Dataset

The PrecepTron benchmark dataset contains 5,250 physician-scored clinical responses drawn from 7 published studies. The full dataset is available on Hugging Face.

Responses by Task

Task	Responses	Score range	Description
Management Reasoning	2,765	Case-specific	Free-text management plans scored against case-specific rubrics
BI Triage	911	Case-specific	Triage-level diagnostic assessment
CPC Bond	853	0--5	Differential diagnosis scored with the Bond score
R-IDEA	312	0--10	Consultation quality scored with the R-IDEA instrument
Diagnostic Reasoning	278	0--19	Multi-axis diagnostic reasoning evaluation
CPC Management	131	0--2	Testing plan evaluation

Schema

Each entry in the dataset contains:

Field	Description
id	Unique entry identifier
case_id	Groups entries from the same clinical case
benchmark	Task name (e.g. cpc_bond, management_reasoning)
study	Source study identifier
model	Model or participant that produced the response
grade	Physician-assigned score
response	The clinical response that was scored
rubric	Scoring rubric (JSON)
case_vignette	Clinical case text (where applicable)
question_text	The question posed (where applicable)
final_diagnosis	Ground-truth diagnosis (where applicable)
max_score	Maximum possible score for this entry