Journalists almost never cite a single source on an internet shutdown. They triangulate — usually Cloudflare Radar (CDN traffic drops), Access Now's #KeepItOn tracker (civil society advocacy), and NetBlocks (BGP / measurement reports). If Voidly Atlas is going to be useful to that audience, we owe a side-by-side answer: on the same events, did we publish before, with, or after the others?

This page is that answer for 20 hand-picked landmark shutdown events from 2019 through 2026. It is intentionally limited in scope, intentionally honest about its limitations, and not the number we lead with internally.

What "lead time" means here

Lead/lag is measured publication-vs-publication. That is: the date the event first lands in Voidly's incidents table compared against the earliest date the same event is referenced in a public URL at Cloudflare Radar, Access Now, or NetBlocks. None of these comparisons say anything about internal detection time at competing orgs — Cloudflare and NetBlocks both run private pipelines that almost certainly catch events before they publish about them.

The headline numbers

• 20 landmark events benchmarked (2019 to 2026).
• 1 event where Voidly led: Lebanon BGP outage, 2026-04-24. Our forecast threshold cross fired on 2026-04-17 — 7 days before Cloudflare Radar's public dashboard registered the withdrawal. This is the only entry in the table that represents a true live-pipeline detection.
• 0 events where Voidly lagged on a like-for-like basis. We have one entry (Bangladesh quota protests, 2024-07-18) where our incident landed 5 days after NetBlocks' BGP-tap report — we document that honestly.
• 19 events are flagged voidly_backfilled: true. That means our DB row was created by ingesting the historical OONI archive after the fact. The first_seen on those rows is not a real-time detection — it is the day OONI measurements first showed anomalies, which our live pipeline didn't exist for. Those entries cannot honestly support a lead-time claim.

Why we publish this with all the caveats

The temptation when you build a forecasting/detection product is to cherry-pick the wins and bury the rest. This is the opposite of what journalists need from a source: they need to know where the source has blind spots so they can compensate.

The honest reading of this table is:

• Voidly's forecast model has exactly one event where the live pipeline was elevated before the event landed (Lebanon, threshold crossed 6.5 days prior). One — and even that is base-rate-consistent for a chronically-unstable network, not proof of timing skill (within-country timing is near chance). The corresponding case study at /atlas/case-studies/lebanon-2026-04-24-isp-outage shows the prediction trace, threshold cross, and SHAP drivers.
• For wholesale BGP-shutdown events (Bangladesh 2024, Tigray 2021), NetBlocks' BGP-tap pipeline is faster than OONI's measurement ingestion — if you need the earliest possible alert on that class of event, cite NetBlocks first.
• For domain-level / TLS-SNI / DNS-poison censorship (which is most of what OONI catches), Voidly is competitive on publication day with NetBlocks and Access Now, and almost always ahead of Cloudflare Radar (which only sees CDN-side traffic drops, not protocol-specific blocking).
• Pre-2021 events: Cloudflare Radar's public event browser didn't exist, so we can't cite a Radar URL for the Kashmir shutdown (2019-08), Bloody November (2019-11), or the June Sudan massacre (2019-06). We mark those as cloudflare_url: null rather than inventing one.

What this is NOT

This is not a statistical sample of Voidly's accuracy. It is a curated list of 20 landmark events that journalists routinely cite. For the non-cherry-picked, live, rolling 30-day production track record across every Voidly forecast model, see /v1/atlas/prediction-track-record .

The drilldown endpoint /v1/atlas/competitive-benchmark supports ?outcome=voidly_led|voidly_lagged|tie|backfilled|no-comparison and ?country=<ISO-2> filters. Each event has its own URL at /v1/atlas/competitive-benchmark/{event_id} with the full per-event payload + a per-event honest_caveats array.

How to cite this

If you reproduce any of these comparisons in a story, please link to the underlying URLs — both ours and the competitor's. The sidecar at /opt/voidly-ai/ml-deploy/competitive_benchmark_v1.json ships every URL we reference, and the generated_at timestamp lets you anchor your citation.