One list, no cherry-picking. Tests move from "Live" once every model has a published result and a blind human read has happened.
Same brief, three AIs, one real business rebuild. Claude's result is done; Gemini and Codex are pending the same brief.
A working web app, built twice, judged on code quality, autonomy, and how much hand-holding it needed.
Claude, GPT, and Gemini each get one campaign brief — copy, targeting logic, and reporting compared side by side.