# AI support experiment plan
Hypothesis and decision this test will support:
Experiment type: pure model comparison / full service comparison
Control and candidate versions:
Dataset and knowledge versions:
Sampling unit, strata and repeat-customer limits:
Assignment unit and stable allocation rule:
One responding arm; shadow write policy:
Primary outcome, denominator and maturity window:
Critical safety and quality gates set before the run:
Judge version, human calibration and repeatability checks:
Actual model route logging and substitution policy:
Attempt limit, cost ceiling and unknown paid-outcome handling:
Sample requirements, uncertainty method and stop rules:
Unscored and excluded reason codes:
Results by segment, limitations and final decision: