# Issue 10517 corrected native evidence — successor index

This index supersedes the F01 claims in the [historical index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/703cbe74be1d6982/index.md).

**WITHDRAWN:** historical `f01-base-dark` and its paired dark difference sheet are contaminated by a clipboard-helper JavaScript error dialog. They cannot establish a product-only before/after delta. The old head-dark recording may stand alone as appearance evidence, but the old pair is rejected. All four old F01 pairs are replaced below by eight new runs.

## Exact desktop-main build attribution

Before: [ef92441313](https://git.cafe/pingdotgg/t3code/commit/ef92441313da327139df75daa8f723c0f42348f6) (`ef92441313da327139df75daa8f723c0f42348f6`). After: [be1903680c](https://git.cafe/pingdotgg/t3code/commit/be1903680c3fd376b0a6fd3bc48e84ab3df544fa) (`be1903680c3fd376b0a6fd3bc48e84ab3df544fa`). Clean source checkouts, actual `git switch --detach ef92441313` then `vp pack` in apps/desktop; restored `fix/issue-10517` and rebuilt head with the same command. Saved to `.t3/evidence-base/main.cjs` and `.t3/evidence-head/main.cjs`. No fixture commit or production source change; final HEAD remains [be1903680c](https://git.cafe/pingdotgg/t3code/commit/be1903680c3fd376b0a6fd3bc48e84ab3df544fa).

Main SHA256 before: `8a322e48872f79c2c151eb0760f932bce3bd812d36468330c42f91c589e03273`. After: `08a06567650aa990960f8b9d7f7763fba09f53371d0fa689b0b8c58d3eb9b1ff`.

These are exact-revision **desktop-main** rebuilds. Existing server/web assets are reused, not complete application rebuilds. Shared artifact hashes and build commands are in [provenance](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f35d013769475e57/provenance.json); [base build log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0a3ff49892369f1b/build-base.log); [head build log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6fcb451c5778b5e8/build-head.log).

The committed capturer still hardcodes older SHAs. These replacement runs use a task-local [corrected capturer](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3618cef2f433c664/capture.py) which reads the exact build refs and asserts the selected main SHA256 before launch. Their manifests contain the correct refs. This provenance explicitly overrides the committed script's old hardcoded labels for these eight runs only. Historical runs retain their original attribution.

## Replacement F01 captures

Whole native X11 display, 1440×900 or 390×900, no crop/renderer styling. Adwaita light/dark; Electron 44.9.3 / Chromium 152.0.7977.167. No window manager. Same zero-project/zero-thread `.t3/healthy-home` SQLite seed, independently backed up per run; literal `projection_threads.branch_pull_request_json` removed, migration ledger retained. All eight child logs contain the literal missing-column error. No live developer data used.

Recording starts before app launch; timestamped captions are instrumentation, not audio narration. Head: keep the native dialog open through later retries, then Enter/Copy Diagnostics, exact same-X xclip readback, closed dialog and continued retries. Base: same first 33-second startup observation, no dialog/action target, then five seconds further observation. Base recordings end around 40 seconds; head around 48 seconds to include copy/readback/post-action observations. The version-only helper does not access clipboard; xclip owns and reads clipboard on the same X session. No clipboard JavaScript helper is launched.

