EPS Dashboard

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a workflow-fix candidate raised on task #1415 (emitting agent: clean-result-critic, hooked-unhooked-decomposition fold round 1 — surfaced as a mechanizable non-blocking follow-up in its verdict marker epm:clean-result-critique v7).

Goal

Decide whether "bit-deterministic" / "bit-determinism" joins the bit_byte_identical ban family in scripts/audit_clean_results_body_discipline.py, or is explicitly allowlisted (a code comment + test pinning it as legitimate determinism vocabulary), so future clean-result critics stop re-raising the ambiguity.

Workflow gap

Bug observed: the #1415 fold body uses "bit-deterministic" / "bit-determinism" (describing bit-identical re-forwards at causally-zero layers); the audit passes it silently, and the reviewing critic could not tell whether that is an allowlisted determinism term or an unwidened escape of the byte/bit-identical ban family.
Why it is a workflow gap: the bit_byte_identical category has been widened twice recently (issue-1423 -equal family, issue-1447 synonym tail) with no recorded decision boundary for "deterministic"; every future critic re-litigates it.
Confidence (emitter): low
verified-at-filing: functional probe of the category regex (?<!-)\b(?:byte|bit)(?:wise)?[\s-](?:identical|equal|exact)\b at compose time: "bit-deterministic" -> no match, "bit-determinism" -> no match, "bit identical" -> MATCH (2026-07-23); grep -n -i "bit.determin" scripts/audit_clean_results_body_discipline.py -> 0 hits in the target file; landed-fix history git log --oneline --since='7 days ago' -- scripts/audit_clean_results_body_discipline.py -> d1afe0730d + ab53595c54 widened the identical/equal/exact tail only, neither covers "deterministic" (not a just-landed duplicate).

Proposed change (candidate diff sketch — refine in planning)

(none — synthesized from prose follow-up; two candidate shapes for the planner:)

EITHER extend the bit_byte_identical alternation with deterministic|determinism (if the register is the same banned claim-shape),
OR add an explicit allowlist comment + a pin test asserting "bit-deterministic" is deliberately NOT flagged (legitimate determinism vocabulary describing exact re-forward reproducibility, as in #1415's jitter-floor evidence).

Scope / surfaces

Primary target: scripts/audit_clean_results_body_discipline.py
Grep the workflow surface for the pattern before editing (grep -rln 'bit_byte_identical' .claude/ CLAUDE.md scripts/ tests/) and update every hit; list them in the plan (expect the audit script + its tests + .claude/rules/clean-result-critic-lens-reference.md prose).

Constraints / invariants

Workflow-surface only — never experiment code, configs/, or tasks/.
scripts/workflow_lint.py --check-asks passes; ruff on touched files passes.
Grandfathered bodies must not be newly hard-FAILed: if the ban is widened, the #1415 body (which legitimately describes bit-identical determinism EVIDENCE, not a results claim) must either pass via context-scoping or be exempted — the planner weighs this; a widening that FAILs #1415's honest determinism evidence is wrong.
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/audit_clean_results_body_discipline.py
fingerprint: 4268862fa320

Verbatim surfaced prose (clean-result-critic, epm:clean-result-critique v7 on #1415): "whether 'bit-deterministic'/'bit-determinism' (used in Methodology + plan v11) should join or be explicitly allowlisted against the byte/bit-identical ban family in audit_clean_results_body_discipline.py (low confidence — likely legitimate determinism vocabulary; mechanizable)."