kookr-shadow-detectionlisted

# Shadow Detection System Kookr has a shadow detection infrastructure that lets you safely validate new stuck-detection strategies. New strategies run alongside the real detector: they see the same data but their verdicts are **logged, not acted upon**. An offline report compares shadow vs. real verdicts using interval-based coverage metrics. ## Purpose The real anomaly detector (needs_input, permission_blocked, stale_agent, etc.) is stable and reliable. When adding new detection strategies (pane pattern matching, PID monitoring, HTTP push), we need to prove they work correctly before activating them. Shadow mode provides this proof: 1. **Strategy runs in shadow mode** — evaluates live data, logs verdicts to `~/.kookr/shadow-detection.jsonl` 2. **Shadow report compares** — reconstructs anomaly intervals, computes coverage/precision/latency 3. **Promotion when validated** — strategy moves from `shadow` to `active` after meeting thresholds ## Shadow Report Endpoint ```bash # JSON format (for programmatic analysis) curl -s http://localhost:4800/api/shadow-report | jq . # Text format (for human reading) curl -s "http://localhost:4800/api/shadow-report?format=text" ``` ### Response (JSON) ```json { "generatedAt": "2026-03-28T12:00:00.000Z", "observationWindow": { "startMs": 1743163200000, "endMs": 1743249600000 }, "strategies": [ { "source": "pane_semantics", "totalObservationMs": 86400000, "realIntervals": 14, "matchedIntervals": 14,