7-day shutdown forecast · calibrated

AT — 7% current-regime risk

Calibrated via isotonic refit on 810 live outcomes (2026-05-20). See calibration drift → · AT country profile →

The Voidly Sentinel model scores AT at a peak 7.4% shutdown-risk level over the next 7 days, highest on day 7 (Sat, May 30). Average across the window: 5%. This is a current-regime risk reading, not a prediction that a new shutdown will start — see the scope note below.

Honest scope: this is a current-regime risk signal, not a shutdown-onset predictor. It reliably flags countries already in a sustained blocking regime; on the days a new shutdown actually begins its skill is at/below chance (transition-row AUC ~0.33). Why →

90% conformal interval: [0%, 42%] — empirical coverage 91.0% (target 90%).

Per-day risk

Dashed line: 50% threshold. Bars colored by risk band.

Per-day drivers

Date	Risk	Drivers
Sat, May 23	3% · Low	no drivers
Sun, May 24	4% · Low	no drivers
Mon, May 25	5% · Low	no drivers
Tue, May 26	5% · Low	no drivers
Wed, May 27	6% · Low	no drivers
Thu, May 28	7% · Low	no drivers
Fri, May 29	4% · Low	no drivers
Sat, May 30	7% · Low	no drivers

What drove the prediction

Top features by signed contribution to today's risk score. Positive (↑) features push the prediction higher; negative (↓) features push it lower.

↑recent_shutdown+0.048
↓risk_tier-0.030
↓gdelt_unrest_30d-0.026

Similar past event

The model found a similar event in the corpus. Reading the incident report tells you what “7% risk for AT” looked like when it actually happened.

AT-2026-0004critical

disruption in AT · Sun, May 3

Honest caveats

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 →

Calibration was refit on 2026-05-20 on 810 live outcomes — it's good for the 30 watched censoring countries but will drift if the world changes. Nightly recalibration runs at 04:30 UTC.
A 70% forecast doesn't mean a shutdown is guaranteed; a 20% forecast doesn't mean it's safe. The 90% conformal interval is the honest uncertainty band.
For long-form methodology see /methodology and the calibration refit finding.

Model version: v1@2026-05-21T16:48:35.555109

/at — full country profile with recent incidents + ISP breakdown
/atlas/elections — all 90-day election forecasts ranked by risk
/sentinel/calibration — live model-honesty dashboard + feature importance
/v1/sentinel/current_risk/AT — raw JSON

Per-day risk

Per-day drivers

What drove the prediction

Similar past event

Honest caveats

Related