You are a frontier consult model for an infrastructure agent...

Prompt

You are a frontier consult model for an infrastructure agent fleet. Answer in two parts. In Part 1 state your explicit assumptions, produce an ordered plan with verification gates and rollback boundaries, and end with a verdict line (ANSWER / CONDITIONAL / HOLD). Part 2 reveals new facts: revise your plan, explicitly retract anything invalidated, and re-issue your verdict. PART 1 SCENARIO: You advise an infrastructure agent migrating unprivileged LXC container ct-214 from oak-a to oak-b. Both run LXC 6.0. The container contains a worker and PostgreSQL 16; its root filesystem is a 24-GiB ZFS dataset. There is no shared storage. A verified root-dataset snapshot has already been received on oak-b. Final root-dataset delta is estimated at 1.8 GiB; measured transfer throughput is 60 MiB/s. The outage allowance is 180 seconds. The service uses 10.40.8.14, which must never be active on both hosts. Root-dataset ownership and UID mappings match. Source retention for rollback is permitted for 48 hours. Recommend an ordered migration with downtime estimates, gates, and explicit rollback boundaries. Separate known facts from assumptions. PART 2 TWIST (revise): Inspection reveals /var/lib/postgresql is a bind mount from source dataset tank/pg214, not part of the received snapshot. It holds 320 GiB. No destination copy exists. The maintenance window starts in ten minutes; its 180-second outage limit is unchanged. Revise your recommendation and explicitly retract anything invalidated.