One hour to write, compile and improve a StarCraft: Brood War bot. Kai McPheeters' StarSkirmish v0.2 then tests the finished bots against a fixed roster.

Across five runs per model, the leading model's bots won 57% overall, but only 7% against the three strongest human-written bots. The page includes short battle clips and traces of how the agents used their practice time.

This evaluates the model together with its coding harness under a one-hour budget. It doesn't establish strength against human players, and results vary between runs.