Computational versus experimental validation

The self-driving lab's binding constraint is a century of footnotes, not the model

Two WashU groups arrive at the same point from opposite ends: the prediction step works, and the corpus of machine-readable synthesis recipes it needs is still being assembled.