Answers and explanations · Rehearse a change and its recovery path

1. Shadow/w1 continues; canary/w2 continues; expand/w3 returns RECOVER. Its critical failure, 0.92 good fraction, 900 ms supplied p95, and six excess bad requests all support stopping under this toy policy. Earlier success does not cancel a later stop condition. The target is fictional-router-v1-digest-a1; no real recovery occurs.
2. Recovery-not-ready returns HOLD with no stage events: a prerequisite failed. Missing fictional review holds at shadow/w1 after reaching a stage with otherwise healthy observations. Both remain simulations, and neither authorizes a live action.

import copy
import opslab
s = opslab.load()
p = copy.deepcopy(s["plan"])
p["rollback_ready"] = False
print(opslab.rollout(p, s["windows"]))

3. Reusing w1 raises InputError because each stage needs a fresh window. The system would otherwise be pretending that one observation justifies several distinct exposure decisions. Fresh evidence does not necessarily mean sufficient evidence, but identical evidence should not masquerade as a new check.
4. Example checklist: fictional owner confirms the baseline digest is available; checks configuration/data compatibility; scopes the paper switch; verifies task and operational signals after recovery; reviews actions already completed. A refund already sent or an email already delivered is not reversed by changing a prompt. Those would need separately authorized remediation in an actual system.

Self-check: a paper COMPLETE_SIMULATION is not a production canary, a pilot, or permission. Keep absolute conditions even when control-relative differences look small.