Project Confirm Oncology

FDA's Project Confirm gives us a public answer key for a hard forecasting problem: which accelerated-approval oncology programs move, stall, verify, or get withdrawn.

The run below uses official accelerated-approval outcomes as the dataset, then adds public clinical-trial, submission, and label events. Whole drug assets are held out of training, so the scores come from programs the model did not see.

If you only read one row, read the Imbruvica cutoff: what the model could see by January 2020, and what FDA did more than three years later.