Give it a terminated trial. It returns a cause of failure across six diagnostic lenses — and tells you plainly which ones it cannot answer.
An autopsy marks what the record can support — and where it goes silent. Not a real program.
Biomarker-unselected enrollment. 41% of the treatment arm lacked detectable target expression at baseline per the reported screening data.
OS primary, adequately powered, consistent with three comparable programs in the same indication.
hERG signal present in the preclinical record at 3× Cmax. Carried into Phase III without a mitigation strategy on file.
MTD-anchored. Exposure held within the Phase I range across both arms.
No archival tissue collection reported. No correlative sub-study published. The evidence required to assess this lens does not exist in the record. Pure Monograph will not infer it.
Two approvals landed in the indication during the enrollment window. Neither altered the comparator arm.
Illustrative readout. Not a real program.
Most systems answer everything. That is the failure mode, not the feature.
The refusal fires at the output boundary — not as a prompt instruction, not as a caveat buried in the wording. A gate that lives in the wording is a gate that eventually opens.
When the underlying evidence is absent, the lens returns DATA GAP and names what is missing. You learn where the record is thin, which is itself a finding.
A confident wrong answer wearing correct citations costs a development cycle. It is more expensive than an obvious error, because nobody catches it in review.
The other direction. Give it a target, and it designs the molecule across ten dimensions — with the same refusal behaviour at every field. See Build Monograph →
No call, no demo booking, no sales sequence. Run one autopsy and judge it on the output.
Start free