Data and training
for scientific discovery.
We provide the datasets, environments, and evaluations that frontier labs use to train models for scientific discovery and recursive self-improvement.
Expert data and post-training.
When a model stalls on scientific work, we build the expert data, rubrics, and environments the lab trains against. We stay on the run with traces, rewards, and evals so the next checkpoint is stronger at the science, and closer to discoveries that move human progress.