EPS Dashboard

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a failure-lesson raised on task #1345 round 6 (emitting agent: experiment-implementer).

Add an artifact-reuse fitness check: parent-branch unmerged fixes + realized-row-count reconciliation.

Bug observed: #1345 reused #825's extractor from MAIN; the parent's own degenerate-row crash-fix (partition_rendered/degenerate_content_turns/span-integrity gate) exists only on the unmerged issue-825 branch. The parent's realized shards embody the filter (naturalistic_s n=4724 vs corpus 5000), so the reuse passed every existing fitness check yet crashed in production at the first unfiltered degenerate row (s57).
Why it is a workflow gap: .claude/rules/artifact-reuse.md's checks (a)-(j) verify the ARTIFACT (recipe match, cells present, sha pins, fetchability, throughput) but never ask whether the reused CODE on main lags the parent branch's fixes, nor whether the realized artifact's row count reconciles with its declared input (a shortfall = filtering happened somewhere). The built-but-stranded-fixes lesson (workflow-fix-on-bug.md) names the class but no reuse-time check operationalizes it.
Confidence (emitter): high
verified-at-filing: grep -c "issue branch\|unmerged" .claude/rules/artifact-reuse.md → 0 hits (no parent-branch check exists; absence-of-guard claim — the 0-hit result IS the evidence) (2026-07-15).

artifact-reuse.md new check (k): when reusing a parent's CODE module (not
just its artifacts), git log --oneline main..origin/issue-<parent> -- <module>
— any unmerged parent-branch commits touching the module must be inspected
and ported or explicitly declared not-needed; AND reconcile the realized
artifact's row/cell count against its declared input corpus — a shortfall
means a filter exists; find it and port it.

Primary target: .claude/rules/artifact-reuse.md
Also consider one line in planner.md step 5 (the self-attested fitness list) — grep first (grep -rln 'fitness check' .claude/agents/planner.md .claude/rules/).