# Proof audit pass 2

1. **R11 — independent rebuild does not reproduce the accepted expansion capture.** Rebuilt `R11-D-K-T2-11-expand-60ms-after.png` twice at [f4bbe57480](https://git.cafe/pingdotgg/t3code/commit/f4bbe57480e4498d279fce1dcf99b1aece1100b5), using the uploaded `more-rows.js` fixture recipe and `capture.mjs` helpers, including the ordinary recording followed by the paused replay. Both rebuilds differ identically from delivered drop `8d9e8b104a613e52`: **4,403 pixels at Diff0, 2,519 at Diff16, maximum channel delta 244**. Of these, **2,395 Diff16 pixels are shelf content**: a fading Loading clone overlaps Settled T2, and shelf content is displaced. Another 124 Diff16 pixels are an unrelated bottom-right update icon, also present as a discrepancy in initial/final/collapse frames (851 Diff0/124 Diff16 each). This exceeds antialiasing. The accepted frame is visually clean, but its reproduction is not. This establishes a proof/reproducibility failure, not by itself a code regression. **Needed:** repeat R11 D-K T2 expansion with the load/cache precondition recorded, matching before/after checkpoints, both diffs, and uninterrupted recording; resolve or accurately qualify the asynchronous transition.

   ![Delivered R11 after on the left; independent same-head rebuild on the right, showing overlapping Loading and Settled text](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/cbc0711755ea5c58/R11-delivered-vs-rebuilt.png)

2. **R13 — after recording has the wrong resolution.** [R13-D-K-r6-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0310f06c911c8de8/R13-D-K-r6-after.mp4) is **390×244**, a downscaled desktop image, while its before recording and stated D-K cell are **1440×900**. Text is unreadable at native size. The full-size checkpoint PNGs cannot replace this priority motion recording. **Needed:** replace this after clip at native 1440×900, showing the same nine narrated shelf actions and the in-flight/rapid-reversal sequence. All other 71 accepted MP4s have their expected desktop/mobile dimensions.

3. **R08 — the disclosed missing multiselection Settled expansion remains insufficient.** `R08-D-K-clean-multiselect-settled-expand-60ms-after.png` in [partial diagnostics](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/de96e5fb712c2bae/R08-partial-diagnostics.zip) contains a Loading clone superimposed on Settled alpha. The [after recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bcd3ff03582eb2b7/R08-D-K-clean-multiselect-after.mp4), after its 2.684s expansion action, also exposes this asynchronous branch. Excluding that pair was correct; the rest of R08 is valid. **Needed:** exactly the D-K multiselection Settled-expand phase, paired with identical loaded data/cache state and a recording covering the real load transition. The analogous independent R11 failure makes this a concrete proof gap rather than a demand to finish every inventory row.

4. **R18 — the Settled reduced-motion recording does not show the stated rows.** [R18-D-K-r2-T-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/af2bc71e4c2e921f/R18-D-K-r2-T-after.mp4) shows Loading during the expanded interval (approximately 0.86–1.00s inspected frames), then collapses. The before clip does likewise. The settled stills and JSON correctly show actual reduced-motion media and zero active animations, but are not proof of the recorded transition with actual Settled rows. **Needed only to retain the matrix's all-three-shelves reduced-motion claim:** replace the T recording pair after the real rows are available. Working/Snoozed/ordinary-move controls remain usable. This extra control is not necessary to establish the narrow normal-motion claim.

5. **Rendering provenance is inaccurate.** Index, proposed Evidence section, provenance and rebuild instructions say Chromium **156**. Geometry JSON says `HeadlessChrome/148.0.0.0`; the specified executable independently reports **148.0.7778.96**. The actual matched environment was Linux x86_64, Playwright-core 1.60.0, Chromium 148, DPR1, en-US, UTC, 1440×900/dark for the rebuild. The public scripts' 15:13Z display clock and native font scale were preserved. No Chromium156 result is established. Correct the environment line throughout; this alone does not require recapturing all rows.

6. **The published PR body still describes superseded proof.** GitCafe [#25412](https://git.cafe/pingdotgg/t3code/pull/25412) currently links the old 11-pair synthetic-projection/DPR2 evidence and says “Final proof recheck: SUFFICIENT.” That is not the accepted real-RPC/DPR1 bundle audited here. Replace those verification/evidence claims and links with the corrected current section; do not carry forward the unconditional sufficiency assertion. Old linked artifacts were not substituted for this pass's accepted catalog.

