Artifacts
What each model built on each task, shown on the held-out cases it was graded on.
Reaction kinetics forecast
Each model worked out a catalysed reaction from eleven lab runs (infrared spectra, HPLC, notebook) and forecast five species over 8 hours in six hidden operating cases, where a value scores only within 1% (or 0.001 mol/L) of the truth. One of those graded cases is shown, with each model's best and median attempt.