Organised by what went wrong, not by what a feature is called. Nobody arrives here typing "character limits" — they arrive typing "why is my ad not delivering".
Audience floors apply per test cell. The published minimums, and the arithmetic that makes a healthy audience deliver nothing.
Two cells with the same copy cannot produce a result. How to spot a null test — including in generated campaigns, where 35 of 48 of ours shared a headline.
The phrases readers have learned to skip, and why a model cannot be trusted to check its own register.
Every figure on these pages is one we counted, with its denominator beside it. Where we have not measured something, the page says so rather than reaching for a number that sounds right.