Sentinel · global forecast

Current-regime shutdown risk by country

30 watched countries, ranked by the calibrated 7-day risk level. This ranking tells you where censorship is entrenched right now — it is not a list of countries about to go dark. The forecast is post-isotonic recalibrated as of 2026-05-20 — predictions in the [0, 0.10) band are now lifted ~15× to match observed outcomes. See /sentinel/backtest for the honest calibration plot.

Updated every 30 min · last refresh Jul 5, 2026 · CC BY 4.0 · Raw JSON

What this forecast is — and is not

The 7-day forecast is a current-regime risk signal: it reliably tells you which countries are currently in a sustained internet-blocking regime. It is not a shutdown-onset predictor — it does not reliably call a new shutdown before it happens.

On an honest forward-temporal split (train on the past, test only on the strictly-future window) the production model scores AUC 0.589 on its raw label — near chance.
The label target_7day is a 7-day sliding window, so it is 98.9% autocorrelated day-to-day. A trivial predict-yesterday baseline scores AUC 0.957 — high numbers here mean “what is blocked stays blocked,” not predictive skill.
On the rows where a shutdown actually begins (transitions) the forecast scores AUC ~0.33 — below chance.

Use it to see where censorship is entrenched today and to read the SHAP drivers + conformal interval — not as an early-warning siren. Read the full honest re-evaluation →

High

≥ 50%

Elevated

30–49%

Watch

15–29%

Low

< 15%

Biggest movers — last 7 days

vs 2026-06-28 → 2026-07-05

↑ Risk increased

↓ Risk decreased

Source: /v1/sentinel/movers joins today's and 7-day-prior snapshots from sentinel_forecasts. Min |Δ| 0.02 to suppress noise. Some current jumps are an artifact of the May 20 isotonic recalibration — see the refit finding.

Full ranking

How to read this page

Max risk is the highest daily forecast probability in the next 7 days for that country. Click into /atlas/forecast/[cc] for the full 8-day chart + SHAP drivers + evidence.
Calibration — predictions for the 30 watched censorship-heavy countries are run through an isotonic calibrator fit on 810 live (predicted, observed) pairs. The other ~120 countries get the raw model output.
Honesty — the model isn't magic. On an honest forward-temporal split it scores AUC ~0.59 and has no skill at calling a new shutdown before onset (transition-row AUC ~0.33). See the forecast onset-skill finding, /sentinel/backtest for current Brier score / per-bin Δ, and /methodology#validation for the honest splits.
Subscribe — push delivery via webhooks at /alerts, or hit POST /v1/forecast/batch for programmatic access.

/atlas/forecast-regions — per-region aggregate 7d forecast (Africa, Asia, MENA, etc.)
/atlas/elections — 90-day election shutdown risk (subset of this index)
/atlas/recent-changes — daily journalist dashboard with delta highlights
/sentinel/calibration — 90-day rolling empirical coverage time series
/sentinel/backtest — reliability diagram + per-country backtest

Current-regime shutdown risk by country

Biggest movers — last 7 days

Full ranking

PakistanPK

UzbekistanUZ

CubaCU

EthiopiaET

VenezuelaVE

MyanmarMM

SyriaSY

KazakhstanKZ

EgyptEG

Saudi ArabiaSA

BelarusBY

IranIR

RussiaRU

NigeriaNG

SudanSD

IndiaIN

ChinaCN

PhilippinesPH

NicaraguaNI

LebanonLB

EritreaER

BrazilBR

MalaysiaMY

VietnamVN

North KoreaKP

ThailandTH

BangladeshBD

IndonesiaID

TurkeyTR

TurkmenistanTM

How to read this page

Related