What we measured, including what did not work.

Our own leave-one-out and generated-molecule measurements: known binders sit 0.41-0.53 from each other, our output sits 0.14-0.17 from them, and recovery AUC falls 0.945 to 0.62 on that slice. Also the negative result that 3D physics lost to 2D fingerprints in this regime.

A measured reaction-search result (+22.9% products per reaction evaluation, 95% CI [+16.2%, +33.1%], 24/24 paired runs) and a worked failure: an azole coupling template that could never fire, invisible because a dead template produces no wrong molecules, only missing ones.

A measured, non-circular separation between the two target sets (1 of 11 versus 6 of 6 carry aging-perturbation data) plus two pre-registered no-effect results published with their confidence intervals, which is the part almost nobody publishes.

A company named after qubits publishing the proof that its own quantum kernel is a classical RBF (max absolute difference 1.44e-15 against scikit-learn), plus three operators killed by pre-declared controls, including one whose apparent signal was entirely a sparsity artifact.

A measured chemical-distance comparison between the four clinically failed CXCR2 compounds and our three designed candidates, showing the failed drugs sit closer to each other than ours sit to any of them, with the exact Tanimoto ranges published.

A per-hallmark druggability read from our own target work: which of the twelve have usable small-molecule chemistry, which are covered by our 21 runnable targets, and the measured finding that 11 hallmarks carry roughly 2.9 independent directions rather than 12.