Experiments
The six daily prompts show what a model makes. An experiment checks something a still can't: a script lays a rebuilt screen over the original pixel by pixel, or plays a game for a minute and a half and counts the coins.
A new model joins after its six daily prompts when it is a big lab's top model or its main coding tier, or when it comes with a loud coding claim worth checking. Small and fast tiers sit these out.

