megaplan-observe

Debug a stalled megaplan run by diagnosing in-flight state and event timelines to determine liveness, detect recoverable transitions, and flag rubric or profile mismatches. This eliminates guesswork when runs get stuck or drift from their intended plan.

Live event tracing uses `megaplan trace --follow` to watch system execution in real time during an active workflow state. It produces a structured view of events to show exactly what the system is doing and why.

The best way to introspect an active plan for fast triage is producing a single structured JSON payload via `megaplan introspect --plan X`. This yields an actionable, structured view of in-flight state and events for immediate diagnosis.

Yes, you can detect rubric or binary profile drift during LLM operations by interpreting `rubric_doc.drift` via `megaplan doctor --plan X`. This validates flags and states to flag mismatches and provide remediation hints.

Megaplan execution gets blocked when liveness fails or transitions stall, but you can recover it by interpreting `block_details.recoverable_via` using `megaplan introspect`. This identifies failure modes and guides recovery actions.