Core concepts
Experiments, variants, metrics, exposures, and verdicts.
Experiment
An A/B/n test on a surface of your product. It has variants (control + one or more treatments) with traffic weights that sum to 100, a traffic allocation (what % of eligible users enter), and a lifecycle: draft → running → paused → archived.
Exposure
The moment a user is actually assigned to a variant. MaxLift logs exposure only on real assignment (trigger-based), which keeps your results undiluted by users who never saw the test.
Metric
What decides the winner. Every experiment has exactly one primary metric; you can attach secondary metrics for context and guardrails that block promotion if they regress.
Verdict
The current statistical readout: collecting / winner / no-winner, with lift, confidence interval, p-value, and the SRM check. No decision statistic is shown before the planned horizon — that's how MaxLift stops you calling a test early on noise.