You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Teams want to start an experiment at a small rollout (for example 5%), confirm that exposure works and that no guardrail metric moves, and then increase the rollout in steps. Today this is a manual loop: change the rollout percentage on the flag, wait, open the exposures tab, check the split and the $multiple count, check the guardrail metrics, and repeat. Nothing in the product guides or records this loop. Existing work covers parts of it: SRM explanation (PR #106289), $multiple warning (#59710), guardrail alerts (#67202), and alerts that can stop an experiment (#101995). No issue covers the ramp itself.
Describe the solution you'd like
A Ramp plan on the experiment: an ordered list of rollout steps, for example 5% → 25% → 50% → 100%, each with a minimum hold time.
A checkpoint between steps. PostHog shows a checklist for the current step: exposure split matches the rollout (SRM check), $multiple rate is under the threshold, exposures are counted in the expected event, no guardrail metric has a significant negative movement.
The user confirms each step by hand. Optionally, the step can advance automatically when all checks pass (opt-in).
Each step change is written to the activity log with the checkpoint results.
Is your feature request related to a problem?
Teams want to start an experiment at a small rollout (for example 5%), confirm that exposure works and that no guardrail metric moves, and then increase the rollout in steps. Today this is a manual loop: change the rollout percentage on the flag, wait, open the exposures tab, check the split and the
$multiplecount, check the guardrail metrics, and repeat. Nothing in the product guides or records this loop. Existing work covers parts of it: SRM explanation (PR #106289),$multiplewarning (#59710), guardrail alerts (#67202), and alerts that can stop an experiment (#101995). No issue covers the ramp itself.Describe the solution you'd like
$multiplerate is under the threshold, exposures are counted in the expected event, no guardrail metric has a significant negative movement.Describe alternatives you've considered
Additional context
Related: #67202, #101995, #59710, #22101, #35750, PR #106289, PR #57767 (significance alerts).