Open-source experiments & tools.

Only one in five tests ever reaches significance. The reasons sit upstream, in the choices you make before launch.

Two in three people abandon a bank signup despite wanting the product. The cause is rarely one wall, but many small ones.

Last-click attribution credits the final touch and quietly starves the channels that did the real work of converting.

Bolt tracking on at the end and the data breaks six months later. Treat the plan like a spec and design it before you build.

AI generates and tests variants at speed, but the bottleneck was never speed. It's knowing what's worth testing at all.