Sample Ratio Mismatch (SRM) Checker
Before you trust an experiment’s results, confirm the traffic actually split the way you intended. Enter the visitors each variation received and their intended split; this runs a chi-square test and tells you whether the difference is within chance or a red flag.
Observed traffic per variation
What a mismatch means and how to fix it
Randomization should send visitors to each variation in the ratio you configured. When the observed counts diverge far more than chance allows, something upstream of your metrics is broken — and a broken split biases every result the experiment produces. The convention is to alarm at p < 0.001. For the full method — how to read the result, diagnose the cause, and prevent it — read Sample Ratio Mismatch: is your A/B test broken?
Common causes
- A redirect or client-side test that drops or delays one variation (flicker/load races).
- Bot or spam traffic filtered unevenly across variations.
- Bucketing or randomization applied after a step that itself differs by variation.
- Logging or tracking that fires inconsistently for one arm.
- A variation erroring or timing out for a subset of users.
Run this check before reading results, as step one of your pre-launch QA. For why a clean split still is not the whole story, see sequential testing and the peeking problem and how to read the results page correctly.