srm-check

To check for sample ratio mismatch (SRM), apply a chi-squared goodness-of-fit test against your designed or default ratios to verify randomization integrity before proceeding with outcome analysis.

SRM occurs when the treatment and control split deviates significantly from the designed ratio, indicating broken randomization. When SRM is detected, treatment-effect calculations cannot be trusted and analysis must be halted.

Yes, SRM detection can auto-fire when experiment data is loaded, automatically identifying common assignment columns like variant, group, treatment, and arm to guard downstream analyses without manual intervention.

You can run per-segment and time-based SRM diagnostics on your experiment data to identify specific segments or temporal windows where randomization integrity breaks and investigate the root cause.

When the chi-squared test yields a p-value of 0.01 or lower, the system reports a BLOCK status, halts all downstream analysis, and offers actionable investigation directions to resolve the randomization failure.

Yes, SRM detection executes a chi-squared goodness-of-fit test against your designed ratios. If custom ratios are not provided, it defaults to standard assumptions to evaluate the experiment data split.