250 systematic reviews × 19 LLMs × 3 input conditions = 8,531 syntheses scored on 6 evaluation dimensions. Accompanies "TruMed: More Context Fixes Fact-Level Errors but Not Directional Ones in LLM Clinical Syntheses" (EMNLP 2025).
Code: anonymized for review
reviews.parquet
250
SR metadata: PMID, DOI, specialty, clinical question, gold conclusion
syntheses.parquet
8,531
LLM outputs + D1–D6 scores per (review × model ×… See the full description on the dataset page:
https://huggingface.co/datasets/TruMed/trumed.
