Dataset
Negation screen, every flagged title
Every record title in the corpus that matched a negation marker, in two categories: discordance candidates, and non-effects on a pharmacokinetic or safety measure, which are counted separately because they usually mean the opposite thing.
6 rows8 columnsGenerated 13 October 2026
The file
6 rows, 8 columns, 1 KB. Comma separated, UTF-8, one header row, quoted where a value contains a comma or a quote. Download discordance-screen.csv.
How it was derived
Ten negation markers were matched against the title of every record in data/studies/*.json by scripts/build-evidence.mjs; this file is the flattened output as stored in data/evidence/*.json. A match in a pharmacokinetic, tolerability, bioavailability, exposure, safety or interaction context is routed to the second category.
Regenerate it with node scripts/sentinel-analysis.mjs --write. The same run produces every figure in 5 flagged titles in 902, so the page and the file cannot drift apart.
Columns
| Column | What it holds |
|---|---|
| slug | The compound whose query returned the record. |
| name | Display name as the corpus holds it. |
| category | discordance_candidate or non_effect_pk_or_safety. |
| marker | Which of the ten markers matched. |
| year | Publication year of the record. |
| pmid | PubMed identifier, where the record carries one. |
| doi | Digital object identifier, where the record carries one. |
| title | The record title, verbatim. This is the only text that was read. |
What it cannot be used for
A row is a title that matched a pattern. It is not a finding, not a conclusion, and not a judgement on the study. Any use of this file that treats a row as evidence that a compound does not work is a misuse of it.
The limits of the analysis built on this file are stated on the investigation itself, and the rules this publication applies to all of its counting are on the methods page.