zifeng-ai/biomedicine-deep-research
Biomedicine Deep Research Complete materialized data for the 13-benchmark biomedicine deep research track. The files preserve the host materializer's directory layout. public/development/: labeled fit and tune examples. public/verifier/: unlabeled evaluation inputs. public/reference/: audited biomedical source allowlist, RiskCalcs, and its notice. private/: evaluation labels, mounted only into the separate grader during benchmark runs. The complete tree is downloadable from… See the full description on the dataset page: https://huggingface.co/datasets/zifeng-ai/biomedicine-deep-research.
Biomedicine Deep Research
Complete materialized data for the 13-benchmark biomedicine deep research track. The files preserve the host materializer's directory layout.
public/development/: labeled fit and tune examples.public/verifier/: unlabeled evaluation inputs.public/reference/: audited biomedical source allowlist, RiskCalcs, and its notice.private/: evaluation labels, mounted only into the separate grader during benchmark runs.
The complete tree is downloadable from this public dataset, including evaluation labels. The labels therefore are not secret outside the task runtime. Do not mount private/ into the agent or verifier containers.
release-manifest.json records the SHA-256 and byte length of every materialized data file. The task manifests contain additional source hashes, split counts, and public/private alignment checks. The task materializer downloads and verifies the current dataset main each time it creates a fresh output directory.
Source data: DeepEvidence at 06b349b24ee18a8d5a2db3e498884d5ad9600f76 and BioDSA at 29aa8a1ffad374b0e6d8db9f6734062c304ac56a. The included RiskCalcs collection is from Clinical-Tool-Learning at 1c64a637b8eecce887cb4b1ee97c8429cdfbfade under the NLM Public Domain Notice retained in public/reference/NOTICE.txt. Check the upstream source terms before redistribution or reuse.