voidly
7-day shutdown forecast · calibrated

Bangladesh5% peak risk

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

The Voidly Sentinel model forecasts a peak shutdown probability of 4.7% for Bangladesh over the next 7 days, with the highest risk on day 7 (Thu, May 28). Average risk across the window: 3%.

90% conformal interval: [0%, 40%]empirical coverage 90.4% (target 90%).

Per-day risk

0%25%50%75%100%1%Thu2%Fri1%Sat2%Sun3%Mon4%Tue2%Wed5%Thu

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

Per-day drivers

DateRiskDrivers
Thu, May 211% · Lowno drivers
Fri, May 222% · Lowno drivers
Sat, May 231% · Lowno drivers
Sun, May 242% · Lowno drivers
Mon, May 253% · Lowno drivers
Tue, May 264% · Lowno drivers
Wed, May 272% · Lowno drivers
Thu, May 285% · Lowno 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.

  • gdelt_unrest_30d-0.018
  • block_rate_roll7_mean-0.011
  • recent_shutdown+0.010

Similar past event

The model found a similar event in the corpus. Reading the incident report tells you what “5% risk for Bangladesh” looked like when it actually happened.

BD-2026-0112warning

censorship in BD · Fri, May 8

Recent evidence

Live measurements from upstream networks supporting this forecast. Click any link to view the raw data on the source.

Honest caveats

  • The model's base AUC on a time-based holdout is 0.50 (random on novel events). LOCO median AUC is 0.91. The calibrated number you see above is the most-honest operational figure but it's NOT a guarantee.
  • 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@2026-05-21T04:20:38.966951

Related