# PR 24406 server evidence index

Scope: [#24406](https://git.cafe/pingdotgg/t3code/pulls/24406), [GitHub mirror #24406](https://github.com/pingdotgg/t3code/pull/24406).

Pinned candidate [f78d4ef](https://git.cafe/pingdotgg/t3code/commit/f78d4efb83406ee495171e1252d176384a5aa582), base [6e79211](https://git.cafe/pingdotgg/t3code/commit/6e792117436662208bf2b6eb71ee8d0d88817e43).

**12/12 matrix rows captured, 40 deterministic cases + 1 real response, each before and after through production getChangeRequest.** Base candidate regressions: 2 fail (47 skipped), exact failures attached. Candidate focused suite: 49 pass. Candidate matrix driver: all expected outputs asserted, exit 0. Worktree restored clean, pinned HEAD unchanged.

Principal incident-shaped stale-count case: base returns cancelled + success / failing; candidate returns success / passing. The real GitCafe payload has a cancelled/replaced run but already-corrected aggregate counts: base 32 rows, candidate 30 rows, both passing. No claim that the live stale-count aggregate was reproduced. No unverified matrix rows.

Environment: fs-18, Linux x86_64, Node v26.12.1, vp v1.3.0, Vitest v5.0.5, Effect 4.0.5. No servers, UI, ~/.t3 reads, commit, push, watch, comment, or PR body mutation.

## Independent pass 2 replay

Keep the driver, matrix-fixtures.json, sanitized-live-response.json, and replay.py together in `/tmp/pr24406-evidence`. Installed dependencies and the exact candidate checkout at `/home/ubuntu/wt/pr-24406` are prerequisites. The replay script checks HEAD, temporarily swaps only the two production source files to base while retaining candidate tests, restores bytes in finally, runs candidate checks, and removes the temporary package test. It writes output beside itself. Do not run concurrently with another worktree edit. Archive extraction, if used, goes to `/tmp/pr24406-evidence`, not the worktree.

```sh
cd /home/ubuntu/wt/pr-24406
python3 /tmp/pr24406-evidence/replay.py
```

Local index: `/tmp/pr24406-evidence/index.md`
Local production HTTP driver: `/tmp/pr24406-evidence/proof-driver.test.ts`
Local replay entry point: `/tmp/pr24406-evidence/replay.py`

`transcript.md` expands all cases. `base-runtime.json` and `candidate-runtime.json` are exact observations from returned provider values; omission is recorded explicitly. `manifest.json` contains SHA256 hashes and final clean status. `commands.json` preserves exact argv and environment flags. Fixtures use only dummy test authentication and sanitized public payload data. The HTTP helper and service layers are copied from the existing fakeGitCafe test setup, not a copy of the production conversion logic. All production conversion/aggregation is imported from the actual checkout.

## Uploaded artifacts

[proof-driver.test.ts](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/69fabcc20f1e6cdc/proof-driver.test.ts)
[matrix-fixtures.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/de6cb9ecfc7d7fe2/matrix-fixtures.json)
[sanitized-live-response.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5b02d599eaa0a2e7/sanitized-live-response.json)
[replay.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/80d90aea8304bb0f/replay.py)
[base-runtime.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f239765273b638de/base-runtime.json)
[candidate-runtime.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e0b17af37f8581e0/candidate-runtime.json)
[base-regressions.log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/05b6389c88ac6529/base-regressions.log)
[base-runtime.log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/dd8d852bcb96463f/base-runtime.log)
[candidate-tests.log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/db7f9cfcdfc52964/candidate-tests.log)
[candidate-runtime.log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6fa2e482292b61d5/candidate-runtime.log)
[commands.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/20d74406f93c6e43/commands.json)
[transcript.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/82a07b6a216050b6/transcript.md)
[manifest.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/92a1d0148e987de6/manifest.json)

## Friction

FRICTION category|happened|prevention
Upstream drift|Remote PR advanced beyond requested candidate|Pin local SHA and record remote separately.
Payload provenance|Live rerun status counts were already corrected|Separate real row-removal evidence from deterministic stale-count proof.
CLI coverage|CI listing empty while checks endpoint accessible|Read typed cafe first, raw response via cafe api only for missing typed output.
Test stdout|Successful vp tests suppress console output|Persist actual returned values to JSON.