| Native display/theme | Before — [ef92441313](https://git.cafe/pingdotgg/t3code/commit/ef92441313da327139df75daa8f723c0f42348f6) | After — [be1903680c](https://git.cafe/pingdotgg/t3code/commit/be1903680c3fd376b0a6fd3bc48e84ab3df544fa) | Diff0 / DiffT (>16 max RGB) |
|---|---|---|---|
| 1440×900 light | [whole-display still](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/64bb93eb5bdce6bd/before-action.png) · [startup video](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/77e229f8cb72ec5d/recording.mp4) | [whole-display still](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/76d71e19b62bad92/before-action.png) · [startup + Copy video](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b1c24d68a1a1bb16/recording.mp4) | 144,460 / 144,460 of 1,296,000; [before/after + masks](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b3b4525252eb5a82/f01-1440-light-sheet.png) |
| 1440×900 dark | [whole-display still](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ca49d67fcef0a2d8/before-action.png) · [startup video](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e2709b47487fa0af/recording.mp4) | [whole-display still](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a0d5077ef92a45b4/before-action.png) · [startup + Copy video](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ced82b5279cf6c82/recording.mp4) | 142,596 / 141,928 of 1,296,000; [before/after + masks](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0de8da105772c238/f01-1440-dark-sheet.png) |
| 390×900 light | [whole-display still](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a55577f05bc88475/before-action.png) · [startup video](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/698a25f5578943d4/recording.mp4) | [whole-display still](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4fae5e83f5d772f4/before-action.png) · [startup + Copy video](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a387521fdc8d96c3/recording.mp4) | 124,800 / 124,800 of 351,000; [before/after + masks](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/30f5889a79ae3cce/f01-390-light-sheet.png) |
| 390×900 dark | [whole-display still](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/97051f29ed40c7de/before-action.png) · [startup video](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/68c574ed3f14e7da/recording.mp4) | [whole-display still](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/dcf5ca14deda3e03/before-action.png) · [startup + Copy video](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f409b7ca796c1a93/recording.mp4) | 124,800 / 124,354 of 351,000; [before/after + masks](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c02fc885e64b144f/f01-390-dark-sheet.png) |

Diff0 means any RGB channel differs; DiffT means max absolute RGB channel difference >16. Masks use original whole-display before-action stills, with no captions burned into stills. These are absent-dialog → native-dialog differences, not subtle rendering regressions; counts include native text, background, controls and antialiasing. Narrow dialog is 390×320; wide is 612×233; both buttons visible. No macOS/Windows/WSL claim.

[Run audit: manifests, timelines, app/child logs, clipboard, ffprobe, cleanup](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d21a80adee03023f/audit.json) · [exact pixel counts](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/276ef70e28301854/diff-metrics.json) · [sampled video frames](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5cba5c4b98f2c826/video-contact.png)

All four new base stills are clean black displays without the prior helper-error box. Sampled video frames at 1/15/32/38 seconds were visually reviewed for all eight runs; this is sampling plus timeline/log correlation, not uninterrupted human playback. Head dialog window identity is stable across retries after threshold; all four Copy readbacks exactly match two diagnostic lines with each run's real child-log path. No replacement dialog by final observation.

## Retained matrix and historical source limits

Only F01 is newly captured at the exact requested desktop-main revisions. All other rows below retain historical earlier-source evidence and their limitations; none are promoted to exact-final-pair proof. The three relevant source files (DesktopObservability, DesktopBackendManager, DesktopBackendPool) have identical git blobs between earlier/requested bases and between earlier/requested heads; that supports source equivalence, not retroactive SHA attribution. [Exact blob comparisons](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d7b8ba3d3a7659fe/source-comparisons.json). Historical earlier head is [ef5a77d2f1](https://git.cafe/pingdotgg/t3code/commit/ef5a77d2f16e5612a97e1e388afbcafe4f7c1422), earlier base [52c2b38ea4](https://git.cafe/pingdotgg/t3code/commit/52c2b38ea4). Historical run links remain in the archive linked above, subject to this withdrawal.

