What it measures
Five hundred medical research questions test study-design selection, evidence-gap identification, PICO formulation, and hypothesis coverage. The release reports heuristic and retrieval baselines with an explicit weighted composite.
Reading the measurement
Partial credit can distinguish a useful approach from an empty attempt, but it is not the same as a completed task. Read the rubric, full-success count and score distribution together. For mathematical work, preserve the difference between expert review and acceptance by a formal proof checker.
Published evaluation
Published deterministic reference baselines on the 500-problem evaluation suite. Composite weights are 30% study-design F1, 25% evidence-gap F1, 25% PICO/estimand score and 20% hypothesis coverage. These rows are heuristics and retrieval baselines, not tested frontier LLMs; the oracle reference upper bound is excluded. Scores assess structured research proposals, not patient-care outcomes. Ulam powers RLTasks; affiliation disclosed.
Where this fits in an evaluation suite
Read accuracy together with error and abstention rates. Domain averages can hide language gaps, specialist failures, and poorly calibrated confidence.
Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.
A reproducible comparison
Published deterministic reference baselines on the 500-problem evaluation suite. Composite weights are 30% study-design F1, 25% evidence-gap F1, 25% PICO/estimand score and 20% hypothesis coverage. These rows are heuristics and retrieval baselines, not tested frontier LLMs; the oracle reference upper bound is excluded. Scores assess structured research proposals, not patient-care outcomes. Ulam powers RLTasks; affiliation disclosed.
Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.
PRIMARY EVIDENCE & FURTHER READING
Reviewed 2026-10-27. Explanation and interpretation by RLTasks. This evaluation is from Ulam, which powers RLTasks.