How the scores compare
The Voidly Atlas Score v2
Composite censorship score per country, 0–100, A–F grades. v2 fixes the v1 bug where chronic-blockers (RU, CN, KP) graded B− despite heavy structural censorship, because v1 over-weighted change-vs-level. v2 adds a 50% structural baseline so chronic blockers grade correctly.
Updated every 10 min · CC BY 4.0 · Raw JSON · v1 (legacy)
No score data right now — feed temporarily unavailable. See /status.
Why v2 — fixing change-vs-level
v1 of the Atlas Score weighted CHANGE (forecast + trend + recent incidents) at 90% and LEVEL (calibration quality) at only 10%. The consequence: countries like Russia and China graded B− despite heavy ongoing censorship, simply because nothing was actively changing this week. Egypt and Iran — same risk tier but spiking — graded D−F.
v2 adds a 50% structural baseline component derived from the past 12 months of censorship-weighted incidents, with a curated risk-tier floor for probe-blind countries (KP, TM, ER). Censorship incidents are weighted 3× heavier than network disruptions to discount upstream-outage noise. Confidence is tracked separately (0–1 field) and does not add to the score.
| Component | Weight |
|---|---|
| Structural baseline (12mo censorship + tier floor) | 50 pts |
| 30-day average forecast probability | 20 pts |
| Current 7-day max risk (active spike) | 15 pts |
| 24h incident density | 10 pts |
| Anomaly disagreement signal | 5 pts |
Scroll to see every column.
Related
- /atlas/score — v1 (still served for backward compatibility)
- /atlas/multi-horizon — 1d/7d/30d forecasts (v2 uses the 30d average)
- /atlas/changelog — full model registry timeline
- /atlas/models — per-model deep dive