# Final-merge visual proof addendum

**SUFFICIENT for the focused four-cell web visual refresh at final merge. No missing affected row.** This addendum governs the current-main Connections UI; the prior full core audit continues to supply unchanged flows and its stated limitations.

Before [eabcfa2fea](https://github.com/pingdotgg/t3code/commit/eabcfa2feaba3be169aeec2da0207b8ed694d79d), full eabcfa2feaba3be169aeec2da0207b8ed694d79d. After [44b82a0d96](https://github.com/pingdotgg/t3code/commit/44b82a0d965053fd8c28fc577fa44507be78f927), full 44b82a0d965053fd8c28fc577fa44507be78f927. Checkout independently observed clean at the after SHA. Capture launcher checks exact revision before starting; driver and index bind these revisions. Video overlay labels BEFORE/AFTER, cell and HARNESS, not the full SHA.

[Delivered index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/649b6425e5dc9e1d/index.md). [Prior full audit](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8111996d77c57b62/audit-final.md) at [65637ec020](https://github.com/pingdotgg/t3code/commit/65637ec020438e9bf200645d36910374f3870279) remains the source of the 442-file verification and independent 0-pixel live-cleanup replay. Those old results were read and reused, not independently rerun or re-hashed in this addendum.

Scoped blob proof independently checked from Git: managedAuth.tsx bfafb5f5556aa8adfbbe5d8b119deda8078039d8, ConnectionsSettings.tsx, connection/catalog.ts and rpc/atomRegistry.ts are byte-identical between prior audited head and final merge:

- managedAuth: bfafb5f5556aa8adfbbe5d8b119deda8078039d8
- ConnectionsSettings: fea873dc4e529fdf852c13fe3cbadfca33dddb85
- catalog: 971fa891106fcabbe039ff4591ef1709d37fcfb6
- app registry: f78b6343f954f2f7ecdc43e6052096afefd545f5

SavedEnvironmentRows.tsx changed upstream since the prior audit, including the signed-out publishing Sign in row. It is identical between current-main BEFORE and final-merge AFTER. Against current main only managedAuth.tsx, its browser tests and vite.config.ts differ. This is scoped equivalence, not certification of all unrelated main changes.

All four paired cells are sufficient: 1440×1000 light/dark and 390×844 light/dark, DPR1, Chromium web. In every cell BEFORE seeds five targets, prematurely persists two at unresolved degraded/client_init launch while rendering relay ghosts, then shows two after same-IDB reload. AFTER retains five through launch and reload with no reseeding, then resource-only client_real notification removes relays live to two direct targets without another reload. Status remains degraded and auth hook state remains signed out; AFTER resource listener count is one, BEFORE zero. Telemetry and actual rendered endpoints agree. No-ID and other original matrix flows were not repeated; scoped blob proof plus prior full audit supplies them.

Only Clerk React state is replaced by the explicitly labeled HARNESS. Actual renderer, catalog, registry, commands and persistence remain in use. This is not fresh live SDK, OAuth, Connect transport, JWT or native-client proof. Synthetic relay failures are expected. Narrow Authorized clients overlap exists on both revisions and remains documented. Preserved rows change card height, control positions and scroll extent; threshold0 also exposes low-amplitude fills/borders. No new layout regression observed in the refreshed publishing row.

Independently inspected all 38 delivered PNGs through source/contact-sheet and comparison views: 26 application stills, four comparison panels, eight threshold images. Inspected all eight videos through each chapter and endpoint; fully decoded every recording without errors after finalization. Four chapters each, all inside duration, 12fps; content viewports are 1440×1000 and 390×844; recording dimensions 1440×1176 and 390×1020 include browser chrome and HARNESS label. BEFORE light/dark durations 23/21s at wide, 19.166667/19s narrow; AFTER 25.333333/21s wide, 21.583333/21.583333s narrow. Boot/reset frames are contextual recording steps, not accepted settled endpoints. All 26 stability files report byte-identical PNG pairs 500ms apart; capture source awaits fonts, fixed time and network idle. Matched comparisons use post-reload unresolved state, same viewport/theme/top scroll; narrow lower-list companions supply full saved-list visibility.

Independent recomputation exactly matches every threshold image and count (absolute maximum RGB-channel delta, image amplified4×):

|Cell|pixels >0|pixels >16|total|
|---|---:|---:|---:|
|1440 light|219952|15768|1440000|
|1440 dark|220326|16367|1440000|
|390 light|38602|11947|329160|
|390 dark|38668|12261|329160|

Public readback: all 98 published files independently downloaded and SHA256-match local packaged bytes, including eight recordings and the evidence bundle. All 97 regular bundle members match the corresponding packaged files. Downloaded index is byte-identical, SHA256 70f1e51e0277498dc570b1b31ed92e1c30cf9e69313efb253708d58477e95f3f. No missing/mismatched public file.

FRICTION: narrow AFTER video was read before capture finalized and returned moov atom not found; final delivered recordings were subsequently fully decoded and hash-checked successfully. Python default HTTP client received403; curl readback succeeded for the complete publication. Earlier in-progress files are not the basis of this verdict. No new independent browser replay was attempted: prior independent0pixel plus scoped blob proof was explicitly authorized for reuse.

No source edits, server/browser launch, checkout, push, comment or watcher by this auditor. Capture owner reports restored clean head and stopped servers/closed ports; that shutdown is owner-reported, not independently claimed here. No merge authorization or general CI verdict. Model/harness: gpt-6.1-sol / Codex.

[Evidence bundle](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9c5c3370dd13a510/finalmerge-evidence.tar.gz)

![BEFORE current main and AFTER final merge, with threshold0 and threshold16 differences](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ba18b2f4daca2e8c/1440x1000-light-comparison.png)