BVCT per-prediction PandaDoc PDFs 287-2138

BACKGROUND. Most medical-AI evaluations are retrospective. We report a prospective adjudicated cohort of 1,427 pre-readout clinical-benefit predictions from a patient-data-free causal virtual-patient model (1,363 unique predictions, 64 with multiple endpoint-readout events). Unit of analysis: per-trial prediction of clinically meaningful benefit vs SoC at registrational endpoint — the upstream gate determining which patients are exposed to investigational drugs. METHODS. BVCT is patient-data-free and first-in-human-blinded for the asset: no patient-level data, sponsor confidential information, first-in-human data, or readouts are inputs. Inputs are drug specification and public trial design; output is a deterministic causal effect-size estimate vs SoC. Binarisation: SUCCESS→GO at HR<0.8; else NO-GO. Past readouts inform knowledge updating within the self-coherent virtual body, not data-driven training. Each prediction was sealed by a PandaDoc Signature Certificate before its readout. Reference: per-(TA, entry-phase) BIO/QLS/Informa 2011-2020 multi-phase Likelihood of Approval (LoA), cohort-weighted. CIs: Agresti-Coull 95%. RESULTS. 1,427 prediction-readout events issued 21 Aug 2022 – 30 Apr 2026 (median ex-ante lead 156 d). Combined cohort (FINAL n=687 + SIMPLE n=740; n_GO=626): TP=561, FP=65, TN=790, FN=11. PPV = 89.6% (Agresti-Coull 99% CI 86.0–92.4). Primary inferential benchmark: vs cohort True-GO prevalence 40.1%, lift +49.6 pp, P<0.001 (construct-equivalent; pre-specified). Contextual benchmark: vs BIO 2011-2020 LoA 28.1%, lift +61.5 pp (not construct-equivalent). FINAL-only sensitivity 89.7%. S4 deterministic envelope (n_GO_S4=247): PPV 89.9% (83.7–93.7) with adjudicator discretion algorithmically removed. 374/1,801 issued predictions (20.8%) remain unread-out at cutoff. CONCLUSIONS. Results are consistent with strong prospective forecasting (PPV 89.6%, NPV 98.6%, F1 0.937, DOR=PLR/NLR 620 as binary-classification metrics, not clinical-diagnostic tests) and stationary performance across five readout years (P=0.48). Adjudication was author-performed; both authors are co-founders/co-CEOs/shareholders of BioinvestGPT (comprehensive CoI). The S4 deterministic envelope (PPV 89.9%, adjudicator discretion algorithmically removed) bounds residual adjudicator-bias at 0.3 pp. Independent academic re-adjudication is invited.