# Supplemental independent proof audit — SUFFICIENT

Evidence-only PASS2 continuation for [#23210](https://github.com/pingdotgg/t3code/pull/23210), mirror [#23209](https://git.cafe/pingdotgg/t3code/pulls/23209), branch `fix/issue-13706`, head [c81e17ac40](https://git.cafe/pingdotgg/t3code/commit/c81e17ac400c3e6c52503906cd69b2e0ee077b5d), supplied base [8acbefefc5](https://git.cafe/pingdotgg/t3code/commit/8acbefefc51b4d281dcd04108c82baa1ab902b0e).

**SUFFICIENT for the narrowed matrix under the explicitly authorized synthetic client-replay boundary and R3 unchanged-behavior CONTROL purpose. No rows remain missing at that proof level. This is evidence sufficiency, not a merge-readiness verdict or proof of natural provider/server reproduction.** The earlier audit remains an accurate record of its then-missing evidence; this supplement closes those rows without replacing it.

Read the completed [supplemental index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/60a2bafab6f11c8e/supplemental-index.md), [capturer report](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/42d42187c839346d/supplemental-report.md) and [append fragment](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/80bfdb741e4a6964/supplemental-append.md) before deciding. Reused the [matrix](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/1d542c8161495ba7/matrix.md) and [prior independent audit](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/cacb8ac028a556f3/final-proof-audit.md). A1 was not rerun: its accepted independent rebuild remains threshold0=0, threshold8=0.

## Independent inspection and provenance

Read natural-R3 base/head JSON, all supplemental W1/W2/W3/W4 traces, replay instrumentation and action scripts, source callers, completed report/manifest, upload inventories, stability data and scratch Git reflog. Independently sampled six timestamped frames per raw W1, W2, W3-complete, W4-branch and W4-palette recording, and both natural-R3 recordings. This is sampled visual inspection plus full trace analysis, not a claim to have watched every frame. Narration is supplied by UTC phase badges/scripts/report; phase labels alone are not credited as successful actions.

Verified local HEAD and recomputed patch, lock, canonical materialized fixture and all six installed bundle hashes against the accepted manifest: all match. [Independent source hashes](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5899f4bade4804b6/audit-supplemental-source-hashes.json). React ESM remains `5f493ab687559b25747ecf3f294008e815e35fc918ac86e20f2271a908ae302f`. Independently regrouped raw trace samples into stable intervals requiring no target, at least three samples, and at least 200ms of identical total/DOM/internal/end geometry: [trace summary](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/17069d245b06e576/audit-supplemental-trace-summary.json).

Supplemental environment is capturer-reported Linux Chromium154, CSS1280×800, DPR2, zoom1, light, system sans; raw video2560×1600 and PNG1280×800. These are final-head controls, not newly paired base/head product runs. Primary A1 paired provenance remains in the accepted audit. No fresh browser, server, fixture execution or test was needed by this auditor.

## Exact row disposition

| Row | Verdict | Evidence and practical limit |
|---|---|---|
| R1 / A1 | SUFFICIENT, preserved | Accepted paired primary-race evidence and independent rebuild, full-image threshold0/8=0. No rerun, altered threshold or relabeling. |
| R2 | SUFFICIENT, preserved | Accepted extracted installed-function size-only callback proof across all six bundles. Execution diagnostic, not natural size-only lifecycle proof. |
| R3 | SUFFICIENT CONTROL | Natural warm-up measures150 rows without setItemSize. Target149 survives mutation on both revisions (base151.2ms/head150.9ms). Both finish with actual row149 height1000, measured total135100, DOM/internal134200, viewport800, end100. This closes unchanged valid-target geometry/size behavior. It does **not** meet absolute end≤1; equal preexisting gap is not a guard regression. |
| R4 | SUFFICIENT, preserved | Accepted bypass recordings and selected-branch diagnostics. |
| W1 | SUFFICIENT client integration replay | Actual history8000→latest, four streaming text chunks, tail append, animated latest, finalization/coalescing. Active target index61/viewPosition1 resolves the tail before authored final snapshot; two post-render frames have targetId=null with total15973. Final DOM/internal15473, extent16221, viewport748, end0; identical geometry848 samples over15.672s. Prior short/underflow evidence reused. No natural provider/server coalescing claim. |
| W2 | SUFFICIENT bypass controls | Prior remembered-middle8700 and find108 reused. Citation109 exact quote highlighted at y172..188.5, DOM/internal12948 stable973 samples/19.781s. Normal latest button with reduce-query shim invokes animated:false and settles DOM/internal16002/end0,44 samples/1.451s. Preference shim is not native OS/browser preference emulation. |
| W3 | SUFFICIENT nested client replay | Actual nested list append through reducer1200→1224/scroll912→936/end0; same-key content replacement and final-key60→61 replacement retain1224/936/end0; removal returns1200/912/end0. Collapse/reopen restores500, total1200,viewport288,end412; stable42 samples/1.232s. Nonanimated maintain-end dataChange=true observed on append. Snapshot replacement/removal are synthetic inputs. |
| W4 | SUFFICIENT ancillary controls | Prior branch400→filter0 and palette keyboard-nearest scroll274 evidence reused. New branch selection chooses existing alternate, reopens selected/current with filter empty/offset0; reverse to main also reopens current. Real scratch Git reflog corroborates both checkouts. Corrected palette retry reaches Project settings and records reopen with empty filter/offset0. Earlier unsuccessful Enter is not credited. |
| B1 | SUFFICIENT, preserved | Accepted six-bundle executed branch diagnostics and identities. No additional native/UI grid implied. |
| N1 | VERIFIED per user | Plain EAS iOS fixture accepted as instructed; not independently reverified here. No Android, desktop or additional iOS obligation added. |

R3 nuance: in these natural traces, the last row's eventual1000px measurement occurs after the active target clears. This visual control establishes unchanged lifecycle/geometry, while the accepted six-bundle valid-resize diagnostic supplies direct active-target +100 correction coverage. Neither should be described as a visual proof of final end0. Changing implementation or timing policy to force end0 would exceed this audit and is unnecessary for this control's regression purpose.

