Autonomous
Identity URL loop: same thread keeps working staging and ratings. Use Live habitat to watch the mutable surface; use Review links for frozen snapshots after a successful stage.
Working staging vs review snapshots (architecture)
Working staging is the mutable surface per thread. Planner/generator/evaluator see working_staging_facts from this live state. Live habitat reflects it in real time.
A staging snapshot is a frozen row created when the graph stages after evaluator pass/partial. It is for human review; it does not replace working staging. Approving or rating a snapshot informs future runs via signals — it does not "paste" that snapshot over the habitat. The canonical flow is: iterations mutate working staging → optional snapshot for review → publication/canon separately.
Server cron autonomous loops (/orchestrator/loops) use the same staging model; this page uses POST /api/runs/start in a client loop.
Identity URL
Planner instructions
Sent once per run as user_instructions, then cleared from the queue.
Run log
No iterations yet.