Skip to content
v4.2 Plan a rollout from 1% to everyone across days, with automatic pauses when error rates rise. Read the release
Plinth
Start free
experiments rollouts

Choosing a guardrail metric

The metric that stops a rollout matters as much as the one you hope will improve.

P

Plinth team

1 min read

A goal metric tells you whether a change helped. A guardrail tells you whether it hurt something you care about more.

Good guardrails are

  • Fast: they move within minutes, such as error rate or latency
  • Hard to argue with: nobody wants more failed payments
  • Few: two or three, or every rollout will pause on noise

Start here

Error rate and p95 latency for the service the change touches. Add a business guardrail, such as checkout completion, once you trust the first two.

Back to blog
Share:

Product updates, once a month

New features and the occasional guide. No spam, unsubscribe any time.