A/B experiments

Compare a new release with what everyone else runs, on crashes, rollbacks and your own events.

npx healkite release … --rollout 20 --experiment new-checkout --metric checkout_success --metric add_to_cart

20% of devices get the new bundle; the rest stay on the current release as the control. The first --metric is the primary one.

#Results

npx healkite experiment rel_…

Or the release page in the dashboard, which shows each metric's lift with a 95% confidence interval and a verdict. Results need at least 100 sessions per side.

#Conclude

npx healkite experiment rel_… --ship    # the new release wins: everyone gets it
npx healkite experiment rel_… --stop    # the control wins: the new release turns off