Make big changes safely
Migrations, redesigns and rewrites should not be all or nothing. Release to a small slice, compare it with the rest and widen only when the data says it is fine.

Where safe rollouts pay off
Database and API migrations
Shadow read, then split reads, then writes, with error and latency checks at each step.
Redesigns
Launch a new navigation to 5% and compare engagement before everyone gets it.
New pricing engines
Run new calculations in parallel and compare outputs before billing anyone.
Third party swaps
Move payment or email providers gradually, with a one click return path.
Mobile releases
Gate new features by app version, so a bad build can be disabled without a store review.
Regional launches
Open a feature market by market and keep the ones you are unsure about behind approval.
Guardrails make the decision for you
You pick the metrics and the tolerances before the release. Kerf holds the line during it. If errors climb or retention dips, the rollout pauses or reverts automatically, and the on call person gets the numbers, not a mystery.
- Pause, notify or auto rollback
- System and business metrics together
- Complete audit trail of every step

“A guardrail caught a 4% retention drop on a launch everyone loved. We rolled it back in one click, before most users ever saw it.”
A rollout checklist
- 01
Flag it
Wrap the change and keep the old path working.
- 02
Pick guardrails
Choose two system metrics and one business metric, with tolerances.
- 03
Ramp
Staff, 1%, 10%, 50%, 100%, with a hold at each step.
- 04
Remove the flag
After a week at 100%, delete the old path. Kerf opens the pull request.
