Interactive graph
Staged Rollout
Table of Contents
Ship to a small slice first so rollback stays cheap if the new state bleeds.
Key Concept
- Full-blast deploy turns every user into a test subject.
- Level 1: If you change everything for everyone at once, one bug hits every person the same day.
- Level 2: A full blast deploy turns your whole user base into testers - harm spreads before anyone can turn back.
- Level 3: Staged rollout limits blast radius because a small group catches failures before the rest of the lane commits.
- Level 4: Ship to a preview branch, one table section, or one scrimmage night before you commit the whole lane.
- Level 5: Staged rollout keeps Reversibility real and Safety by Design at release time - harm hits fewer people before you commit the whole lane.
- Staged rollout - preview branch, one table section, scrimmage Tuesday - keeps Reversibility real.
- Safety by Design at release time: harm hits fewer people before you commit the whole lane.
Examples
- Cloudflare preview branch caught the broken layout before main - rollback was not touching production.
- New inbound play ran in scrimmage Tuesday before Friday's game - point guard reverted to last week's call at halftime without reinventing.
Note Relationships
| Relationship | Wikilink | Reason |
|---|---|---|
| alternative | Practice Small Experiments | Controlled slice when micro-trials are too small for the risk |
| alternative | Preparedness | Rehearse on a slice vs rehearse the undo drill on paper |
| contradicts | Ship It | when blasting to everyone beats waiting for the small-slice signal |
| extends | Reversibility | Small blast radius makes undo cheap |
| extends | Risk Management | Limit exposure before you bet the whole lane |
| extends | Rollback Principle | Staged slice is where you practice undo before everyone's affected |
| extends | Safety by Design | Fewer people hit before the hazard is proven safe |
| implements | Free Tier Hosting Stack | PR preview deploy before Pages merges to main |
