Voidly
Atlas · 1d / 7d / 30d forecasts

One country. Three horizons.

Compare current-regime risk at 1, 7 and 30 days. Read the uncertainty ranges; these are not shutdown-onset predictions.

Model trained 2026-05-21 · Recommendation: ship_all_three · model info

20 measurement-covered spotlight countries · ranked by 30-day model score

#Country1d7d30dCoverage / interval statusCons.
1Iran IR22%36%95%Covered; intervals unvalidatedok
2China CN22%36%95%Covered; intervals unvalidatedok
3Russia RU15%64%95%Covered; intervals unvalidatedok
4Venezuela VE36%36%95%Covered; intervals unvalidated!
5Belarus BY15%64%95%Covered; intervals unvalidatedok
6Cuba CU15%64%95%Covered; intervals unvalidatedok
7Azerbaijan AZ22%65%95%Covered; intervals unvalidatedok
8Tajikistan TJ22%59%95%Covered; intervals unvalidatedok
9Kazakhstan KZ22%65%95%Covered; intervals unvalidatedok
10Uzbekistan UZ22%65%95%Covered; intervals unvalidatedok
11Afghanistan AF22%59%95%Covered; intervals unvalidatedok
12Bangladesh BD22%65%95%Covered; intervals unvalidatedok
13India IN22%49%95%Covered; intervals unvalidatedok
14Turkey TR36%59%95%Covered; intervals unvalidatedok
15Saudi Arabia SA22%59%95%Covered; intervals unvalidatedok
16Egypt EG33%68%70%Covered; intervals unvalidatedok
17Ethiopia ET22%65%70%Covered; intervals unvalidatedok
18Myanmar MM33%68%70%Covered; intervals unvalidatedok
19Turkmenistan TM15%36%70%Covered; intervals unvalidatedok
20Pakistan PK33%68%70%Covered; intervals unvalidatedok

Ranks compare model scores only within this watched panel. Covered means measurement evidence exists; scores are model calculations, not observed shutdowns. The API generated_at time is calculation time, not source observation time.

Cons. = numerical order (P(1d) ≤ P(7d) ≤ P(30d)). Interval bands are withheld pending validation.

1d LOCO skill
AUC 0.907
F1 0.448
7d LOCO skill
AUC 0.883
F1 0.660
30d LOCO skill
AUC 0.845
F1 0.701

Leave-one-country-out results cover the watched panel only. AUC can overstate practical skill; read F1 alongside it. Rare positive days and overlapping labels limit onset claims.

Why three horizons?

A single 7-day number is what most journalists ask for, but it hides two important things: the current-regime risk level tomorrow, and the operational ceiling over the next month. We train three independent XGBoost models and publish all three with LOCO AUC numbers and per-horizon SHAP attribution. Note: each horizon uses the same target_Nday sliding-window label as the production v1 forecast, so the LOCO AUCs are inflated by label autocorrelation the same way — these are current-regime risk signals, not shutdown-onset predictors. See the onset-skill finding.

The monotonicity check is a free honesty signal: longer windows must contain shorter windows, so if P(1d) > P(30d) the three models are disagreeing and you should treat the headline numbers with caution. See /atlas/models for the full registry, or /methodology for the full pipeline.

About this view

Three model horizons per country — each with its own XGBoost model and feature attribution. Interval coverage is unvalidated, so no interval is shown. Click a country for per-horizon detail.

Forecast scope & limitations