Wire
Codex Astra wins 18 of 18 Brood War games
Codex Astra / xhigh won all 18 of 18 matches in Ben Swerdlow’s Brood War Bench results, a 19-configuration round robin that still rated every agent at beginner level. The same report puts the winner at $10.54 per game and says a human using a photon rush would beat every system, so the 100% score measures a narrow harnessed matchup—not general real-time strategy competence. Builders should file this beside the archive’s evaluation-safety case for 122 agent runs: game wins are useful regression signals, but not permission to infer reliable long-horizon action.