## Product source and injection boundaries

The actual source path is [useThreadProjection](https://git.cafe/pingdotgg/t3code/blob/c81e17ac400c3e6c52503906cd69b2e0ee077b5d/apps/web/src/state/entities.ts#L153) through the shared thread-detail projection atom into [ChatView](https://git.cafe/pingdotgg/t3code/blob/c81e17ac400c3e6c52503906cd69b2e0ee077b5d/apps/web/src/components/ChatView.tsx#L464), then MessagesTimeline/useTimelineFollow. Replay publishes to that live client state. Item updates call the real applyOrchestrationV2ProjectionEvent turn-item.updated reducer; final coalescing/key replacement/removal publish authored projection snapshots. The existing renderer derives rows, keys, measurements and follow behavior. The inspected scripts do not substitute a fake list or force LegendList target/size/cache geometry. Fiber inspection reads runtime state; ordinary DOM scrolling positions the replay.

[MessagesTimeline](https://git.cafe/pingdotgg/t3code/blob/c81e17ac400c3e6c52503906cd69b2e0ee077b5d/apps/web/src/components/chat/MessagesTimeline.tsx#L1093) owns MVCP/follow gates; [useTimelineFollow](https://git.cafe/pingdotgg/t3code/blob/c81e17ac400c3e6c52503906cd69b2e0ee077b5d/apps/web/src/components/chat/useTimelineFollow.ts#L261) reads matchMedia when normal latest is invoked. The W2 wrapper logs/delegates the real scrollToEnd call rather than replacing it. Citation input uses the actual assistantCitationNavigation helper and mounted router, then the real quote highlighter; it is not a recorded click on a rendered citation link. [TimelineWorkGroupRows](https://git.cafe/pingdotgg/t3code/blob/c81e17ac400c3e6c52503906cd69b2e0ee077b5d/apps/web/src/components/chat/TimelineWorkGroupRows.tsx#L245) supplies the nested real LegendList and data-change maintain-end. Branch filtering explicitly resets offset; palette uses DOM nearest or scrollIndexIntoView.

Thus W1 contains an actual stale target in a **synthetic client replay**. It is not natural provider repro, backend finalization proof or wire-contract validation. The permitted negative-smoke fallback would also have been acceptable with R1 positive cause proof; this supplement is stronger at the client boundary without enlarging its claim. The reconstructed W1/W2 action files are readable recipes, not byte-exact logs of tool-boundary timing; actual UTC/frame traces determine observed timing.

## Visual evidence

[W1 sampled frames](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b8016c1551de7945/audit-supp-supplemental-W1-sheet.jpg), [W2 sampled frames](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9d371294b48155ba/audit-supp-supplemental-W2-sheet.jpg), [W3 sampled frames](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d9d8d78390f688a4/audit-supp-supplemental-W3-complete-sheet.jpg), [W4 branch samples](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4a9ec9b2dacb326c/audit-supp-W4-sheet.jpg), [palette samples](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a61870e475095d3e/audit-supp-palette-sheet.jpg). Raw recordings, traces and supporting archive are linked in the final capturer report. W4 combined recording is a disclosed montage with retained cuts, not one seamless synchronized interaction.

![Actual nested client replay: append, replacement, removal, collapse and reopen sampled from the complete recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d9d8d78390f688a4/audit-supp-supplemental-W3-complete-sheet.jpg)

## FRICTION

- One packaging mismatch: supplemental-manifest.json records the derived W4 montage as1048624bytes, SHA256 `ce57df4d2cccd999fdc568118034aef13d3bfb4be198c9e5ea8dc7b46d5d5919`; completed local/upload inventory is1404063bytes, actual SHA256 `56e03401da552b249055b162b39869b91819ae93cdc0fa112f332552a48ee426`. Of28 manifest entries,27 match and1 is stale. The montage was finalized after its manifest entry. The final [checksum inventory](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e48cf3b74223a938/supplemental-checksums.json) carries the correct montage hash and independently verifies in full. Rely on verified raw recordings/traces for the verdict; this report records the completed montage identity without modifying the original manifest. This metadata flaw does not remove required behavior evidence.
- Failed citation attempts precede correction to the32-character context contract; only the exact highlight and stable position are credited. First palette Enter occurred while searching and caused no navigation despite its phase label. Corrected retry actually navigates to Project settings; do not call it Open settings.
- Capturer disclosed trace serialization limits and repeated W3 with compact trace storage. Complete W3 repeat and parseable traces, rather than filenames or stills alone, support the row. Empty W4-server-events JSON is not checkout proof; real scratch reflog and visible selected/current states are.
- Source-path guesses for the shared thread-detail file initially missed; rg resolved it. No browser fallback or verification expansion was needed.
- Capturer report retains existing server67043 for lead cleanup and says its own supplemental servers stopped. Auditor started/stopped no server and did not independently certify listener teardown.
- No implementation/test/repo edits, test execution, push, comments, PR body edits, merge, watch, CI polling or child delegation by this auditor. Only scratch audit artifacts and authorized public uploads written; required app PR registration performed. No new A1 rebuild. Original sufficient results and previous report preserved.

Audited independently by **gpt-6.1-sol, Codex harness in T3 Code**, using pr-audit and ghdrop. Final evidence verdict: **SUFFICIENT — R1,R2,R3,R4,W1,W2,W3,W4,B1; N1 accepted verified per user.**