7-day shutdown-risk estimate · validation pending

ML — 46% risk

A seven-day view of the current regime, with drivers and supporting evidence. This is not a shutdown-onset prediction.

Per-day risk

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

Reading this forecast & citation

The Voidly Sentinel model scores ML at a peak 45.6% shutdown-risk level over the next 7 days, highest on day 7 (Thu, Nov 12). Average across the window: 2%. This is a current-regime risk reading, not a prediction that a new shutdown will start — see the scope note below.

May 2026 legacy-model audit: the audited forecast mainly tracked existing blocking regimes, not new shutdown onsets. On the days a new shutdown began, its transition-row AUC was ~0.33. This historical result is not a current production scorecard. Why →

Per-day drivers

Date	Risk	Drivers
Thu, Nov 5	1% · Low	no drivers
Fri, Nov 6	2% · Low	no drivers
Sat, Nov 7	3% · Low	no drivers
Sun, Nov 8	1% · Low	no drivers
Mon, Nov 9	1% · Low	no drivers
Tue, Nov 10	1% · Low	no drivers
Wed, Nov 11	2% · Low	no drivers
Thu, Nov 12	5% · 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.

↓week_of_year-0.069
↓block_rate_roll30_mean-0.057
↓block_rate_roll7_std-0.027

Similar past event

The model found a similar event in the corpus. Reading the incident report tells you what “46% risk for ML” looked like when it actually happened.

ML-2026-0011critical

disruption in ML · Tue, Nov 3

Recent evidence

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

ooniinterferencehttps://explorer.ooni.org/search?probe_cc=ML&test_name=web_connectivity&since=2026-11-05&until=2026-11-062026-11-05
ooniinterferencehttps://explorer.ooni.org/search?probe_cc=ML&test_name=web_connectivity&since=2026-11-04&until=2026-11-052026-11-04
oonitor-blockinghttps://explorer.ooni.org/search?probe_cc=ML&test_name=tor&since=2026-11-04&until=2026-11-052026-11-04

Honest caveats

Legacy forecast: May 2026 audit

A May 2026 audit of the then-production 7-day forecast found that it mainly tracked existing blocking regimes and did not reliably predict new shutdown onsets. The figures below describe that legacy model, not a current production scorecard.

In that audit, a forward-temporal split (train on the past, test only on the strictly-future window) gave the legacy model AUC 0.589 on its raw label — near chance.
The audited label target_7day was 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 began (transitions), the legacy forecast scored AUC ~0.33 — below chance.

Read any current 7-day risk number as a source-reported signal. These historical audit figures do not validate current onset skill. Interval coverage is unvalidated and the bounds are withheld here; do not use this as an early-warning siren. Read the full honest re-evaluation →

The 2026-05-20 isotonic refit is historical. Its 810 archived pairs do not establish current calibration or onset skill.
A 70% forecast doesn't mean a shutdown is guaranteed; a 20% forecast doesn't mean it's safe. Source-reported interval fields have no verified empirical coverage here.
For long-form methodology see /methodology and the calibration refit finding.

Model version: v1@2026-11-01T02:18:10.429395

/ml — 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/ML — raw JSON

About this view

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