EPS Dashboard

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a prose follow-up raised on task #1901 (emitting agent: clean-result-critic).

Goal

Widen verify_task_body.py check 30's count-noun regex family to cover chunk/sidecar nouns and modifier-separated nouns so footer file-count claims phrased that way stop being invisible to the check.

Workflow gap

Bug observed: check 30 (check_hf_file_count_claims) reported "no file-count claims adjacent to HF tree links" on #1901's body while the footer carried TRUE count-claims in parentheticals following the pinned issue1901_wildchat /tree/<sha> link — the claims used the noun "chunks" ("3 activation chunks") and a modifier-separated noun ("3 chunk files plus 1 over-length-skip sidecar"), neither matched by _COUNT_NOUN_RE's files?|shards? vocabulary or its adjacent count-noun shape.
Why it is a workflow gap: the count-noun vocabulary/shape in _COUNT_NOUN_RE (and its siblings _COUNT_PAREN_LINK_RE, _HF_LINKTEXT_THEN_PAREN_RE) is narrower than the footer phrasings bodies routinely use, so count claims phrased with chunk/sidecar nouns or a modifier token between count and noun are unverifiable-by-silence — a FALSE claim in that phrasing would ship unflagged (the #1901 instance happened to be true; the critic verified it manually via list_repo_tree).
Confidence (emitter): low
verified-at-filing: grep -n "_COUNT_NOUN" scripts/verify_task_body.py + code read of lines 9439-9456 → vocabulary is exactly files?|shards? with no modifier allowance; #1901 body footer (tasks/followups_running/1901/body.md) carries "3 activation chunks" + "3 chunk files plus 1 over-length-skip sidecar" adjacent to the pinned HF tree link, and check 30's output on --issue 1901 reads "no file-count claims adjacent to HF tree links" (2026-07-31). Per-target: scripts/verify_task_body.py 6 hits for _COUNT_NOUN.

Proposed change (candidate diff sketch — refine in planning)

- r"\b(?P<count>\d{1,3}(?:,\d{3})+|\d{1,6})\s+(?P<noun>files?|shards?)\b"
+ r"\b(?P<count>\d{1,3}(?:,\d{3})+|\d{1,6})\s+(?:\w+[ \t]+)?(?P<noun>files?|shards?|chunks?|sidecars?)\b"

(applied consistently across the _COUNT_NOUN_RE regex family, with the check's documented precision guards — per-namespace lookahead, folder-inflation tolerance — reviewed for the widened vocabulary; the planner decides whether the single-modifier allowance needs a stopword guard to protect precision.)

Scope / surfaces

Primary target: scripts/verify_task_body.py
Grep the workflow surface for the pattern before editing (grep -rln '_COUNT_NOUN' .claude/ CLAUDE.md scripts/) and update every hit; list them in the plan. Update tests/test_verify_task_body.py pins for the widened shapes (a "3 activation chunks" positive + a modifier-separated positive + a precision-guard negative).

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 any of its own subagents' workflow-fix candidates (recursion guard).

Provenance

workflow_fix_target: scripts/verify_task_body.py
fingerprint: 04e69aa91f07

Surfaced prose (verbatim, clean-result-critic on #1901): "scripts/verify_task_body.py's 'HF file-count claims match the Hub tree' check reported 'no file-count claims adjacent to HF tree links' while the footer carries exactly such claims in parentheticals following the pinned issue1901_wildchat tree link ('3 activation chunks', '3 chunk files plus 1 over-length-skip sidecar'). The claims are TRUE (I verified via list_repo_tree), so no body defect — but the check's adjacency window apparently misses count-claims phrased inside a parenthetical after the link, a phrasing footers use routinely. Concrete change: widen the check's claim-binding window to include the sentence/parenthetical span immediately following a pinned /tree/<sha> link. confidence: low" — orchestrator triage corrected the mechanism: the adjacency/parenthetical binding exists (#1005/#833 shapes); the actual gap is the count-NOUN vocabulary + modifier separation.