Country shutdown risk
Where to
look closer.
Seven-day model estimates.
Compare countries. Inspect their evidence.
Country risk ranking
Source JSONCompare countries. Exact shutdown dates remain unreliable.
Source snapshot
shutdown_risk_v93 country records have old or unconfirmed data. See the flags beside each date.
- 01IndiaAs of45.3%Elevated
- 02PakistanAs of33.2%Elevated
- 03MyanmarAs of27.8%Elevated
- 04IranAs of23.2%Elevated
- 05RussiaAs of22.7%Elevated
- 06IraqAs of16.6%Watch
- 07TanzaniaAs of12.9%Low
- 08JordanAs of8.5%Low
All 60 country records
| Country | Model estimate | Source band | As of / freshness | Structural input | OONI anomaly input |
|---|---|---|---|---|---|
| IndiaIN | 45.3% | Elevated | As of | 46.6% | 46.1% |
| PakistanPK | 33.2% | Elevated | As of | 18.9% | 81.6% |
| MyanmarMM | 27.8% | Elevated | As of | 46.6% | 30.3% |
| IranIR | 23.2% | Elevated | As of | 18.9% | 63.2% |
| RussiaRU | 22.7% | Elevated | As of | 16.6% | 48.3% |
| IraqIQ | 16.6% | Watch | As of | 8.1% | 65.1% |
| TanzaniaTZ | 12.9% | Low | As of | 2.8% | 83.9% |
| JordanJO | 8.5% | Low | As of | 2.8% | 80.5% |
| GNGN | 5.0% | Low | As of | 1.8% | 77.9% |
| KenyaKE | 5.0% | Low | As of | 1.8% | 77.8% |
| SyriaSY | 5.0% | Low | As of | 3.4% | 60.8% |
| EthiopiaET | 3.2% | Low | As of | 1.0% | 82.8% |
| BelarusBY | 1.9% | Low | As of | 1.0% | 68.8% |
| TurkeyTR | 1.7% | Low | As of | 2.8% | 29.4% |
| KyrgyzstanKG | 1.6% | Low | As of | 0.4% | 67.9% |
| NigeriaNG | 1.6% | Low | As of | 0.8% | 59.4% |
| OmanOM | 1.6% | Low | As of | 0.4% | 75.7% |
| VenezuelaVE | 1.6% | Low | As of | 1.8% | 24.7% |
| VietnamVN | 1.2% | Low | As of | 0.4% | 40.0% |
| TGTG | 0.8% | Low | As of | 0.4% | 11.2% |
| KazakhstanKZ | 0.7% | Low | As of | <0.1% | 79.5% |
| UzbekistanUZ | 0.7% | Low | As of | <0.1% | 80.8% |
| United Arab EmiratesAE | 0.3% | Low | As of | <0.1% | 74.3% |
| AzerbaijanAZ | 0.3% | Low | As of | <0.1% | 62.2% |
| BFBF | 0.3% | Low | As of | <0.1% | 76.9% |
| BahrainBH | 0.3% | Low | As of | <0.1% | 62.3% |
| ChinaCN | 0.3% | Low | As of | 0.4% | 0.8% |
| CubaCU | 0.3% | Low | As of | <0.1% | 74.0% |
| DenmarkDK | 0.3% | Low | As of | <0.1% | 26.7% |
| EgyptEG | 0.3% | Low | As of | <0.1% | 62.2% |
| SpainES | 0.3% | Low | As of | <0.1% | 56.5% |
| FranceFR | 0.3% | Low | As of | <0.1% | 59.1% |
| Sri LankaLK | 0.3% | Low | As of | <0.1% | 55.1% |
| MLML | 0.3% | Low | As of Older source data | <0.1% | 42.3% |
| NorwayNO | 0.3% | Low | As of | <0.1% | 27.0% |
| Saudi ArabiaSA | 0.3% | Low | As of | <0.1% | 45.4% |
| SudanSD | 0.3% | Low | As of Older source data | 8.1% | <0.1% |
| UgandaUG | 0.3% | Low | As of | <0.1% | 77.0% |
| ZimbabweZW | 0.3% | Low | As of | <0.1% | 71.0% |
| IndonesiaID | 0.2% | Low | As of | 0.4% | 0.5% |
| AustriaAT | 0.1% | Low | As of | <0.1% | 4.6% |
| CanadaCA | 0.1% | Low | As of | <0.1% | 2.3% |
| SwitzerlandCH | 0.1% | Low | As of | <0.1% | 5.1% |
| FinlandFI | 0.1% | Low | As of | <0.1% | 7.5% |
| United KingdomGB | 0.1% | Low | As of | <0.1% | 3.8% |
| IrelandIE | 0.1% | Low | As of | <0.1% | 5.1% |
| ItalyIT | 0.1% | Low | As of | <0.1% | 3.4% |
