Voidly
All findingsPublished research

Curated shutdown source-date comparison: 20 historical events

Archive correction (2026-10-08): This hand-curated May 2026 table juxtaposes dates and public references for 20 landmark shutdown events; it is not a validated live lead/lag benchmark. Nineteen rows are marked as historical OONI backfills, so their first_seen dates are not Voidly live detection or…

Read the full published summary

Archive correction (2026-10-08): This hand-curated May 2026 table juxtaposes dates and public references for 20 landmark shutdown events; it is not a validated live lead/lag benchmark. Nineteen rows are marked as historical OONI backfills, so their first_seen dates are not Voidly live detection or publication timestamps. The Lebanon row pairs a published Sentinel row labeled 2026-04-17 with an IODA North Lebanon BGP observation on 2026-04-24. Original forecast issuance and matching model lineage are unverified, so this date comparison does not establish advance warning or a scored win. The Bangladesh row assigns an OONI-derived 2024-07-23 date and a NetBlocks 2024-07-18 public reference; that historical difference is not a live Voidly lag. Competitor reference dates do not establish internal detection time. The selected set is not a statistical sample; inspect the original source URLs and per-event caveats. A public empirical forecast scorecard remains on hold pending issue-time, model-version, target, and full-window qualification.

Archive correction — 2026-10-08. This May 2026 article compares attributed dates and source links for 20 hand-picked shutdown events. It does not establish that Voidly detected or published first in any of them.

Journalists can use the table to find public references from Cloudflare Radar, Access Now, and NetBlocks alongside archived Voidly records. Each underlying source should be checked for the event and timestamp being cited.

What the dates mean

Nineteen rows in the source table are marked voidly_backfilled: true. Their first_seen dates come from later ingestion of historical OONI evidence. They are not timestamps of live Voidly alerts or public Voidly reports. Competing dates were entered from public references. Subtracting these different kinds of dates does not measure detection or publication speed.

The remaining Lebanon row compares a published Sentinel row labeled 2026-04-17 with IODA's North Lebanon BGP connectivity observation on 2026-04-24. The original forecast issuance and matching model lineage are unverified. The date interval is an attributed historical comparison, not verified advance warning or a scored competitive win. The Lebanon case study explains the row and its provenance limits.

What this set can show

  • 20 selected events: this is a curated reference list, not a statistical sample of Voidly incidents or forecasts.
  • 19 historical backfills: archived OONI dates can help locate event evidence, but cannot establish when Voidly would have alerted in production.
  • One Lebanon historical comparison: its archived row label and outage date cannot be used for a live lead/lag count without original issuance and model-version receipts.
  • Bangladesh illustrates the timestamp mismatch: the source table assigns a 2024-07-23 OONI-derived date and a 2024-07-18 NetBlocks public reference. The five-day difference describes those attributed dates; it does not measure a live Voidly lag.

Some older rows lack a cited Cloudflare Radar event URL. A missing reference does not show when Cloudflare detected an event. Public references from any organization also do not reveal its internal detection time.

Using the source table

The curated API table provides per-event links and caveats. Its outcome labels or date arithmetic should not be cited as validated live lead/lag performance while provenance is being corrected. Check the source URLs and generated_at value before citing a historical date.

A public empirical forecast scorecard is on hold pending verification of issue times, model versions, matching targets, and complete outcome windows. This curated table cannot substitute for that scorecard.

The source record.

Original source links and labels. Linked responses may have changed since publication.