Sample Ratio Mismatch Calculator (A/B Test SRM Check)

You planned a 50/50 test, but 12,000 users saw A and 12,700 saw B. Is that random noise or a bug in your assignment? A sample ratio mismatch (SRM) is a split imbalance too large to be chance, and when it happens the whole experiment is usually compromised: something is filtering, redirecting or mis-bucketing users before they reach one variant. Enter your intended split and the users each variant actually received, and this runs a chi-square goodness-of-fit test to tell you whether the result is safe to trust.

The convention among experimentation teams is to flag an SRM when the chi-square p-value drops below 0.001. That threshold is deliberately strict: a genuine mismatch points to a data or assignment fault, not a marginal effect, so you want very few false alarms. If this tool flags a mismatch, do not read the conversion numbers. Find the cause first: bot filtering hitting one variant, a redirect that loses users, uneven caching, a broken randomisation seed, or an analytics tag that fires on only one side.