Autonomy Scorecard

Can it operate itself? — scored honestly

Every node can make blocks. Few systems remember, explain themselves, and recover. This is TSUNAGI's real autonomy status — ✅ proven · 🟡 partial/controlled · ⬜ not yet. Updated only when real evidence exists.

Observe — KAGAMI

capability	impl	tested	prod-proven
Status feed + dashboards (/status, /live)	✅	✅	✅
Acceptance tracker (internal vs Koios)	✅	✅	✅
KES expiry predictive alert	✅	✅	🟡
L2 pollwatch detector	✅	✅	🟡
Operational memory (history.json)	✅	✅	✅

Protect — YAMORI

capability	impl	tested	prod-proven
In-node forge gates (era/sigma/nonce/parent/opcert/self-verify)	✅	✅	✅
Guardian L1 crash detection	✅	✅	✅
Crash-loop breaker	✅	✅	⬜

Recover — TATE

capability	impl	tested	prod-proven
Evidence capture (incident package)	✅	✅	🟡
Auto-restart (3.8s in rehearsal)	✅	✅	🟡
Rollback recommendation	✅	✅	⬜
Verified rollback chain	✅	✅	✅

Learn — KINTSUGI

capability	impl	tested	prod-proven
Incident catalogue (INC-001…013)	✅	✅	✅
signatures.json knowledge model + graph	✅	✅	✅
Operational replay (incident → full record)	✅	✅	✅
Auto incident generation + classification	✅	✅	🟡

The 🟡s on Recover/Learn share one cause: proven by a controlled recovery rehearsal (3.8s, no human) — and the recovered node then forged real canonical blocks — but not yet by a spontaneous production failure. That single event is the next milestone; when it lands, the 🟡s flip to ✅. See incidents · KINTSUGI · live.