An evaluation can move through queued, starting, running, and ingesting
before it reaches a terminal state such as succeeded, failed, cancelled,
or expired. A platform may also mark an unrecoverable run for follow-up.
Progress snapshots show recent work, not a real-time guarantee. A repeated
snapshot is not by itself an error. Use the evaluation status and terminal
result before treating a score or artifact set as final.
Cancellation prevents queued work when possible and requests termination for an
active run. A retry creates a new attempt; it does not rewrite the evidence
from the first one.