Discovery Loop · Real Measured Numbers
Quantitative Benchmark — beyond fabricated numbers
The yes/no benchmark tests whether the platform can call a drug response correctly. This one goes further: it tests whether the platform can predict a real, measured quantity — the order of drug potencies — and be scored against actual laboratory measurements. Every number here is a real measured value pulled from a public experimental database, each with a citation. None is estimated or invented.
The answer key is real laboratory potency data (ChEMBL) for approved HER2 inhibitors — each value carries a verified publication. No number is fabricated.
The platform commits its full ranking first. Only then are the measured values attached and the ranking graded. The order is enforced by the data itself.
A pair of drugs is only scored when their measured potencies are clearly separated (≥3× apart). Pairs too close to call are recorded but never counted.
The platform ranks the inhibitors blind, then the real measured values are revealed and the ranking is scored against them.
No quantitative run yet. Press Run benchmark to have the platform commit a potency ranking and score it against the real measured values.