Protocol Reasoning
Transform trial objectives and clinical context into coherent protocol concepts.
TrialDesignBench
Rigorous benchmark for evaluating AI agents in clinical trial design
TrialDesignBench measures whether AI agents can complete high-stakes trial design workflows with clinically grounded reasoning, auditable outputs, and reproducible evaluation.
TBD
Benchmark tasks
TBD
Evaluation checkpoints
TBD
Agent submissions
What TrialDesignBench Evaluates
Agents must synthesize evidence, reason about design tradeoffs, and produce outputs that survive clinical, statistical, and safety review.
Statistical Design
Select endpoints, estimands, assumptions, sample sizes, and adaptive design choices.
Operational Feasibility
Balance enrollment, follow-up, site capacity, cost, and timeline constraints.
Leaderboard Preview
Public results will appear here when the benchmark is released.
| Rank | Model | Agent | Pass Rate | Status |
|---|---|---|---|---|
| - | Pending | Pending | TBD | Not yet released |
Methodology
End-to-end completion, not isolated answers.
Clinically anchored
Tasks reflect realistic design decisions and expert review.
Checkpoint graded
Rubrics capture critical reasoning, artifacts, and safety constraints.
Reproducible
Versioned inputs and recorded trajectories support auditability.
High stakes
Unsafe or unsupported recommendations block successful completion.