EPS Dashboard

Overview / Motivation

Auto-filed by the /daily 2026-07-28 problem sweep (transcript miners over the day's 55 session transcripts). Source: group-I P2.

Probe arm-gating real-corpus structural assumptions in the smoke slice so their fail-loud asserts fire before production.

Bug observed: #1768's production run crashed mid-flight on plan assumption 4 (n_distinct_prefix==1), which is false on the real corpus (n=2) — ~55 min lost, hot-fixed in-session. unverified hypothesis — verify at plan time: the exact assert location + the claim that a 1-row manifest read at smoke time exercises it (read from the transcript narrative + the session's hot-fix marker v18, not re-run).
Why it is a workflow gap: §-assumptions entries are verified at plan time by the fact-checker where checkable, but REAL-corpus structural claims are only checkable against the corpus — the smoke slice is the first place that data exists, and no duty routes assumption probes there.
Confidence (emitter): medium
verified-at-filing: mechanism labeled above; verify at plan time markers included (the planner spec's §-assumptions section is the named target — re-read it before wording the duty).

One §-assumptions clause in planner.md (+ possibly a smoke-checklist mirror in the implementer spec).

Workflow-surface only — never experiment code, configs/, or tasks/.
scripts/workflow_lint.py --check-asks passes; ruff on touched files passes.
The spawned session runs under a workflow_fix_target: Provenance line — recursion guard applies (it parks, never auto-routes, its own subagents' workflow-fix candidates).