The MIT License covers everything in this repository that this project authored: the
tool surface, the evaluation and construction pipelines, the scenario specifications
(benchmark/_scenarios/, including scenario.toml, rubrics.toml and policy_rubrics.toml), the
task templates, the judge prompts, and the documentation.
It does not and cannot cover the third-party material described below. Nothing here restricts your use of the MIT-licensed code; this file exists so the provenance of the rest is unambiguous.
The policy-grounding axis is backed by 67 clinical, regulatory and operational guideline documents
authored by third parties. benchmark/corpus_manifest.csv is the authoritative per-document record:
its license column classifies every one of them, and source_url names where it came from.
| Class | Count | Shipped here? |
|---|---|---|
PD-USGOV — United States Government works |
16 | Yes, as Markdown under benchmark/corpus_public/ |
RESTRICTED / RESTRICTED-NO-REDIST — publisher terms permit retrieval, not redistribution |
40 | No |
CC-BY-NC-ND-4.0, CC-BY-NC-ND, CC-BY-NC-4.0 — NonCommercial and/or NoDerivatives |
6 | No |
PMC-AUTHOR-MS — PubMed Central author manuscripts |
3 | No |
CC-BY-4.0 |
1 | No |
CC-BY-NC-SA-3.0-IGO — WHO intergovernmental licence |
1 | No |
Every document carries exactly one classification, and the classes above are disjoint and sum to
67. The 51 documents not shipped are assembled on your own machine by
construction/scripts/acquire_corpus.py from the sources the manifest names.
Redistribution is blocked for different reasons across these classes, which is why the recipe ships instead of the corpus. Forty documents are under publisher terms that permit retrieval but not redistribution, and three are PubMed Central author manuscripts, which are free to read but carry no redistribution grant. Seven are Creative Commons but NonCommercial. Five of those seven are also NoDerivatives and could not be redistributed in this form even with a redistribution right, because converting a PDF to Markdown produces a derivative work; one is ShareAlike, which would require any derivative to carry the same licence and so could not sit in an MIT-licensed repository.
The single CC-BY-4.0 document is not blocked by its licence at all. It is fetched rather than
shipped so that every document committed here is a United States Government work, leaving no
attribution or licence obligation attached to anything in this repository.
Each of the 51 documents retrieved during a corpus build remains governed by the terms of its own
publisher, which the license column names per document. Complying with those terms is the
responsibility of whoever builds the corpus. See
DATA_ACCESS.md.
Works of the United States Government are not subject to copyright protection in the United States (17 U.S.C. § 105). The 16 documents below are therefore in the public domain. They are redistributed here as a convenience, converted to Markdown; they are not original works of this project and are not covered by the MIT grant in LICENSE, which cannot be applied to them. No endorsement by any United States Government agency is claimed or implied.
| File | Title |
|---|---|
BPCIAdvanced_AltFS_MY4_Sepsis_08112020_508.md |
Severe Sepsis and Septic Shock: Management Bundle Measure — BPCI Advanced Quality Measures Fact Sheet |
antibiotic-use-in-the-united-states.md |
Antibiotic Use in the United States |
care-transitions.md |
Care Transitions (AHRQ PSNet Perspective) |
caring-for-patients-with-sepsis.md |
Caring for Patients with Sepsis |
cdc-core-elements-hospital-antibiotic-stewardship.md |
The Core Elements of Hospital Antibiotic Stewardship Programs: 2019 |
cdc-hospital-sepsis-core-elements.md |
Hospital Sepsis Program Core Elements: 2023 |
cdc-naloxone-distribution-eval.md |
Evaluation Profile for Naloxone Distribution Programs (OD2A) |
cdc-nhsn-ssi-definitions.md |
Surgical Site Infection Event (SSI) — NHSN Patient Safety Component Manual, Chapter 9 |
cdc-protect-patients-from-sepsis-factsheet.md |
Get Ahead of Sepsis: Protect Your Patients From Sepsis |
cms-appropriate-use-criteria-program.md |
Appropriate Use Criteria Program |
cms-cctp-fact-sheet.md |
Community-Based Care Transitions Program Fact Sheet |
cms-hrrp-planned-readmission-algorithm.md |
CMS Planned Readmission Algorithm, Version 2.1 |
cms-hwr-readmission-methodology.md |
Hospital-Wide (All-Condition) 30-Day Risk-Standardized Readmission Measure: Methodology Report |
cms-reducing-disparities-in-readmissions-2024.md |
Guide to Reducing Disparities in Readmissions |
hospital-readmissions-reduction-program.md |
Hospital Readmissions Reduction Program (HRRP) |
samhsa-tip63-moud.md |
TIP 63: Medications for Opioid Use Disorder |
No patient data appears in this repository. MIMIC-IV, MIMIC-IV-Note and MIMIC-IV-ED are distributed by PhysioNet under the PhysioNet Credentialed Health Data Use Agreement, which prohibits redistribution; the cohort bundle carrying case identifiers and reference labels is a MIMIC derivative and is distributed separately to credentialed users. See DATA_ACCESS.md.
Third-party Python packages are declared in pyproject.toml and pinned in uv.lock, each under its
own license. They are neither vendored nor redistributed here.