Confidence-calibrated, citation-backed read on every target and result in the portfolio — what's already known in the literature, and whether each result agrees with it.
Literature is real — 77 PubMed records, real PMIDs, extracted affinities. Experiment results are authored test cases, one per verdict path, chosen to probe the real priors: there is no Foundry API token behind this page.
Every target resolved against NCBI Gene at snapshot-build time — "HER2" is a display name, not an identifier. The expected KD range is the log-space interquartile core of extracted affinities, not the full envelope, which on real literature spans orders of magnitude legitimately.
Results below are authored test cases — one per verdict path, with KD values chosen against the real published priors. The verdicts, ranges and citations are computed; the inputs are not lab data.