EPS
← All tasks·#1539Completed

workflow-fix: never-run claims scan live pods+markers

kind: infra#wf-fix#wf-fix-fp:244c82ba276e#daily-auto-filed
track:

Highlight text on any card (body, plan, or an event) to anchor a comment, or leave a whole-task comment here. Mention @claude to summon a reply.

No comments yet.

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from the 2026-07-18 /daily transcript problem sweep (c4-P5). Route-2 filing.

Goal

Extend the CLAUDE.md "Verify before asserting" bullet so a never-run / nothing-live-generating claim additionally scans LIVE compute — running pods (pod.py list-ephemeral), GCE instances, and recent epm:run-launched / epm:followup-scope / epm:free-analysis-followup-run markers on PARKED tasks — before asserting nothing is producing a cell.

Workflow gap

  • Bug observed: a chat claim that nothing live was generating a cell was read from a task-status-folder scan, while #1345's inline onpolicy_assistant_story round (a PARKED task, signalled only pod-side) had already generated 2,019 stories with a character named Assistant that night; Thomas caught it.
  • Why it is a workflow gap: the "Verify before asserting" bullet mandates an eval_results/ artifact check for a number / winner / never-run claim, but says nothing about scanning LIVE generation (pods, GCE, parked-task run markers) — a never-run/nothing-live claim can be false while every task-status folder looks idle.
  • Confidence (emitter): medium
  • verified-at-filing: grep -n 'never-run\|nothing-live\|live pods\|list-ephemeral' CLAUDE.md → the "Verify before asserting" bullet (line ~185) scopes verification to eval_results/ JSON / named HF/WandB artifacts only; the only list-ephemeral / GCE hits (lines 311, 328) are the pod-CLI doc + the GCE janitor cron, NOT the verify-bullet — so no live-generation-scan duty exists there (2026-07-19). (.claude/agents/experiment-status.md lives under ~/.claude/ — user-global, out of the project workflow surface — so it is NOT a target here.)

Proposed change (candidate diff sketch — refine in planning)

# In CLAUDE.md "Ad-hoc results summaries ... Verify before asserting:" bullet,
# add after the eval_results artifact-check sentence:
+ A never-run / nothing-live-generating claim ALSO scans live compute before
+ asserting: `pod.py list-ephemeral`, the GCE instance list, and recent
+ epm:run-launched / epm:followup-scope / epm:free-analysis-followup-run
+ markers on PARKED tasks (an inline pod round on a parked task signals
+ pod-side only and leaves every task-status folder idle-looking; #1345).

Scope / surfaces

  • Primary target: CLAUDE.md
  • Single bullet edit; keep it one sentence appended to the existing Verify-before-asserting clause.

Constraints / invariants

  • Workflow-surface only. Do not add a mechanical gate — this is an assertion- discipline rule, same register as the surrounding bullet.
  • scripts/workflow_lint.py --check-asks passes.
  • Recursion guard applies (workflow_fix_target: Provenance line).

Provenance

  • workflow_fix_target: CLAUDE.md
  • fingerprint: ecd694722516

Surfaced problem (c4-P5): a "nothing live is generating that cell" chat claim was refuted by #1345's parked-task inline pod round (2,019 stories generated, character named Assistant); the user had to correct it.

Activity