EPS Dashboard

Overview / Motivation

Auto-filed by the /daily 2026-07-28 problem sweep (transcript miners over the day's 55 session transcripts). Sources: group-J P3 + P7.

Close two gaps in the user-chat inline round duties (Step 9a-ter / the CLAUDE.md carve-out).

Bug observed: (a) While #1773 (a stronger SAE-label judging instrument) was being designed, three live rounds using the known-weak labeling rubric kept running toward Batch-API judge spend; the freeze happened only after Thomas asked twice. (b) On the same afternoon Thomas had to re-state 'parallel + vectorized' twice before a throughput addendum landed — the compute-character statement binds round DISPATCHES, and scope-extension addenda slipped past it.
Why it is a workflow gap: the carve-out's duties enumerate dispatch-time checks; neither instrument-supersession awareness nor addenda inheritance is named, so both defaulted to user vigilance.
Confidence (emitter): medium (inferred from interaction patterns)
verified-at-filing: grep -c 'Compute-character pre-launch statement' .claude/skills/issue/SKILL.md → 5 anchors (block exists; addenda not named) (2026-07-29 UTC).

Two short additive clauses at Step 9a-ter (+ the CLAUDE.md carve-out mirror).

Primary targets: .claude/skills/issue/SKILL.md (Step 9a-ter), CLAUDE.md (§ User-chat inline free analysis)

Workflow-surface only — never experiment code, configs/, or tasks/.
scripts/workflow_lint.py --check-asks passes; ruff on touched files passes.
The spawned session runs under a workflow_fix_target: Provenance line — recursion guard applies (it parks, never auto-routes, its own subagents' workflow-fix candidates).