Measure experiment exposure before interpreting results
Check who actually encountered each version of a test.
Distinguish assignment from exposure
A visitor can be assigned a page variant without reaching the feature being tested. Track assignment and, where appropriate, feature exposure separately. For a checkout test, changing to an exposed-only comparison can change the question and introduce selection effects.
Investigate unexpected group sizes
If a 50–50 assignment yields 6,000 and 4,000 records, inspect randomization, event loss, and eligibility before interpreting the result. Verify whether the imbalance appears at assignment or later in delivery.
Report the basis explicitly
State which users are included in the primary analysis. Show exposure as a diagnostic alongside results for the planned assignment groups. If implementation compromised the test, record the issue and rerun after fixing it.
Try a related tool
Find the share of eligible visits that completed a defined action. Count at most one conversion per visit for this calculation.
