Voidly
All findingsPublished research

Pre-shutdown network signal detector: do BGP, TLS-reset and new-ASN precursors lead a blocking event?

A user-visible shutdown is the end of a process — by the time domains stop loading, the routing and filtering infrastructure has often already moved. This finding ships a per-country composite "pre-shutdown signal score" built from three technical precursors that TEND to precede user-visible…

Read the full published summary

A user-visible shutdown is the end of a process — by the time domains stop loading, the routing and filtering infrastructure has often already moved. This finding ships a per-country composite "pre-shutdown signal score" built from three technical precursors that TEND to precede user-visible blocking, computed daily over the trailing 90 days. bgp_signal: IODA BGP-datasource critical/warning alert counts (route withdrawals), normalized to that country's own 90-day baseline. tls_fail_spike: TLS / TCP-reset interference evidence rows (http-blocking-tcp-reset etc.) against a trailing 7-day mean — a DPI box being switched on shows here before the blockpage does. new_asn_signal: count of distinct ASNs lighting up with anomaly evidence today that were not seen the prior week — blocking infrastructure spreading. The composite is a 0.5/0.3/0.2 weighted sum, then z-scored within each country so every country is on a comparable scale. We back-tested the composite against 2,240 evaluable confirmed shutdowns (incident_type censorship/mixed/disruption, severity critical) in the window: for each shutdown, did the composite cross z>=1.5 in the 72 hours before the shutdown timestamp? 522 of 2,240 did — a 23.3% true-positive rate, with a median lead time of 31.3 hours (mean 39.9h). The honest other side: of 585 country-days where the signal fired, 180 had no shutdown in the following 72h — a 30.8% false-positive rate. So the signal is real but partial: it clears the promote-floor (>=3 historical shutdowns with >=1h lead — we got 522) and ships as promoted:true, but it catches under a quarter of shutdowns and fires falsely about a third of the time it fires. The reason is structural and stated in every API response: many shutdowns are SUDDEN with no measurable precursor, IODA BGP has a ~6h ingest lag that hides any sub-6h lead, and an elevated composite reflects anomaly activity that frequently resolves without a shutdown. Use it as a supporting signal alongside the 7-day forecast, never as a standalone trigger. Live at GET /v1/sentinel/pre-shutdown-signal/{info,leaderboard,<cc>}. Built by scripts/build-pre-shutdown-signals.py.

A user-visible internet shutdown is the end of a process, not the start of one. By the time domains stop loading and users notice, the routing and filtering infrastructure behind the shutdown has often already moved — routes withdrawn, DPI boxes switched on, blocking spreading to new networks. The pre-shutdown network signal detector asks a narrow, falsifiable question: can we see those technical precursors in our own evidence stream before the blocking becomes visible to users?

We build a per-country composite "pre-shutdown signal score" from three precursor signals, computed daily over the trailing 90 days:

  • bgp_signal — IODA BGP-datasource critical/warning alert counts (route withdrawals and announcements), normalized to that country's own 90-day daily baseline. A spike here means the country's networks are being re-routed.
  • tls_fail_spike — counts of TLS / TCP-reset interference evidence rows (http-blocking-tcp-reset and related signal types) measured against a trailing 7-day mean. When a deep-packet-inspection box is switched on, the connection resets show up here before any blockpage does.
  • new_asn_signal — the count of distinct ASNs appearing in anomaly evidence today that were not seen in the prior 7 days. New networks lighting up with anomalies is blocking infrastructure spreading across a country.

The composite is a weighted sum (0.5 BGP, 0.3 TLS, 0.2 new-ASN), then z-scored within each country over its own 90-day history so every country sits on a comparable scale. A score of z ≥ 1.5 is called "signal up."

The back-test. We took every confirmed shutdown in the window — 2,240 evaluable events (incident_type in censorship / mixed / disruption, severity critical) — and for each one asked: did the composite cross z ≥ 1.5 at any point in the 72 hours before the shutdown timestamp? 522 of 2,240 did — a 23.3% true-positive rate, with a median lead time of 31.3 hours (mean 39.9h). When the precursor exists, it tends to exist a day or more out.

The honest other side. A lead signal is only useful if it doesn't fire constantly. Of 585 country-days where the composite fired, 180 had no shutdown in the following 72 hours — a 30.8% false-positive rate. So the detector is real but partial: it catches under a quarter of shutdowns and, when it fires, is wrong about a third of the time. It clears our promote-floor (≥ 3 historical shutdowns with ≥ 1h lead — we got 522) and ships as "promoted": true, but "promoted" here means the signal is genuine and worth exposing, not that it is a reliable standalone trigger.

Why it only catches a quarter. The ceiling is structural, and we state it in every API response. Many shutdowns are sudden: a government issues an order, an operator flips a switch, and there is no measurable technical run-up at all — the detector cannot warn on those because there is nothing to see. IODA's BGP feed also carries a roughly 6-hour ingest lag, so any precursor with a real lead time shorter than ~6h is invisible to this back-test even when it physically exists. And the back-test crosses days, not hours, so reported lead times are day-resolution.

What this is, what this isn't. It is a supporting early-warning signal — a per-country composite that, when elevated, raises the prior that a shutdown is coming and can feed into the broader forecast stack alongside the 7-day shutdown forecast and the pre-protest GDELT correlator. It is NOT a standalone predictor: an elevated composite reflects raised BGP / TLS / new-ASN anomaly activity, which frequently resolves with no shutdown at all. The leaderboard surfaces which countries are elevated today; the per-country endpoint returns the full 90-day daily series so a journalist or analyst can see exactly what drove the score. Every response carries the caveats inline — signal up is not the same as shutdown imminent.

The source record.

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