >16 pixels, and 0 stability changed pixels over the whole surface**. Both attempts and their scripts/diffs are preserved; no masking or scrollbar CSS was used.

Exact rebuild: audit-pass2/rebuild-exact/after.png. Script: audit-pass2/rebuild-exact.mjs. Recompute script: audit-pass2/audit.py. Supplemental checks: provenance-check.mjs, check-audio.py, checks.json, provenance.json, extra-checks.json, audio-checks.json and archive-equality.json. Raw-byte script identity is in rebuild-exact/audit-provenance.json; inherited digest fields retain the old harness convention and are explicitly distinguished.

## Isolation violation and provenance limits

The capture report and public probe-boundary-audit metadata disclose two discovery requests that reached native usage scanning before the adapter. I accept that disclosure as a real violation. I did not inspect the quarantined native responses/caches, so I do not independently attest to their contents or deletion. Directory exclusion checks establish upload packaging boundaries, not a retrospective clean-room guarantee.

Retained evidence remains sufficient for this narrow visual change because the final sessions' visible summaries/config/limits are deterministic synthetic fixtures, both revisions get the same request-bounded summary data, every comparison is stable, source scope is restricted to header responsiveness, and an independent fail-closed exact rebuild reproduces the entire H2 image byte-for-pixel. This supports the retained web render comparison despite the discovery violation. It cannot support a claim that the whole execution was synthetic-only, that native providers were never touched, or that provider ingestion was tested.

A separate metadata issue: 32 current static manifests' `originalHarnessHashes` were overwritten with canonical values by a subsequent provenance-script run. Uploaded static archives preserve the earlier legacy values. All other fields, including canonical adapter/master hashes, SHAs, geometry, data and images match. The underlying `.original.json` records and archived scripts retain the provenance chain. This needs disclosure, not another screenshot run; the audit bundle preserves the exact difference inventory.

No required visual rows are missing. No additional recapture is required for retained web visual sufficiency. If clean execution isolation is independently required, no number of extra screenshots repairs this historical run: start fresh isolated homes, block every native/provider usage path before any navigation, install the fail-closed synthetic adapter before creating the first page, and rerun H1–H4/D1–D4 in both themes plus I1/I2 on both revisions/themes. Do not reuse native caches or inspect the quarantine. That would be a new execution-compliance proof, not a correction to these images.

Not covered: desktop/mobile, remote/relay/tunnel, provider ingestion, loading/error/permissions/empty-selection supplemental branches. The specification explicitly excluded those supplemental states absent observed risk. No new risk requiring them appeared. This is no broader source or merge approval.

Disk checks remained about 14 GB available, above 5 GB. Retained dev-runner PID2250261 and Xvfb PID2216157 were reused and left running. Browser contexts launched by this audit closed normally. No stalled shell or unrelated process termination.

Model/harness: gpt-6.1-sol / Codex in T3 Code.

VERDICT: SUFFICIENT