1. **W02 is insufficient: all three pairs contain a spinner.** The six files `W02-W-light-three-before.png`, `W02-W-light-three-after.png`, `W02-W-dark-three-before.png`, `W02-W-dark-three-after.png`, `W02-N-dark-three-before.png`, and `W02-N-dark-three-after.png` visibly say “Syncing messages…” below the final message. Recapture after reconciliation. The Evidence claim that W02 is complete is contradicted. A clean diff does not make these valid endpoints.
2. **Coverage remains incomplete.** Seven of the eight delivered row IDs have missing cells/subcases or invalid endpoints. All remaining matrix rows lack accepted observation. Missing details are listed per row below. Explicitly rejected M01/H01 diagnostics were not promoted to passing evidence. No delivery from the parallel capturer was supplied to this audit.
3. **The unbadged overlap reversal matches the stated intent.** A01 shows earlier avatars on top on both web and Android. Both changed implementations apply the order unconditionally. This agrees with the PR sentence “Grouped headers on web and mobile stack earlier avatars on top so each bottom-right badge stays visible.” It is a visible consequence, not evidence of extra scope relative to that sentence.

Inspected the five-file diff at [fd6f84adeb](https://github.com/pingdotgg/t3code/commit/fd6f84adebd5592fff1df47f903ad260c389f2bd) against [32ecd71726](https://github.com/pingdotgg/t3code/commit/32ecd71726fee16f335484c1f95cbce126d461f6), all 38 before/after pairs, their 76 diff panels and 38 sheets, and all 24 recordings through every decoded source frame (603 frames), using both full-scene and enlarged avatar review sheets. The delivered column labels match the requested full SHAs. All 38 pairs have matching dimensions and both stated diffs recompute exactly: Diff0 is saturated RGB absolute difference ×16, DiffT16 is per-channel difference >16/255. The changed pixels are confined to the avatar areas listed under “what else changed”; no unrelated layout or clock drift appeared in accepted diff panels.

Web cells use CSS 1440×900 / 390×844, Chromium156/Linux, DPR2, zoom100%, 16px root font; the preview exports wide images1280×800 and narrow780×1688. Android cells are1080×2424, the declared Pixel9/API36 logical411.428571×923.428571, DPR2.625. Native metadata comes from delivered provenance; no new device was leased. Both themes match the labels.

W01 narration matches pointer expansion/collapse and keyboard-focused expansion/collapse across both widths/themes. W03 delivered clips show five members, +2, the failure specimen and disclosure/collapse without blank navigation frames. A02 clips show dark accent, yellow accent, then badge removal, and web status dots return bottom-right. A05 web clips show badge appearance and removal following project-setting changes. The A02 dark-after single save at14.68s is visible as badge removal by15.860933s; its long subsequent stable-frame interval is not a timing defect. Blue circles in disclosure videos are pointer-click indicators, not app loading spinners. Recordings have written narration in Evidence, not audio. W01/W03 exact key/action telemetry was not retained, so the visible behavior is corroborated but exact physical keys cannot independently be recovered from video alone.

The visible ACP logo, initials, light/dark accent text contrast, status-dot placement and badge/ring separation are supported for the delivered specimens. Claims about other provider logos, long path detail on W03, active clocks, remote environments, iOS or handoff behavior remain unproven. W03 puts the path/prose in titles and repeats an ordinary result detail; this does not exercise the required no-space path-detail branch. The PR body's older revision images are not substituted for this exact-head delivery.

