EPS
← All tasks·#1082Completed

daily-fix: task-corpus search recipe - children, follow-up l

kind: infra#wf-fix#wf-fix-fp:5ad6c03b2b21#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 /daily 2026-07-05 (route 2: behavior/logic change -> independent review), from the nightly transcript problem sweep.

Goal

Add a search recipe to research-pm.md (and a vocabulary alias table to the glossary doc): before any 'not run / no experiment exists' claim, (1) enumerate CHILDREN of candidate parents via REGISTRY parent lineage and grep their bodies + events.jsonl; (2) sweep follow-up round labels (epm:followup-scope notes + eval_results/issue_*// dirs); (3) never treat a crashed sub-search inside a compound command as 'no hits' - rerun it; (4) grep alias variants (per-example map / question-averaged map / single-context map / query-averaged) alongside the standardized prefix-map/context-map terms.

Workflow gap

  • Bug observed: Two separate 'did we run X?' recall failures today, each needing 2-3 user pushbacks before the right answer: (a) the assistant twice reported a next-token activation-mapping experiment unrun before finding #922 (child of #841) - a REGISTRY-title one-liner had crashed (AttributeError: int has no attribute get) and printed nothing, silently degrading the search, and #922's evidence hides behind parent #841's 'affine layer map' label (session eaec45a8); (b) the assistant asserted 'no experiment fits both maps on the same matched grid' through two pushbacks before finding #813's same-day follow-up round 'per-example-vs-averaged-map' - follow-up rounds do not surface in title/body greps, and #813's vocabulary ('per-example' vs 'question-averaged') missed the glossary-term grep net (session 3a3ac9db).
  • Why it is a workflow gap: the failure originates in the workflow surface / shared helper named below, not in any one experiment.
  • Confidence (emitter): medium

Scope / surfaces

  • Primary target: .claude/agents/research-pm.md, docs/glossary_context_answer_map.md
  • Negative-existence claims are load-bearing for what gets filed next; both incidents risked duplicate experiments.

Constraints / invariants

  • scripts/workflow_lint.py --check-asks + --check-references stay green; ruff clean on touched files; relevant tests pass.

Provenance

  • workflow_fix_target: .claude/agents/research-pm.md, docs/glossary_context_answer_map.md
  • source: /daily 2026-07-05 problem sweep (transcript-mined)
Activity