Independent tests of AI models on real product management work. Graded by Jev, reviewed blind by Martin.

Releases

Results are published as versioned releases, not a table that changes under you. Earlier releases stay available and unchanged; corrections ship as documented patch versions.