| Row | Status | Scope / reason |
|---|---|---|
| F01 | PASS Linux exact desktop-main revisions | Eight replacements above: 1440/390 light/dark, literal missing-column, native controls and Copy. Native macOS/Windows unverified. |
| F02 | PASS on Linux at 1440 light | Literal non-SQLite bytes; child log says not a database. Static chrome/reason shape collapses into F01; additional platform/theme timing not claimed. |
| F03 | UNVERIFIED branch; fixture counterexample captured | server-child.log directory does NOT produce unavailable: lazy writer advertises an unusable path. Blocking logs parent instead aborts tracing before backend startup in both revisions. No production hook/injected dialog used. No 390 unavailable proof. |
| F04 | PASS on Linux 1440/390 light/dark | Real missing built server entry; absolute reason path copied. See run list for 1440/390 light/dark completion. No macOS/Windows evidence. |
| F05 | UNVERIFIED | No deterministic real-app seed for configuration resolution failure after desiredRunning=true; no synthetic callback injected. |
| F06 | UNVERIFIED | No deterministic initial configuration-resolution failure seed. Parent-log failure is earlier tracing initialization and is not this row. |
| F07 | CANNOT VERIFY | Windows/WSL and actual unbounded preflight configuration unavailable on Linux host. |
| F08 | CANNOT VERIFY | Windows/WSL bounded preflight/fallback environment unavailable; focused spy test is not native proof. |
| F09 | CANNOT VERIFY | Windows/WSL fatal preflight and persisted fallback environment unavailable. |
| F10 | PASS Linux subset | New F01 runs keep the same dialog open across later retries; base retries without dialog. Recovery/reset/global dedupe unverified. |
| F11 | PARTIAL: Linux path-present PASS | Default Enter and mouse click return exact two diagnostic lines in same-X-session xclip readback. Initial light readback helper failed; dark/narrow Enter and f11-click independently succeed. Unavailable text branch and macOS/Windows keyboard order unverified. |
| F12 | PARTIAL historical | Escape and Close via Tab+Return captured; actual mouse Close/title-bar dismissal, native macOS/Windows semantics unverified. |
| F13 | UNVERIFIED | No deterministic real-app ready→new-failure-streak seed established within budget. Possible overlapping dialogs after ready/reset not tested. |
| F14 | UNVERIFIED | DB cases are ordinary code=1 exits; missing entry is exists=false. Neither is a distinct readiness/spawn-error fixture. |
| F15 | UNVERIFIED | No controlled native dialog rejection or clipboard rejection injection. Five existing pool tests passed but do not establish real OS rejection behavior. |
| C01 | PASS on Linux 1440 light/dark | Fresh snapshot of clean zero-project/zero-thread home. Matched 1440 light/dark recordings and main-window stills when complete. No 390 main-window/mobile claim. |
| C02 | UNVERIFIED full sequence | Capture cleanup deliberately stops own app, but repeated ready-on-every-restart plus normal-shutdown sequence was not separately seeded/recorded. |
| C03 | UNVERIFIED | No competing server launched; only isolated desktop processes, no live developer home accessed for testing. |
| C04 | CANNOT VERIFY | Secondary/WSL backend environment unavailable. |
| C05 | SOURCE COVERED / out of capture scope | Diff affects desktop observability, backend manager/pool and focused pool tests only. Callback supplied only for primary local desktop. No web/mobile/provider/relay/tunnel UI changes; no browser used. |

## Cleanup and FRICTION

[Cleanup verification](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/30758a553fb3c63f/cleanup.json): all 32 captured application/version-helper/xclip/recorder PIDs gone; all eight Xvfb/D-Bus sessions exited and no session keyring remains. Pre-existing system/session services untouched. `fix/issue-10517` restored, clean git status, head main restored byte-for-byte, server entry present. No push, PR or commit. Old logs and saved lead bundles preserved; originals backed up under task scratch.

FRICTION: committed capturer's hardcoded SHAs corrected only in an external task-local copy and explicit provenance manifest; the repo fixture remains unchanged. Longer task output paths make the new narrow native dialog 320px tall (historical F01 303px); both buttons fit. Shared server/web assets reused and fully disclosed. No unsupported fixtures/platforms chased; all unverified rows retained. No dependency installation, browser, or live-home access.

Scratch retained at `/tmp/issue10517-correction` (build logs, original bundles, corrected capturer, fixture homes, raw logs, recordings, frame samples, audit, upload lists and this index); preserve until lead cleanup. Existing `/tmp/issue10517-capture` and proof-pass files unchanged.