Run 187 batch: apply mode, one test policy, rules past compaction, code review diff, and 15 more #471

Closed
opened 2026-11-09 22:36:26 -05:00 by cmoriarty · 1 comment
Owner

What happens. Run 187 (21 UX tasks on Scratch) took 417 min and never reached the lanes. apply ran all 41 tasks in one session (335 min, 10 compactions, no subagents). The briefs gave opposite test policies, rules dropped out of context at the first compaction, the code review saw none of the modified code, and a step waiting on its own background command was failed. The findings are on braid#467.

Fix. The 19 items the owner approved at the top of braid#467, as one change. Briefs must not grow: an added rule replaces text.

Done when. ./tools/test.sh passes with a test per fix, ./tools/sim.sh --smoke passes, the model-facing changes are replayed on run 187's steps, and a re-run of run 187's brief is compared with run 187.

**What happens.** Run 187 (21 UX tasks on Scratch) took 417 min and never reached the lanes. apply ran all 41 tasks in one session (335 min, 10 compactions, no subagents). The briefs gave opposite test policies, rules dropped out of context at the first compaction, the code review saw none of the modified code, and a step waiting on its own background command was failed. The findings are on braid#467. **Fix.** The 19 items the owner approved at the top of braid#467, as one change. Briefs must not grow: an added rule replaces text. **Done when.** `./tools/test.sh` passes with a test per fix, `./tools/sim.sh --smoke` passes, the model-facing changes are replayed on run 187's steps, and a re-run of run 187's brief is compared with run 187.
Author
Owner

Shipped in PR 472 (b1029f55) and archived in PR 473: OpenSpec change run187-batch, now openspec/changes/archive/2026-11-10-run187-batch/. Its 23 requirement changes are synced into openspec/specs/.

What shipped: the 19 fixes approved at the top of braid#467, with the owner's changes. auto splits a change of more than six task groups. The delegation section is tightened without growing. The compaction overshoot was left out.

Evidence from the laptop

  • ./tools/test.sh passes (690 backend, 119 vitest, 51 browser journeys) with a test for each fix, among them run 187's eight groups, its eight spec files, its waitForTimeout(50) poll, its re-seeded fullGame.test.ts and its two Depends on: 1 findings.
  • All 87 sims pass, including the new apply-auto-many-groups. The full set caught a regression in one-shot-tracked, now fixed.
  • Scripted replays of run 187's own steps, on today's code (run 187 → now):
step run 187 now
explore 21,684 21,363, no file count
round-one review 174,798 146,365, plan whole
round-two review 100,444 11,479
code review 243,959 235,235, modified code diffed tests first, re-seed named
  • No brief grew on the fixture (explore −321, the Visual check −44, apply −16).
  • A free replay of run 187's explore could not test anything: four Free models refused it in 1.1 min. The model-facing check is the re-run of run 187's brief on vLLM, starting after this deploy, compared on braid#467.
Shipped in PR 472 (`b1029f55`) and archived in PR 473: OpenSpec change `run187-batch`, now `openspec/changes/archive/2026-11-10-run187-batch/`. Its 23 requirement changes are synced into `openspec/specs/`. **What shipped:** the 19 fixes approved at the top of braid#467, with the owner's changes. `auto` splits a change of more than six task groups. The delegation section is tightened without growing. The compaction overshoot was left out. **Evidence from the laptop** - `./tools/test.sh` passes (690 backend, 119 vitest, 51 browser journeys) with a test for each fix, among them run 187's eight groups, its eight spec files, its `waitForTimeout(50)` poll, its re-seeded `fullGame.test.ts` and its two `Depends on: 1` findings. - All 87 sims pass, including the new `apply-auto-many-groups`. The full set caught a regression in `one-shot-tracked`, now fixed. - Scripted replays of run 187's own steps, on today's code (run 187 → now): | step | run 187 | now | |---|---|---| | explore | 21,684 | 21,363, no file count | | round-one review | 174,798 | 146,365, plan whole | | round-two review | 100,444 | 11,479 | | code review | 243,959 | 235,235, modified code diffed tests first, re-seed named | - No brief grew on the fixture (explore −321, the Visual check −44, apply −16). - A `free` replay of run 187's explore could not test anything: four Free models refused it in 1.1 min. The model-facing check is the re-run of run 187's brief on vLLM, starting after this deploy, compared on braid#467.
Sign in to join this conversation.
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
cmoriarty/braid#471
No description provided.