EPS Dashboard

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a workflow-fix candidate raised on task #1315 (emitting agent: experiment-implementer, crash-fix round 4).

Goal

Extend experiment-implementer.md item 5's PASS_UNIFIED definition: any smoke-scale slicing of a non-cell axis must satisfy every downstream phase's minimum-N asserts (grep consumers for assert len(...) >= shapes; name the floor per sliced axis in the attestation note).

Workflow gap

Bug observed: A PASS_UNIFIED smoke (cell subset threaded through every phase) still crashed by construction at its LAST phase because a NON-cell smoke axis (question slice) was sized below a downstream consumer's min-N assert (#1315 r4: questions[:1] vs split_half_self_cosine's len(qs)>=2).
Why it is a workflow gap: The § pre-flight item-5 PASS_UNIFIED definition requires per-phase CELL-subset threading but is silent on non-cell smoke-scale slices (questions/rows/steps) vs downstream min-N asserts, so an un-passable-by-construction smoke can still attest PASS_UNIFIED.
Confidence (emitter): medium
verified-at-filing: grep -c "PASS_UNIFIED" .claude/agents/experiment-implementer.md → 5 hits (presence of the item-5 attestation site confirmed); grep -ciE "non-cell|min-N|minimum-N" .claude/agents/experiment-implementer.md → 0 hits (absence-of-guard claim — the 0-hit in-target result IS the evidence) (2026-07-15)

Proposed change (candidate diff sketch — refine in planning)

In experiment-implementer.md § Before writing code, item 5, after "Per-phase subset threading is part of the PASS_UNIFIED definition":

The same duty covers NON-cell smoke axes: for every axis the smoke slices
(questions, rows, steps, draws), verify the sliced size satisfies each
downstream phase's minimum-N asserts and name the floor per axis in the
attestation note; a sliced axis below a downstream min-N is un-passable
by construction — FAIL_NO_CANARY (#1315 r4).

Scope / surfaces

Primary target: .claude/agents/experiment-implementer.md
Grep the workflow surface for the pattern before editing (grep -rln 'PASS_UNIFIED' .claude/ CLAUDE.md scripts/) and update every hit that defines (not merely consumes) the attestation; list them in the plan.

Constraints / invariants

Workflow-surface only — never experiment code, configs/, or tasks/.
scripts/workflow_lint.py --check-asks passes; ruff on touched files passes; if workflow.yaml or CLAUDE.md change, they stay consistent with the rule file.
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: .claude/agents/experiment-implementer.md
fingerprint: 55ea303235bb

target_file: .claude/agents/experiment-implementer.md bug_observed: A PASS_UNIFIED smoke (cell subset threaded through every phase) still crashed by construction at its LAST phase because a NON-cell smoke axis (question slice) was sized below a downstream consumer's min-N assert (#1315 r4: questions[:1] vs split_half_self_cosine's len(qs)>=2). why_workflow_gap: The § pre-flight item-5 PASS_UNIFIED definition requires per-phase CELL-subset threading but is silent on non-cell smoke-scale slices (questions/rows/steps) vs downstream min-N asserts, so an un-passable-by-construction smoke can still attest PASS_UNIFIED. proposed_change: Extend item 5's PASS_UNIFIED definition: any smoke-scale slicing of a non-cell axis must satisfy every downstream phase's minimum-N asserts (grep consumers for assert len(...) >= shapes; name the floor per sliced axis in the attestation note). diff_sketch: | In experiment-implementer.md § Before writing code, item 5, after "Per-phase subset threading is part of the PASS_UNIFIED definition":

The same duty covers NON-cell smoke axes: for every axis the smoke slices
(questions, rows, steps, draws), verify the sliced size satisfies each
downstream phase's minimum-N asserts and name the floor per axis in the
attestation note; a sliced axis below a downstream min-N is un-passable
by construction — FAIL_NO_CANARY (#1315 r4). confidence: medium related_task: #1315