workflow-fix: pre_reg audit — verdict-lattice head nouns + range tokens escape

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a formal <!-- workflow-fix-candidate v1 --> block raised by the clean-result-critic on task #1090 (round-6 fold review, epm:clean-result-critique v10).

Goal

Extend the pre_reg audit branch's bare-registered <noun> head-noun alternation (cuts?/paths?/clauses?/controls?/levers?/bars?/smokes?) and widen its intervening-token regex to admit multi-dot numeric range tokens (e.g. 0.60-0.85), re-measured against the promoted-body corpus per the #1419 method.

Workflow gap

• Bug observed: #1090's folded body carried ~12 bare registered-noun forms ("the registered 0.60-0.85 band", "registered kill path", "registered per-arm abort clause", "registered 0.30 cut", "registered install-strength control", "registered unrun lever", "registered 10% kill bar") that all escape the #1419 pre_reg branch — caught only by the LM critic (Lens 7).
• Why it is a workflow gap: the #1419 head-noun list was measured against the 2026-07-16 corpus; the verdict-lattice vocabulary now spreading through the factory line (cut / kill path / abort clause / kill bar) is the same Lens-7 jargon class and currently rides only on the LM critic — the exact escape mode #1419's own comment documents. Range tokens like 0.60-0.85 additionally fail the intervening-token class [\w%/<>=≤≥−+-]+(?:\.\d+)*, so "registered 0.60-0.85 band" never reaches the listed head noun "band".
• Confidence (emitter): medium
• verified-at-filing: sed -n '<pre_reg alternation lines>' scripts/audit_clean_results_body_discipline.py → the alternation at the verdicts?|lattices?|… lines contains none of cuts/paths/clauses/controls/levers/bars/smokes (presence-of-branch + absence-of-nouns both confirmed at source); git log --oneline --since='7 days ago' -- scripts/audit_clean_results_body_discipline.py → #1419 (9ef47afcfc) is the latest pre_reg touch, no later fix covers this (2026-07-17)

Proposed change (candidate diff sketch — refine in planning)

•    r"(?:[\w%/<>=≤≥−+-]+(?:\.\d+)*[ \t]+){0,3}?"
  

•    r"(?:[\w%/<>=≤≥−+-]+(?:[.\-−]\d+)*[ \t]+){0,3}?"
  

•    r"(?:verdicts?|lattices?|margins?|reads?|criteri(?:on|a)|thresholds?|bands?"
  

•    r"|gates?|rules?|endpoints?|contrasts?|floors?|companions?|hypothes[ei]s|alpha)\b",
  

•    r"(?:verdicts?|lattices?|margins?|reads?|criteri(?:on|a)|thresholds?|bands?"
  

•    r"|gates?|rules?|endpoints?|contrasts?|floors?|companions?|hypothes[ei]s|alpha"
  

•    r"|cuts?|paths?|clauses?|controls?|levers?|bars?|smokes?)\b",
  

("test"/"interval" stay deliberately absent per the #1419 measurement method; re-measure the false-positive rate against the promoted-body corpus before landing.)

Scope / surfaces

• Primary target: scripts/audit_clean_results_body_discipline.py
• Grep the workflow surface for the pattern before editing (grep -rln 'registered <noun>\|pre_reg' .claude/ CLAUDE.md scripts/) and update every hit; check the audit's test file for pinned pattern expectations.

Constraints / invariants

• Workflow-surface only. workflow_lint.py --check-asks passes; ruff passes; the audit's pinned tests updated alongside; false-positive re-measure against the promoted corpus recorded in the plan.
• This session runs under EPM_WORKFLOW_FIX_SESSION=1 / carries a workflow_fix_target: Provenance line — recursion guard applies.

Provenance

• workflow_fix_target: scripts/audit_clean_results_body_discipline.py
• fingerprint: 38a0acee2c04

(Verbatim candidate block preserved in the origin_prompt; raised by clean-result-critic, task #1090 round-6 fold review.)