Time-travel debugging: re-run a job's code against its recorded journal.

case Belay.Replay.dry_run(MyBelay, job_id) do
  {:ok, result, trace} -> ...      # code path fully covered by the journal
  {:blocked, why, trace} -> ...    # hit something the recording never saw
  {:raised, error, trace} -> ...   # the code itself raised during replay
end

In a dry run every memoized read (steps, sleeps, spawns) returns its recorded value, emit/2 and debit/3 are inert, and nothing else touches the world. That makes two workflows possible:

• Post-mortem — replay a failed job locally and step through the exact decisions it made, with the exact intermediate values it saw.
• Divergence detection — after changing worker code, replay recorded jobs: {:blocked, {:missing_step, name}, _} pinpoints where the new code's path departs from what production actually executed.

Dry runs never ack, never consume attempts, and are safe against live jobs.