# PASS2 proof audit — issue24504

Read-only audit of worktree `/home/ubuntu/wt/issue-24504`, branch `fix/issue-24504`, exact HEAD `203a30bacc0dbb269fa68b66f12b9c5e7cb6d0fc`, base `b3d382e1009d5430de8c528aef99a39854a7cbc4`. No implementation edits or forge writes. No confirmed implementation bug; the verdict concerns incomplete proof against the supplied matrix.

Public index and bundle byte-match the local delivery. Inspected all 26 root images (including supplied contact sheets), all 36 delivered frame images, all nine recordings across their full durations using one-second extraction plus transition/tail frames, and all eight delivered Diff0/threshold20 images. Inspected key originals separately. Prior main captures are explicitly prior baseline, not stale head proof. Current worktree and independent reproduction establish the requested head; original capture SHA provenance remains the supplied verification transcript.

## Row results

| Row | Status | Observed proof and exact remaining scope |
|---|---|---|
| A | INSUFFICIENT | Primary opens and closes; recording ends in composer target submenu, before the alternate dialog opens. It does not record changing selections/query followed by close/reopen/reset. secondary-runtime.json reports defaults but cannot replace the required interaction recording. |
| B | SUFFICIENT | Matched base/head large-fixture captures at both requested CSS widths and both themes. Defaults 1/15B, first200, disabled oversize, truncation/search hint, footer all present. Independently reproduced wide/light head. |
| C | INSUFFICIENT, narrow gap | Trimmed uppercase205 past cutoff, Enter without clone request, rapid final204 and clear are supported. Whitespace-only returning the initial list is not demonstrated in delivered interaction evidence. |
| D | SUFFICIENT | Recording shows .env+205+204, retained205 on revisit, count3/31B and filtered clone. Destination transcript exactly those three files with expected contents and unselected001 absent. Clear/hidden-retention behavior additionally supported by F; destination was not independently re-cloned during this audit. |
| E | INSUFFICIENT | Shown-only operations preserve hidden205; summary variants have runtime/visual support. Partially selected multirow results and None specifically at cap are missing, as disclosed. Zero-byte variant is supported; optionality does not excuse the other missing cases. |
| F | SUFFICIENT | Recording demonstrates global200, disabled new eligible row, All cannot exceed cap, selected row unchecks to199, notice disappears, beyond-cutoff204 restores200, filtered clone. Destination transcript validates all200 expected contents/1607B, excludes001/oversize, includes.env/204/205. |
| G | INSUFFICIENT | Empty/small static branches pass. The cited search-enter-debounce-zero recording has205→204→env→zero selection, not nonsense/no-match→clear. No delivered recording shows the required no-match message with retained hidden picks and recovery. Service no-match transcript is a different proof obligation. |
| H | INSUFFICIENT | Clearly labeled synthetic writable-query-atom pending/error/recovery and initial error are visible, including the short final initial-error frame. They do not exercise controlled RPC/service delay/failure required by the matrix. No claim of a real transport failure is justified. |
| I | INSUFFICIENT, narrow gap | Clearly labeled SOURCE capability overrides, modern target, no-search/no-listing UI and ordinary successful clones supported. Synthetic override is an allowed matrix fallback; absence of an actual old binary is not itself a finding. Old-listing All reaches199 because oversize is in first200; exact200 compatibility cap remains unexercised. |
| J | SUFFICIENT | Independently reran WorkspaceEntries.test.ts:19/19 passed. Contract transcript accepts omitted/empty/256 and rejects257. Authenticated service transcripts cover omitted/empty, trim/casefold, beyond cutoff, broad truncation, no-match, wholly ignored directory exclusion, size eligibility. Exact1MiB destination copy was not required by this row. |

## Confirmed findings

P2 evidence completeness: A, C, E, G, H, I have the exact gaps above. In particular the index's A/G Passed claims exceed the recordings. These are missing proof, not established application defects. H/E/I limitations are candidly disclosed; they still leave matrix obligations open.

P3 metric precision: delivered `differentPixelsThreshold0` counts PIL luminance>0, not every changed RGB pixel. All delivered Diff0 and threshold20 images recalculate pixel-exactly. Correct any-RGB counts and delivered luminance counts are below; this does not invalidate layout proof.

| Comparison | Any RGB change | Luminance >0 | Luminance >20 |
|---|---:|---:|---:|
| 1440 light, raster1280×800 |293196|290253|64564|
| 1440 dark, raster1280×800 |253495|251037|43061|
| 390 light, raster780×1688 |271265|269641|78771|
| 390 dark, raster780×1688 |297774|297770|76591|

Visible changes: new search field, global selected-count wording, expanded truncation hint. Wide dialog grows30 raster pixels (top15 up, bottom15 down). Narrow sheet expands upward76 raster pixels while list/footer positions remain stable, accommodating search. Background relative timestamps, backdrop/shadow geometry, and caret also contribute differences. Narrow search/cap variants in both themes remain legible; no horizontal overflow or inaccessible footer observed. No unaccounted visible implementation regression established. Responsive web is not native mobile. Desktop shell/hosted/relay/tunnel remain unverified conditional surfaces, not failed mandatory matrix rows.

## Independent public-seed reproduction

Ran the public seed/driver in the requested head on fs-15, Linux x86_64, Node26.12.1, with separate isolated source/target homes and actual Settings pairing. Source Vite7797/backend15837, target15838 with matching dev origin; local synthetic git remote19454. Opened the real panel caller in light theme at CSS1440×900/DPR2; transferred raster1280×800 matches the delivery. Confirmed200 checkboxes, one selected default.env/15B, search/truncation and oversize states.

Reproduction versus delivered head-initial-1440-light.png: whole image126110/1024000 pixels differ in any RGB channel (12.315%);9414 exceed luminance20 (0.919%). Within dialog crop [384,96,896,704],898/311296 differ in RGB (0.2885%);543 exceed luminance0;16 exceed luminance20 (0.00514%). Differences concentrate in background/sidebar and, inside dialog, caret/minor edge pixels. This is close reproduction, not a zero-diff claim. Capture, Diff0, threshold20 mask, and exact metrics accompany this report.

## Cleanup

Owned browser tab closed; preview_status reports no attached tab. Owned source/target/git-daemon stopped; ports7797/15837/15838/19454 verified closed. Worktree clean at requested HEAD. Audit temporary homes, fixture repositories and image extraction scratch removed. Only audit artifacts outside the repository retained. No implementation files changed, push, forge write, merge, watcher, or delegation.

## FRICTION category|happened|avoidance

- browser selectors|Unsupported exact attribute and assumed placeholder/button labels caused timeouts.|Inspect settled accessibility/DOM state and use actual labels before interaction.
- infrastructure|Owned source runner exited143 during audit; cause not established.|Check tracked session and own ports, restart only isolated owned runner, reload and revalidate state.
- inspection|One-second sampling initially missed the very short final initial-error state.|Inspect recording tails and transitions before declaring a missing capture.
- coverage|A/G index claims exceeded recorded outcomes.|Map every mandatory subcase to a visible timestamp and preserve settled outcome before stopping recording.

VERDICT: INSUFFICIENT A C E G H I