Mistake: rotating offers with no holdout control

Mistake: rotating offers with no holdout control

Failure mode: your traffic-distribution rule splits across three offers and you crown a winner after a day. But you never held a control share constant, so you cannot tell if the "winner" won on merit or on a lucky traffic-quality swing.

Fix — distribution rules:
— Split evenly at first. Equal weights remove your bias from the test.
— Set a minimum-data gate: no offer gets paused until it has, say, 100+ clicks or 5+ conversions. Pausing on 12 clicks is noise.
— Keep one offer as a stable baseline across the whole test so you can normalize against traffic-quality drift.
— Change one variable per test. Rotating offers and prelanders at once makes the result unattributable.
— Use the tracker's built-in rotation/weight feature, not manual link swaps that lose historical data.

Verify: before declaring a winner, confirm each arm cleared the data gate and the baseline held steady.

Save this SOP. Run this before every launch.