Daily outcome reports mixed executive summary and operational detail in a flat structure, making it hard to scan. The conversation also surfaced a gap: no guidance for classifying whether a workflow is genuinely stuck vs. its outcome model simply not yet mature.

outcome-collector.md — report format

Restructured into two layers:

Executive section (always visible):

• One-sentence Executive read (quality signal, concentration, stuck/underdefined flags)
• Per-workflow visual table — no numeric counts; the bar communicates volume

| Workflow | Status | Lifecycle health |
|---|---|---|
| Chaos PR Bundle Fuzzer | <span style="white-space: nowrap;">🟨🟨🟨🟨🟨</span> | 🔴 stuck |
| PR Description Updater  | <span style="white-space: nowrap;">🟩</span>           | 🟢 resolving |

• Legend immediately below the table
• Rows sorted by management attention (most pending → most unknown → resolved-only last)

Details section (<details>): all scorecard metrics, numeric breakdowns, per-workflow counts, evidence quality, trend signals, and action items

Added lifecycle health classification table with cache-memory-based decision rules so the agent can assign the correct state each cycle.

Also clarified required inputs for status-bar rendering by explicitly documenting /tmp/gh-aw/outcome-evaluations.jsonl as the source of per-item outcome_status.

Scope cleanup from review feedback

• Reverted .github/workflows/scout.lock.yml to remove unrelated changes from this PR.
• Removed .github/copilot-instructions.md per review request.