Each challenge is one UI brief. Its inputs are frozen and versioned, so every output can be read against exactly what the tool was given.
The launch cohort: model and reasoning settings could be the tool’s defaults, and subagents were not ruled out. Runs are compared only within one protocol. How the protocols differ