Showing Sample Ratio Issues Without Panic
Surface assignment problems early so experiment reviews do not invent stories from broken traffic splits.
Sample ratio mismatch is one of the most common reasons a beautiful readout lies. Put expected versus observed assignment percentages near the top of every experiment view.
Pair the ratio with exposure window notes: when users enter, when they become eligible for the primary metric, and whether bots were filtered.
If the ratio drifts beyond agreed tolerance, freeze the decision field and route to instrumentation. Interpreting lift while the split is broken wastes leadership time.