# Incomplete evidence capture for PR 25697

**No paired row is verified. No before/after behavioral diff exists.** The budget elapsed while dependency installation and tool calls were outstanding. Do not accept this as completed M1 proof or a successful regression run.

PR: [#25697](https://git.cafe/pingdotgg/t3code/pull/25697), [GitHub #25697](https://github.com/pingdotgg/t3code/pull/25697).
Before: [6f6c6c14b7](https://git.cafe/pingdotgg/t3code/commit/6f6c6c14b76e72d7abd43b8e6ebffff17eadc89e). After: [39da8b2458](https://git.cafe/pingdotgg/t3code/commit/39da8b24584ab2f453dd4835713d4b221e0df955).

Environment: fs-14; Node v26.12.1; Linux x86_64 6.1.102; Vitest v5.0.5; real runtime/event sink/projection store; in-memory SQLite; registered Codex stub provider; synthetic data; injected completion cohorts and terminal events; no real model execution. No server/browser was started. No live userdata or lead checkout was read.

Counts: 0/17 affected rows have complete paired evidence; 5/22 matrix rows have partial after-only captures (R01,R02,R03,R19,R09); 21 rows are runtime-eligible and R21 is excluded by the supplied source audit. R09 is a control, so 4 affected rows have partial after-only observations.

## Commands and outcome

Both detached worktrees were created at the requested SHAs, and global `vp install` completed in each. Reported installation durations: before 8m52.7s; after 1h26m9.6s. Tool host generation became stale during a wait, preventing timely recovery; first clock was 07:29:14 UTC and next successful clock after the stale-host error was 08:56:31 UTC. Budget was 55 minutes. No more runtime captures were attempted after detecting the overrun.

Executed from after/apps/server: `vp test run src/orchestration-v2/queue-evidence.test.ts -t 'R01|R02|R03|R19|R09'`. Suite result: 5 failed, 6 filtered out. Shell exit code was not returned by the stale tool host; raw log contains the failed-suite report. Before test command was not executed. The exact same driver bytes were copied into both worktrees; no per-side API shim was used. Only after-side M1 executed.

## Harness failure and interpretation

The observer incorrectly queues every run.updated whose payload status is starting, including updates to blocker A's completion cohort that retain its existing starting status. When the drain loop consumes one of those old A entries after injecting A's terminal event, the current projection correctly says A is completed, causing `expected 'completed' to equal 'starting'`. This is a harness bug, not an established PR defect. The label `PERSISTED PROMOTION` is consequently overbroad for A: those entries do not prove a transition. Actual first queued promotions (U/C1/W/U/C1 for R01/R02/R03/R19/R09) appear later as genuine persisted events. Full ordered promotion sequences were not recorded.

The retained R01 regression assertion exists in the exact driver, but was never reached because the drain assertion failed first. It has not been demonstrated failing on before/passing on after. Rows beyond M1 in the driver were not executed. Goal, steering and resume fixtures were not implemented within budget.

## What else changed

Unknown: without a baseline runtime log, behavioral differences cannot be enumerated or attributed. No claim that changes are confined to the PR title is justified by this capture. The observed harness observer failure is not evidence of an additional production change. `comparison-unavailable.diff` explicitly records this limitation; it is not a fabricated unified diff. The normalized after log is supplied for later comparison.

## Evidence

| Row | Before (6f6c6c14b7) | After (39da8b2458) | Log section |
| --- | --- | --- | --- |
| R01 | UNVERIFIED: not run | Partial: reorder accepted; U=true; order U,C1,C2,C3,V; persisted U→starting. Remaining promotions unverified. | [R01 raw log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/661894fa389bd808/after-m1-attempt1.raw.log), CAPTURE SECTION R01, line 6 |
| R02 | UNVERIFIED: not run | Partial: accepted between completions; order C1,U,C2,V; persisted C1→starting. Remaining promotions unverified. | [R02 raw log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/661894fa389bd808/after-m1-attempt1.raw.log), CAPTURE SECTION R02, line 88 |
| R03 | UNVERIFIED: not run | Partial: ordinary-anchor moves accepted; W,V,U,C1,C2; persisted W→starting. Remaining promotions unverified. | [R03 raw log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/661894fa389bd808/after-m1-attempt1.raw.log), CAPTURE SECTION R03, line 159 |
| R04 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R05 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R06 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R07 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R08 | UNVERIFIED: not run | UNVERIFIED control: neither captured nor separately validated with existing tests | No runtime section |
| R09 | UNVERIFIED: not run | After-only: C2→C1, C2→null, N→U rejected with exact automatic-delivery error; no command run events. Baseline unverified. | [R09 raw log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/661894fa389bd808/after-m1-attempt1.raw.log), CAPTURE SECTION R09, line 367 |
| R10 | UNVERIFIED: not run | UNVERIFIED control: neither captured nor separately validated with existing tests | No runtime section |
| R11 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R12 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R13 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R14 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R15 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R16 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R17 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R18 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |
| R19 | UNVERIFIED: not run | After-only real handler list/reorder/read: cursor pages 0/2/4, sequence response, forbidden automatic move, truncation/read captured. No paired baseline. | [R19 raw log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/661894fa389bd808/after-m1-attempt1.raw.log), CAPTURE SECTION R19, line 251 |
| R20 | UNVERIFIED: not run | UNVERIFIED control: neither captured nor separately validated with existing tests | No runtime section |
| R21 | Excluded by supplied source audit | No changed client render path | Source audit only; no browser launched |
| R22 | UNVERIFIED: not run | UNVERIFIED: budget exhausted before this priority group | No runtime section |

## Artifacts

[queue-evidence.test.ts](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4d1e75fa156f8832/queue-evidence.test.ts)
[after-m1-attempt1.raw.log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/661894fa389bd808/after-m1-attempt1.raw.log)
[before-not-run.txt](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c7ba8eaca3836cd1/before-not-run.txt)
[after.normalized.log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/710b4410e8d98b3e/after.normalized.log)
[comparison-unavailable.diff](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/decb2f977427886f/comparison-unavailable.diff)


`before-not-run.txt` is an availability record, not a raw runtime log. No before runtime log can be uploaded because none exists. The driver and log contain synthetic data only. None of these artifacts was committed.

## Cleanup

Both evidence worktrees are being removed using the exact literal requested paths. All original worktree-add/install/test processes had exited when checked. No long-lived server or model processes were started. Final cleanup/disk confirmation is in the handback.

FRICTION:
- infra | after `vp install` reported 1h26m9.6s, exceeding the entire 55-minute budget; before install took 8m52.7s | Warm dependencies or reserve more installation budget; check the clock after every blocking operation.
- tool | code-mode wait timed out, then reported a stale host generation; execution results became unavailable while work proceeded | Bounded calls with separately durable PID/status capture and elapsed-time monitoring.
- model | driver subscribed before blocker A and classified all starting-status updates as promotions; all five M1 tests failed on stale A queue entries | Track queued→starting transitions, register after seeding, or explicitly exclude the active blocker before asserting delivery order.

VERDICT: INSUFFICIENT. All 17 affected rows remain unverified as paired captures.