EPS Dashboard

Overview / Motivation

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

Goal

Strengthen /paper-plots' interim/user-facing figure defaults: no hatching / per-bar n annotations / estimated-value markers unless asked; CV figures label fold structure explicitly; interim figures state the mapping arm and per-behavior disaggregation in the caption/setup line.

Workflow gap

Bug observed: a cluster of figure corrections in one day: 5 separate user fixes in 09f28ede 05:12-08:15 (hatching, per-bar n labels, "guess" markers, unexplained negative y-axis, unwanted rotation arm); "n=4998" misread as train size because a CV figure's caption omitted fold structure (b7150177 22:06); interim figures shipped without stating the mapping arm or per-behavior disaggregation (28d0874a 06:36-06:44).
Why it is a workflow gap: the skill's no-editorial-annotations rule covers arrows/effect-size labels but not hatching / per-bar n / estimated-value markers, and it has no CV-caption or arm-disclosure defaults — so each interim figure relearns Thomas's preferences via correction rounds.
Severity: medium
verified-at-filing: grep -n 'hatch\|interim\|fold' .claude/skills/paper-plots/SKILL.md → 0 relevant hits (no hatching / interim-figure / fold-structure guidance; "annotation" hits at L166-198 cover the existing no-arrows/no-effect-size-labels rule only) — proposed defaults absent; plain-English label rule present at L121 (adjacent, not covering these cases) (2026-07-16 UTC).

Proposed change (refine in planning)

Extend .claude/skills/paper-plots/SKILL.md's figure-hygiene section (anchor: the "Figures present DATA" block at L166): (a) interim/user-facing default = no hatching, no per-bar n annotations, no estimated/"guess" value markers, no gratuitous label rotation unless asked; (b) any cross-validated metric figure labels its fold structure explicitly in the caption (n_contexts, k folds, held-out size) so aggregate n is never misread as train size; (c) interim figures state the mapping arm (prefix vs context) + per-behavior disaggregation in the caption/setup line — the figure-side mirror of the CLAUDE.md ad-hoc-summaries provenance rule.

Scope / surfaces

Primary target: .claude/skills/paper-plots/SKILL.md (anchor: L166 figure-hygiene block; L121 label rule)

Constraints / invariants

Workflow-surface only — never experiment code, configs/, or tasks/.
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: a8a55c08f9ed
workflow_fix_target: .claude/skills/paper-plots/SKILL.md

Mined from 2026-07-15 session transcripts by the /daily problem sweep. Evidence: 09f28ede 05:12-08:15 (batch 08 P5); b7150177 22:06 (batch 01 P3); 28d0874a 06:36-06:44 (batch 04 P5).