EPS Dashboard

Overview / Motivation

Auto-filed by /daily 2026-08-01 (route 2: behavior/logic change → independent review) from consolidated problem sweep entry C12 (miner 7, P5; sessions 4a1a7d1f (#1917), 6dbf040f (#1895 sibling)).

Goal

Decouple test_corpus_replay_all_historical_markers from stale-branch parser vintage: replay through the MAIN-resident parser (or pin the corpus at the branch's merge-base snapshot), OR fold the parser+test pair into the Step 5a spec-freshness sync family so a live fleet marker cannot red a branch forked before a parser fix.

Workflow gap

Bug observed: #1917's gate round 1 red: FAILED tests/test_workflow_followup_labels.py::test_corpus_replay_all_historical_markers — "Left contains one more item: (1900, '2026-08-01T03:30:56Z')". The branch forked at 9cc4a47487, BEFORE main's #1984 parser fix 0aad8f32c7 ("widen leading version-stamp stripper (dash-led form)"); a #1900 dash-led marker in the live tasks/ corpus failed the stale branch's parser while root-main passed. Cost: one ~30-min gate round + diagnosis + merge origin/main + a second ~40-min gate round. Same-family same-morning: #1895's test_no_new_torch_before_dotenv_vm_entrypoints red until main-side 322c519ab7 synced.
Why it is a workflow gap: The test replays LIVE fleet task state (_corpus_events_by_task() walks tasks/) through the BRANCH-resident parse_followup_note_field, coupling every stale branch to fleet-wide marker churn — any branch forked before a parser fix goes red the moment a new-form marker lands anywhere in the fleet, with no code defect on the branch.
Confidence: high (mechanism read directly from the test source + verified commits; incident chain miner-read from the transcript)
verified-at-filing: grep -rn 'test_corpus_replay_all_historical_markers' tests/ → 1 hit, tests/test_workflow_followup_labels.py:1206 (target CONFIRMED — the consolidated entry's named file is the real site; the test body reads _corpus_events_by_task() over live tasks/ and calls the branch-local parse_followup_note_field, construction site confirmed per clause (g)); git rev-parse → 9cc4a47487, 0aad8f32c7, 322c519ab7 all resolve; git log --oneline --since='7 days ago' -- tests/test_workflow_followup_labels.py → 0aad8f32c7 (the stripper widen — fixes the PARSER, not the stale-branch coupling), 6eb0866aa2; Step 5a family-atomic sync block confirmed at .claude/skills/issue/SKILL.md ~line 2451 (2026-08-02).

Proposed change (refine in planning)

Two candidate designs, same intent — planner picks:

(a) Test-side: pin the replayed corpus at the branch's merge-base with origin/main (or import the parser from the fetched origin/main tree for this test only), so branch runs validate the branch-vintage corpus/parser pair; or
(b) Sync-side: add tests/test_workflow_followup_labels.py + the parser module (src/explore_persona_space/task_workflow.py followup-label functions, or wherever parse_followup_note_field lives) to the Step 5a family-atomic spec-freshness list, so the pre-gate sync freshens the pair from origin/main (the #1742-class mechanism already runs before the selector). Option (b) is likely smaller and matches the existing family-atomicity machinery (test_step10d_family_atomicity_matches_step5a).

Scope / surfaces

Primary target: tests/test_workflow_followup_labels.py, .claude/skills/issue/SKILL.md
If (b): update the Step 5a FAMILY_OF map + its drift-guard pin tests atomically (the de974d95ce precedent). If (a): keep the KNOWN_MALFORMED_RUN_MARKERS allowlist semantics intact.

Constraints / invariants

Workflow-surface only — never experiment code, configs/, or tasks/.
scripts/workflow_lint.py --check-asks passes; ruff/bash -n on touched files passes.
Recursion guard: this task's session carries the workflow_fix_target Provenance line and MUST NOT auto-route its own subagents' workflow-fix candidates.

Provenance

fingerprint: d1e4befaaa2d
workflow_fix_target: tests/test_workflow_followup_labels.py, .claude/skills/issue/SKILL.md
origin: /daily 2026-08-01 problem sweep, CONSOLIDATED.md entry C12.