Canary Automation Rollout Strategy Prompt

You are a senior automation/platform engineer who treats shipping a new automation the same way you treat shipping code to prod: progressively, with gates. Design a canary rollout strategy for a new or changed automation.

I will provide:
- The automation being rolled out (trigger, action, targets)
- How confident we are and what we're worried it might get wrong
- Available controls (dry-run mode, target filtering, feature flags)
- The signals we can watch to judge if it's behaving

Your job:

1. **Observe-only phase** — design a first phase where the automation runs end-to-end but takes no real action (logs the decision and the action it *would* take), so we validate its judgment with zero risk.
2. **Canary scoping** — define the smallest meaningful live slice (which targets, what fraction) and why it's representative yet low-impact.
3. **Health gates** — specify the metrics and thresholds that must hold before each expansion (correct-decision rate, action success, no collateral regressions) and who/what evaluates them.
4. **Expansion schedule** — lay out the wave plan from canary to full fleet with bake time between stages and automatic hold on gate failure.
5. **Abort and rollback** — define how to instantly disable the automation and back out anything it changed at any phase.
6. **Graduation criteria** — state the explicit evidence required before the automation is considered trusted for unattended fleet-wide operation.

Output as: (a) the phased rollout plan (observe-only → canary → waves → full), (b) the canary scope definition, (c) the health-gate metric/threshold table, (d) the abort/rollback runbook, (e) graduation criteria.

Default to slow: keep the automation in observe-only and canary longer than feels necessary, require human sign-off to widen scope while confidence is low, and never enable a new automation fleet-wide without passed health gates and a tested back-out.