EPS Dashboard

Overview / Motivation

Auto-filed from a formal workflow-fix candidate emitted by clean-result-critic (single-Claude round) on task #1768 (fold-round review, 2026-08-02).

Goal

Add a pm_inline rule to the clean-result body audit matching value ± err / bare ±<num> in reader-facing prose, reusing interval_inline's caption-blockquote + GFM-table blanking.

Workflow gap

Bug observed: median ±0.16 displacement, ±0.06 read-out sat in #1768's ## Results prose through a prior clean-result gate and its Codex twin; caught only on a fresh LM read.
Why it is a workflow gap: the audit's own comment (L892) implies value ± err coverage but no live rule matches the character — the Lens 7 sub-category has no mechanical backstop at all.
Confidence (emitter): high
verified-at-filing: grep -n $'±' scripts/audit_clean_results_body_discipline.py → 2 hits, BOTH comment lines (L311, L892); live-rule occurrences = 0 (2026-08-02, re-run by filer; matches the emitter's grep).

Proposed change (candidate diff sketch — refine in planning)

diff_sketch: |

"pm_inline": (
```
   r"\d\s*±\s*\d|±\s*\d*\.?\d+",
```

   "Inline credence interval as `value ± err` in reader-facing prose (banned)",

), reusing interval_inline's caption-blockquote + GFM-table blanking so chart annotations and parameter tables stay legal.

Scope / surfaces

Primary target: scripts/audit_clean_results_body_discipline.py (+ its test file); grep .claude/skills/clean-results/SPEC.md for the Lens 7 wording and keep consistent.

Constraints / invariants

Workflow-surface only; existing grandfathered bodies not newly hard-FAILed (WARN-first or forward-only per the audit's conventions — planner's call).
This session runs under EPM_WORKFLOW_FIX_SESSION=1 and carries a workflow_fix_target: Provenance line — recursion guard applies.

Provenance

workflow_fix_target: scripts/audit_clean_results_body_discipline.py
fingerprint: 7511c5db4555 Verbatim candidate block preserved in the emitting critic's report (task #1768 events context, 2026-08-02).