| Row | Status | Delivered / missing |
|---|---|---|
| W01 | SUFFICIENT | Wide/narrow × light/dark; collapsed/expanded pairs and before/after disclosure recordings. |
| W02 | INSUFFICIENT | Wide light/dark and narrow dark all contain “Syncing messages…”; six filenames above. |
| W03 | INSUFFICIENT, partial | Six settled pairs and six clips; wide dark and required long prose/path-detail specimens missing. Controlled failure correctly labelled HARNESS. |
| W04 | INSUFFICIENT, missing | Active clock increments, settlement, mixed status and inactive hover. |
| W05 | INSUFFICIENT, partial | Wide light/dark static component HARNESS only; narrow dark and real/status transitions missing. |
| W06 | INSUFFICIENT, missing | Standalone/nonactionable, tooltip, focus and local open/return. |
| W07 | INSUFFICIENT, missing | Notification outcome matrix. |
| W08 | INSUFFICIENT, missing | Resumed notification, live status and fallback. |
| W09 | INSUFFICIENT, missing | Phased native workflow members and interactions. |
| W10 | INSUFFICIENT, missing | Unphased workflow, actionable/nonactionable and empty controls. |
| M01 | INSUFFICIENT, missing | Accepted Android/iOS interactions; rejected diagnostics do not count. |
| M02 | INSUFFICIENT, partial | Android light/dark pairs support all three initials; iOS light/dark missing. |
| M03 | INSUFFICIENT, missing | Native overflow, failed/long details and disclosure/scroll. |
| M04 | INSUFFICIENT, missing | Native live clocks and settlement. |
| M05 | INSUFFICIENT, unverified control | Native singleton and empty work-log observations absent. |
| M06 | INSUFFICIENT, missing | iOS wide inspector and phone navigation control. |
| A01 | INSUFFICIENT, partial | Grouped wide light/dark, narrow dark, Android dark; standalone web, Android light, iOS light/dark absent. |
| A02 | INSUFFICIENT, partial | Web light/dark and Android dark all three states + recordings; Android light and iOS light/dark absent. |
| A03 | INSUFFICIENT, missing | Default/custom/Unicode label resolution. |
| A04 | INSUFFICIENT, missing | Disabled historical/current siblings and re-enable. |
| A05 | INSUFFICIENT, partial | Web light/dark enabled/removed pairs + recordings; Android light and iOS light absent. |
| A06 | INSUFFICIENT, missing | ACP identity equivalence and live changes. |
| A07 | INSUFFICIENT, missing | Other provider logo morphology and fallbacks. |
| A08 | INSUFFICIENT, missing | Hydration, rename, invalid accent and status/error pushes. |
| R01 | INSUFFICIENT, missing | Actual remote environment scoping and open/return. |
| R02 | INSUFFICIENT, missing | Remote unavailable/config absence and reconnect. |
| R03 | INSUFFICIENT, missing | Cross-environment single-account control. |
| H01 | INSUFFICIENT, missing | Accepted strict native pairs, endpoint alerts, plural-source and iOS. |
| H02 | INSUFFICIENT, missing | Pending footer → committed divider. |
| H03 | INSUFFICIENT, missing | Failed/legacy/zero/multiple-source handoffs. |
| H04 | INSUFFICIENT, missing | Missing provider/model/config fallbacks and alerts. |
| D01 | INSUFFICIENT, missing | Electron integrated smoke. |
| Z01 | INSUFFICIENT, unverified control | Guards code-traced; required surrounding guard observations absent. Matrix requires no dedicated positive capture. |

Independent rebuild: ran the uploaded schema-decoding RPC seeder against the exact head from `/home/ubuntu/wt/pr-25031`, with `T3CODE_DEV_SEED=0` and explicit `--home-dir` inside the audit scratch directory. Historical seed phase used the uploaded frozen clock, then the owned server restarted without it before settings/captures. Changed only scratch paths in the driver. Used real pooled config and evaluated the uploaded production-component palette harness with an actual thread mounted. Environment: Chromium156/Linux, no emulation, CSS1440×900, DPR2, zoom100%, root16px, light theme; saved1280×800. Against `W05-W-light-HARNESS-palette-after.png`: **0/1,024,000 pixels differ**, maximum channel delta0, threshold16 count0, no diff bounding box. Two independent rebuilt captures also differ by0 pixels. This satisfies one affected-capture rebuild; it is explicitly a component HARNESS, not missing W05 transition proof.

Captured process groups360726 and369214 stopped; owned preview tab closed; both backend19223 and web13953 confirmed no longer listening. No worktree edits, git mutations, pushes, PR comments or new provider subscriptions. The lead's worktree and parent scratch directory are preserved. Audit scratch is removed only after artifact upload. Browser-managed snapshot artifacts are copied into the uploaded bundle; their tool-owned locations in the live T3 artifact store are not manually cleaned.

FRICTION:
- tool | download host rejected Python urllib's default request with403; curl with browser User-Agent succeeded | document the host's required User-Agent.
- model | initially misread a small A02 frame tile; enlarged frame comparison corrected it before verdict | inspect enlarged transition frames before making timing findings.
- tool | browser snapshots can only save to T3's managed artifact directory | allow an explicit scratch output path.

VERDICT: INSUFFICIENT W02–W10, M01–M06, A01–A08, R01–R03, H01–H04, D01, Z01