Is any of this true?
Start with the validation status and the limitations, then read the results with their artefacts.
Documentation
The canonical documents live in the repository. This page only routes to them — it deliberately does not copy them, so a number cannot drift between two pages.
Start here
Read them in this order if you intend to quote a number.
Mission, validation status, what is implemented, the results including the negative ones, the nine limitations and the reproduce commands.
README.mdWhere the canonical version of each thing lives, what state it is in, and what has been superseded. It links; it does not copy.
docs/INDEX.mdMetric definitions, estimators, multiple-testing correction and threats to validity. A new metric updates this file before it updates any output.
docs/METHODS.mdThe generated data audit: counts per source, coverage by year, timestamp quality, freshness and database size.
docs/DATA.mdThe metrics.json and report contract. Change this file before changing a
metric, or the artefacts will disagree.
The state source, the local gates and the freeze register. Note what is not applied: branch protection is still off.
docs/governance/GOVERNANCE.mdQuickstart
Standard library only; Python 3.10, 3.11 and 3.12 are tested in CI.
1,160 collected, 1,157 passed, 3 skipped. No network, no API key.
Offline build from the committed snapshot in under 60 s:
out/demo/index.html.
Needs network, about three minutes, still no API key. Writes the report, the dashboard and the CSV.
source: README.md → Reproduce; PROJECT_STATE.yaml → tests; .github/workflows/ci.yml
git clone https://github.com/ChazY11/medradar.git
cd medradar && pip install -e .
# 1,160 offline tests · no network · no API key
python -m pytest -q -o addopts=""
# offline demo → out/demo/index.html (< 60 s)
medradar demo
# live run → report, dashboard, summary.csv (~3 min)
medradar run --months 24 --with-trade --with-dashboard
Reading map
Pick the route that matches why you are here; each ends at the document that owns the explanation.
Start with the validation status and the limitations, then read the results with their artefacts.
The procedure is reproducible offline; the 1.3 GiB snapshot is not committed, so exact counts are not.
The highest-value contributions are curated knowledge files — company aliases, track dictionaries, product-code mappings — not code.
The state file is the single source of truth; the research index and the status blocks are generated from it. Branch protection is not applied.
source: docs/INDEX.md → How to read this repository; README.md
Contributing
The core package stays standard-library-only. Every new metric updates
docs/METHODS.md and docs/SCHEMA.md first, so a metric cannot enter the
codebase without a definition.
source: README.md → Contributing; CONTRIBUTING.md
Licence and citation
Cite the repository and the version, not a published paper: no paper here is submitted or peer reviewed. Paper drafts and their claim maps live in papers/.
MedRadar (2026). Research infrastructure for public medical-technology evidence, v0.5.0, dataset snapshot 2026-10-31. https://github.com/ChazY11/medradar
source: README.md → Licence; papers/README.md
A fact is stated once, in its canonical file, and linked everywhere else. If you find the same number in two places without a pointer, one of them is already drifting — report it rather than copy it.
source: docs/INDEX.md → Maintenance rules for this page