Research

Benchmarks and technical work on AI for clinical development operations.

Selected work

01
Evidence changes. Oversight must keep up.A conceptual timeline connects an initial source record, newly available evidence, and an updated oversight record. It illustrates the evaluation premise, not benchmark results.EVIDENCE / TIMETRIAL OVERSIGHTInitial recordNew evidenceUpdated viewReviewRevisitTracet₀t₁t₂
Conceptual illustration · evidence arrives over time
Benchmark

ClinOpsBench:
Trial Oversight v1.0

The strongest frontier agents catch most critical issues. None met the complete oversight standard.

15 models. 45 runs in the published, expert-curated trial simulation. Explore the results, source evidence and representative trajectories. Private held-out worlds are available on request.

Explore the benchmark