| Row | Status | Coverage and relevance to the narrow claim |
|---|---|---|
| R01 | SUFFICIENT | Reported overlap visibly reproduced before and absent after; all four neighbor states, both directions, 1440/390 widths, dark/light, desktop Enter/Space. |
| R02 | SUFFICIENT | Classic design, D-L, all three shelves both directions. |
| R03 | SUFFICIENT | Snoozed, D-K, all neighbor combinations and keyboard cycles. |
| R04 | SUFFICIENT | Settled, D-K, all neighbor combinations and keyboard cycles. |
| R05 | SUFFICIENT | Routed Working retained; shared backend disclosed, full base frontend used. |
| R06 | SUFFICIENT | Routed Snoozed retained. |
| R07 | SUFFICIENT | Routed Settled retained. |
| R08 | INSUFFICIENT / partial | Singleton outside/retained and multi W/Z/T-collapse valid; multi T-expand specifically needed as above. |
| R09 | SUFFICIENT | No Working: Snoozed owns bottom anchoring. |
| R10 | SUFFICIENT | No Snoozed: Working/Settled transitions. |
| R11 | INSUFFICIENT | Delivered T2/T0 captures look clean, but independent T2 expansion reproduction fails. |
| R12 | SUFFICIENT | Working beta off, Snoozed/Settled transitions. |
| R13 | INSUFFICIENT | Checkpoints valid; priority after motion clip is downscaled and unreadable. |
| R14 | SUFFICIENT | Ordinary pin/unpin, settle/unsettle, snooze/unsnooze, archive/unarchive/delete and real runtime start/stop movements still animate. Expected entry/exit fades are ordinary-motion behavior. |
| R15 | UNCAPTURED / inventory | Crowded/mixed-height/long-name/draft/min-width extension. Not needed: key guard is independent of row size; full/ slim cards and 390-wide drawer already covered. No typography/draft code changes. |
| R16 | PARTIAL / not credited | Pagination diagnostics have unmatched bottom clipping/scroll and hover state; `R16-D-K-r2-final-Diff16.png` and retained expansion show unrelated differences. No accepted page-growth pair. Inventory beyond this claim: pagination/projection/retained-page logic unchanged; retained slim rows covered R07. Do not claim pagination verified. |
| R17 | SUFFICIENT | Search results/no-results/clear, Settings return, desktop/mobile drawer remount. Blank no-results area is intentional. |
| R18 | INSUFFICIENT / partial | Actual media preference established; T recording shows Loading. Optional for narrow claim; replace T clips if retaining all-toggle reduced-motion assertion. |
| R19 | UNCAPTURED / inventory | Retry/failure harness not needed: response/error handling unchanged. The observed asynchronous Loading issue still needs focused R08/R11 follow-up; no need to commission the whole failure/Retry row. |
| R20 | UNCAPTURED / inventory | Drag/context-drop/sweep/undo extension not needed: existing drag suspension and separate release reconciliation unchanged; membership moves covered R14. No claim of verified undo-notice geometry. |
| R21 | UNCAPTURED / inventory | Cold mount/empty/draft/persisted reload not needed: initial key equals initial ref, existing mount suppression unchanged; remount exercised R17. |
| R22 | UNCAPTURED / inventory | >40 entering/exiting candidates: existing controller suppression already calls the nonanimated path. Cap and candidate calculation unchanged. |
| R23 | CANNOT VERIFY / platform extension | macOS Electron not run. Shared-web inference only; not needed for the stated shared-component claim, and no macOS-verified claim allowed. |
| R24 | UNCAPTURED / inventory | Filter/reveal/remote-decoration/min-width extension not needed: scope/projection/contracts unchanged; actual project-scoped transitions and search already covered. No remote-environment verification claim. |

Coverage: inspected all **201 accepted phase pairs** (402 before/after PNGs and 402 Diff0/Diff16 images), all 56 delivered sheets, and all **72 accepted recordings** through decoded initial/action/final frames matched to their timestamped written narration; seven clips additionally reviewed frame by frame. Native sidebar detail sheets cover the entire sidebar, with full-viewport originals/diffs also inspected. Opened both labelled partial diagnostic bundles and their R16 recordings. Superseded/uncredited attempt archives are not accepted proof and were not fully reviewed. Recomputed all 402 diff images: their pixels match the stated formulas. Only two accepted phases have any outside-sidebar Diff0 noise (8 antialias pixels each); outside-sidebar Diff16 is zero. R01 settled endpoints are byte/pixel equal. SHA columns and checked source/dependency identity match the requested head and full merge-base. Before motion overlap is the expected bug, not a capture defect. The “what else changed” description otherwise matches the code and accepted diffs, including cancellation of preexisting ordinary motion on toggle.

Rebuild limits: isolated fresh RPC-seeded R11 fixture, no provider turns required; original normal-recording/checkpoint functions retained, side loop restricted to after and absent-shelf guard taken from more-rows.js. Only filesystem/port/auth bindings adapted. Additional global fixture threads were not created; the inspected project scope contains the same R11 rows and unchanged main-chat fixture. Two complete successful replays give identical counts above. This is a real pixel discrepancy, not a claimed byte-identical rebuild. Credentials are excluded from the archive; server log is redacted.

No source edits, pushes or comments. Own browser and server stopped by captured PID; no provider turn started; own fresh worktree `.t3` deleted. Lead-owned worktrees/mirror/capture directory preserved. Audit ended within the 60-minute budget; coverage limits are stated above. Audit scratch removed after upload.

FRICTION:
- infra | Intermittent DNS blocked original downloads and the first audit-image upload; local lead mirror plus retry restored access | Include a local complete artifact bundle in the handoff.
- tool | Shared Browser panel DPR2 could not match the delivered DPR1 environment; exact uploaded Playwright executable/driver was used | Expose DPR control or hand off the precise runnable driver.
- forge | Some GitCafe review/status reads returned ConnectionRefused; PR body/head read succeeded | Stable read connectivity.

VERDICT: INSUFFICIENT R08, R11, R13, R18; rendering provenance and published Evidence section