The Voidly Atlas Score
One citable composite score per country, 0-100. Like FICO for internet censorship. Combines our calibrated forecast, recent incident density, 7-day trend, calibration quality, and unsupervised anomaly signal into a single number with a letter grade.
Updated every 15 min · CC BY 4.0 · Raw JSON · Per-country
Honest known limitation
v1 of the score weights CHANGE more than LEVEL. Chronic- censorship countries (RU, CN, KP, SY) score lower than acutely-active ones (EG, IR) despite higher absolute baselines, because nothing is shifting this week. v2 (in the lab) will add a base-rate component. Until then, cross-reference with the forecast index for absolute levels.
1 watched countries, ranked
| # | Country | Score | Grade | Composition |
|---|---|---|---|---|
| 1 | Venezuela VE | 32.1 | B- |
forecast incidents trend calibration anomaly
Composition formula
Composite formula: 40% forecast + 25% incident density + 20% trend + 10% calibration + 5% anomaly. Grade A (0-20) = open internet, Grade F (80-100) = censorship-heavy.
| Component | Weight |
|---|---|
| Calibrated 7-day forecast | 40 pts |
| Recent incident density (24h, log-scaled) | 25 pts |
| 7-day forecast trend delta | 20 pts |
| Calibration quality (more outcomes = better) | 10 pts |
| Anomaly disagreement signal | 5 pts |
Related
- /atlas/forecast — calibrated 7-day forecast (the biggest input)
- /atlas/recent-changes — recent incidents (24h density input)
- /sentinel/backtest — calibration quality (model honesty input)
- /atlas/models — full ML model registry