EPS Dashboard

Overview / Motivation

Auto-filed by the /daily 2026-07-15 problem sweep (route 2 — behavior/logic change, independent review required).

Goal

When a clean-result body or figure cites a sibling issue's headline measured under a different eval protocol, the protocol delta is stated inline next to the number.

Workflow gap

Bug observed: mentor-facing R² headlines across #779/#823 were not protocol-comparable (single-split vs k-fold, different layer-selection rules); Thomas needed ~6 clarifying questions to draft a one-line reply, and the strictly comparable cell "doesn't exist yet" (b7150177 21:04-22:07Z question chain).
Why it is a workflow gap: SPEC.md governs how a body cites its OWN numbers (what-is-plotted-exactly, footer provenance) but has no rule for cross-issue headline citations — so two sibling headlines sit side by side in mentor-facing prose with no signal that their protocols differ.
Severity: medium
verified-at-filing: grep -n 'protocol delta\|protocol-comparable\|sibling.*headline\|headline.*sibling' .claude/skills/clean-results/SPEC.md → 0 hits — proposed rule absent (2026-07-16 UTC).

Proposed change (refine in planning)

Add to .claude/skills/clean-results/SPEC.md (the ## Goal **This experiment in context:** guidance and/or the Results three-beat spec): when a body or figure cites a sibling issue's headline number measured under a DIFFERENT eval protocol (split scheme, fold structure, layer-selection rule, eval distribution), the protocol delta is stated inline next to the number (e.g. "#823 R²=0.63 (k-fold, predictivity-selected layer) vs this issue's 0.71 (single split, steering layer) — not directly comparable"). Consider a companion clean-result-critic lens note (Lens 7 statistical-framing) so the critics enforce it.

Scope / surfaces

Primary target: .claude/skills/clean-results/SPEC.md
Secondary: .claude/agents/clean-result-critic.md / .claude/rules/clean-result-critic-lens-reference.md (statistical-framing lens enforcement, if trivially co-editable)

Constraints / invariants

Workflow-surface only — never experiment code, configs/, or tasks/.
ALWAYS read SPEC.md before changing anything about the report structure; keep CLAUDE.md summary / verify_task_body.py / critic lenses in sync per the standing SPEC.md sync rule.
scripts/workflow_lint.py --check-asks passes; ruff on touched files passes.
This session runs under EPM_WORKFLOW_FIX_SESSION=1 and carries a workflow_fix_target: Provenance line — it MUST NOT auto-route its own subagents' workflow-fix candidates (recursion guard).

Provenance

fingerprint: 4268e2012483
workflow_fix_target: .claude/skills/clean-results/SPEC.md

Mined from 2026-07-15 session transcripts by the /daily problem sweep. Evidence: b7150177 21:04-22:07Z question chain (batch 01 P1).