| JapanJP | 0.1% | Low | As of | <0.1% | 13.6% |
| NetherlandsNL | 0.1% | Low | As of | <0.1% | 2.5% |
| New ZealandNZ | 0.1% | Low | As of | <0.1% | 10.2% |
| PortugalPT | 0.1% | Low | As of | <0.1% | 2.2% |
| SwedenSE | 0.1% | Low | As of | <0.1% | 2.3% |
| TajikistanTJ | 0.1% | Low | As of | <0.1% | 14.2% |
| TurkmenistanTM | 0.1% | Low | As of Source flags stale data | 1.8% | <0.1% |
| United StatesUS | 0.1% | Low | As of | 0.4% | 0.1% |
| BangladeshBD | 0.1% | Low | As of | <0.1% | 1.5% |
| GermanyDE | 0.1% | Low | As of | <0.1% | 1.1% |
| AustraliaAU | 0.0% | Low | As of | <0.1% | 0.3% |
| BelgiumBE | 0.0% | Low | As of | <0.1% | <0.1% |
| ThailandTH | 0.0% | Low | As of | <0.1% | 0.2% |
Per-country pages retain shutdown history, component details and source JSON. Open the complete response.
Page checked the data service at . Country “as of” dates refer to the source data, not this page visit.
The question
matters.
Comparing countries and calling the day of a shutdown are different jobs.
Inspect the model cardForward validation · AUC
Country ranking
0.94Within-country day timing
0.380.5 is the random-ranking reference. AUC measures ranking skill, not the percentage of correct predictions. Values come from the leaderboard’s evaluation fields.
Use the country comparison. Within-country day timing has not been established as reliable.
Follow the changes.
Choose a country and alert settings on the existing subscription page.
Set up risk alertsKeep the score honest.
Inspect archived predictions, outcomes and the model’s own accountability record.
Inspect accountabilityFollow it to the source.
Sources, scope and freshness
The shutdown-risk model uses Access Now KeepItOn STOP shutdown labels and OONI measurement signals. Country-level estimates are less informative about subnational events; sparse records also limit what can be learned.
An automatic daily check picks which model version serves. Some versions also use Cloudflare Radar outage records, which are licensed CC BY-NC 4.0. While one of them serves, the model card names Cloudflare Radar in its sources. No paid Voidly product reads these scores.
“Source snapshot” is the response’s generation timestamp. Country dates, reported data age, and the source’s stale flags are separate. A fresh page request does not make old measurements fresh. Missing flags remain unconfirmed; dates more than three days old receive an additional warning.
Source score formula: isotonic(sigmoid(0.6*logit(shutdown_risk_v4_prob) + 0.40*logit(honest_forecast_v1_prob)))
- Forward cross-country AUC
- 0.91
- Forward cross-country precision–recall lift
- 8.54×
Country bands are copied from the leaderboard, not recalculated from an invented threshold. The table preserves zero estimates and distinguishes missing inputs from zero.
Related Atlas forecasts and case studies
Atlas Sentinel answers a different question with a different model. Its heatmap, validation and case studies are separate from the shutdown-risk ranking above. A worked example explains the alert mechanism; it does not establish this model’s day-timing skill.