Design a feature flag rollout plan with kill-switch

Description

Ship-on-merge sounds brave until the first regression takes down prod. This designs a flag-gated rollout — cohort-by-cohort exposure, metric-based promotion, one-click kill — so you ship faster with less blast radius, and a junior engineer can kill a bad feature without a war room.

You are a release engineer designing the flag rollout for {{feature_name}} on {{product_name}}. Expected user reach: {{user_reach}}.

## Step 1 — Cohort plan
| Cohort | % traffic | Duration | Promotion criteria |
|--------|-----------|----------|-------------------|
| Internal employees | 100% internal | 48h | No P0 regressions, internal NPS ≥3/5 |
| Beta opt-in | 100% of opt-in list | 1 week | Error rate 2x baseline for 15 min → auto-rollback
- p99 latency >1.5x baseline for 30 min → auto-rollback
- Support tickets >3x normal volume → manual review
- Activation metric drops by >10% → manual review

## Step 4 — Kill-switch design
- Single flag toggle returns to pre-rollout state
- No deploy required to kill
- Tested in a dry run before first cohort

## Output
1. Cohort schedule with promotion criteria
2. Metric monitoring list
3. Kill criteria with automation vs. manual decisions
4. The one metric most likely to move before user-visible regressions — our canary-in-the-coalmine

Customize This Prompt