Archive
Data
Every number on this site is computed from one of four tables, and every table is downloadable.
Every figure this publication has printed comes from one of these four tables. They are generated by a single script against files in the open, and they are the point: a number nobody can check is not reporting.
| Dataset | Rows | Columns | Size | Used by |
|---|---|---|---|---|
| Retrieval coverage, per compound | 77 | 12 | 7 KB | Index depth |
| Derived evidence tier, per compound | 77 | 13 | 14 KB | Evidence mix |
| Negation screen, every flagged title | 6 | 8 | 1 KB | Negative findings |
| Federal Register, four compounding terms | 673 | 10 | 286 KB | The register |
Each dataset has a page of its own describing what every column holds and what the file cannot be used for. The methods page states the rules that apply to all four.
Format and reuse
All four are comma-separated UTF-8 with a single header row, quoted where a value contains a comma or a quotation mark. They can be re-generated from this repository's corpus with one command, which is stated on each dataset page.
They may be re-used with attribution. The one thing they must not be used for is stated on every dataset page individually, because it differs: a derived label is not a clinical judgement, and a flagged title is not a finding.