By the numbers
Voidly Atlas aggregates 4,060 citable censorship incidents from 195,848 evidence rows. The corpus draws on 2.2B+ raw measurements from OONI, IODA, and CensoredPlanet, plus 40+ probe nodes operated directly by Voidly and community partners.
Where the most incidents happened
Where the evidence comes from
No single measurement network captures the full picture. Voidly aggregates four independent sources so every confirmed incident has ≥2 corroborating signals — and structurally cites the upstream so journalists can audit our chain.
How accurate is our shutdown forecast?
Voidly Sentinel — the forecast model that powers /atlas/elections and per-country risk scores — publishes three honest accuracy splits so readers can choose which figure matches their risk tolerance.
| Split | AUC | F1 | Use for |
|---|---|---|---|
| Stratified (inflated) | 0.98 | 0.79 | Sanity check only |
| Time-based (floor) | 0.50 | 0.00 | Conservative baseline |
| LOCO median (honest) | 0.91 | 0.55 | Cite this |
Full methodology including live calibration drift is published at /methodology#validation and live at /sentinel/calibration.
Cite
Numbers in this edition reflect the corpus as of the page's dateModified — they will continue to grow. For the moving live numbers see /atlas; for the freeze-frame this annual edition represents, see the BibTeX + 5 other citation formats above. License: CC BY 4.0.
Related Atlas surfaces
- /atlas — live hub (this annual edition is the freeze-frame view)
- /atlas/recent-changes — what shifted in the last 24h / 7d
- /atlas/methods — 8 blocking techniques explained
- /atlas/elections — 90-day election shutdown forecast
- /atlas/cost-of-shutdown — economic-impact calculator
- /atlas/findings — curated deep-dives w/ researcher bylines
- /methodology — full pipeline + 3 honest accuracy splits