Voidly
All findingsPublished research

Voidly Score: one continuous 0-100 daily number for "how censored is this country today"

Journalists asking Voidly Atlas "how censored is Iran today" kept hitting seven separate numbers — a supervised classifier probability, a 7-day forecast, a DBSCAN anomaly score, a source-agreement rate, an incident count, a blocked-domain tally, a mobile-messenger probe skew. A headline needs one.…

Read the full published summary

Journalists asking Voidly Atlas "how censored is Iran today" kept hitting seven separate numbers — a supervised classifier probability, a 7-day forecast, a DBSCAN anomaly score, a source-agreement rate, an incident count, a blocked-domain tally, a mobile-messenger probe skew. A headline needs one. The Voidly Score is that one number: a continuous 0-100 daily index per country, built so a newsroom can write "Iran censorship intensity 78 today, up from 62 yesterday." It is deliberately different from Atlas Score v2 (an A-F structural rating dominated by a 50% base-rate term that captures chronic blockers like CN/KP even on a quiet week). Voidly Score answers "what is happening today" and moves. The composite: 30% v3.3 classifier probability + 20% 7-day forecast max risk + 15% DBSCAN anomaly (normalized, saturation cap 3.0) + 10% cross-source agreement + 10% log-scaled 24h incident rate (censorship/mixed weighted 3x, disruption 0.3x) + 10% log-scaled 30d unique-blocked-domain count + 5% mobile-messenger probe skew. Each component is normalized to [0,1], weighted, scaled to 100, then smoothed with a 3-day EMA (alpha=0.5) so a single noisy probe day cannot fake a 15-point headline swing. The endpoint returns raw_score, smoothed_score (the headline number), and delta_vs_yesterday. First run 21 May 2026 scored 30 watched countries: top five EG 42.4 (anomaly-led), PK 36.3 (forecast-led), VE 35.0 (classifier-led), IN 32.0 (anomaly-led), UZ 29.8 (forecast-led) — no single signal dominates the board, which is the point of a composite. Honest caveats baked into every response: weights are hand-tuned not learned (no ground-truth "intensity" label exists); the seven components are correlated (forecast leans on the same OONI block-rate the classifier uses) so the sum is not a clean information-theoretic average; "intensity" is editorial framing — the metric is a censorship-RISK-WEIGHTED average across live signals, not a direct measurement of how blocked any user is; mobile skew is approximated from OONI test_name= URL params and gated to >=10 probes; the EMA needs ~3 days to warm up; sparse-data countries (low OONI coverage, no incidents, no DBSCAN window) float near 0 not because they are uncensored but because the signals collapse there — use Atlas Score v2 with its risk-tier floor for chronic-blocker country pages. Live at GET /v1/atlas/voidly-score + /v1/atlas/voidly-score/{cc} + /v1/atlas/voidly-score/info. Built daily 04:35 UTC. The two scores are complements: Atlas Score v2 for "where is censorship the worst", Voidly Score for "what is the headline number today".

Journalists asking Voidly Atlas a simple question — "how censored is Iran today?" — kept hitting the same wall. We had a supervised classifier probability, a 7-day forecast, an unsupervised DBSCAN anomaly score, a source-agreement rate, an incident count, a blocked-domain tally, and a mobile-messenger probe skew. Seven numbers. A headline needs one.

The Voidly Score is that one number: a continuous 0–100 daily index per country, built so a newsroom can write "Iran censorship intensity 78 today, up from 62 yesterday." It is deliberately different from Atlas Score v2, which is an A–F structural rating dominated by a 50% base-rate term. Atlas Score v2 answers "where is censorship the worst" (chronic blockers like China and North Korea rank high even on a quiet week). Voidly Score answers "what is happening today" — it moves.

The composite

Each of the seven components is normalized to [0, 1], then combined with hand-tuned weights and multiplied by 100:

ComponentWeightWhat it captures
v3.3 classifier probability30%Supervised country-day censorship probability
7-day forecast max risk20%Forward-looking shutdown risk
DBSCAN anomaly (normalized)15%Unsupervised shape anomaly vs the country's own 45-day baseline
Cross-source agreement10%Fraction of country-days where ≥2 of OONI / IODA / CensoredPlanet / Voidly corroborate
Recent incident rate10%Log-scaled 24h incident count (censorship/mixed ×3, disruption ×0.3)
Domain diversity blocked10%Log-scaled count of unique blocked domains, last 30 days
Mobile blocking skew5%Share of OONI probes targeting messenger apps (Signal / WhatsApp / Telegram / Messenger)

The raw composite is then smoothed with a 3-day exponential moving average (α=0.5) so a single noisy probe day does not produce a fake 15-point headline swing. The endpoint returns both raw_score (today's component sum) and smoothed_score (what to put in the headline), plus delta_vs_yesterday.

First run — 21 May 2026

Thirty watched countries scored. Top five by smoothed score:

CountryVoidly ScoreBiggest contributors
EG Egypt42.4DBSCAN anomaly 11.5, forecast 10.6, agreement 6.2
PK Pakistan36.3Forecast 19.0 (saturated), agreement 6.9, domain diversity 6.1
VE Venezuela35.0Classifier 13.0, agreement 7.2, domain diversity 6.1
IN India32.0DBSCAN anomaly 15.0 (saturated), domain diversity 6.7
UZ Uzbekistan29.8Forecast 19.0 (saturated), incident rate 4.3

Notice the components are doing genuinely different work: Egypt's score is anomaly-led, Pakistan and Uzbekistan are forecast-led, Venezuela is classifier-led, India is anomaly-led. No single signal dominates the top of the board — which is the point of a composite.

The honest caveats

  • The weights are hand-tuned, not learned. 30/20/15/10/10/10/5 is an editorial judgement about which signals deserve the most headline weight. It is not the output of an end-to-end optimization against a ground-truth "intensity" label — no such label exists.
  • The seven components are correlated. The forecast leans on the same OONI block-rate that feeds the classifier; the DBSCAN anomaly and the incident rate both spike on the same bad days. The weighted sum is not a clean information-theoretic average of independent signals. A reader should treat Voidly Score as a risk-weighted blend, not seven orthogonal measurements added up.
  • "Intensity" is editorial framing. The metric is a censorship-risk-weighted average across the signals Voidly Atlas currently runs. It is not a direct measurement of how blocked any specific user is. A score of 42 does not mean "42% of the internet is down."
  • Mobile blocking skew is an approximation. It is derived from the test_name= parameter in OONI Explorer URLs and gated to countries with at least 10 probes in the last 7 days, so a country with thin OONI coverage contributes 0 rather than a phantom boost.
  • The EMA needs ~3 days to warm up. On the first run the smoothed score equals the raw score (cold start) and delta_vs_yesterday is null. Deltas stabilize after day 3.
  • Sparse-data countries float low. A country with low OONI coverage, no recent incidents, and no DBSCAN window will score near 0 — not because it is uncensored, but because the signals collapse there. North Korea is the canonical example. Use Atlas Score v2 (with its risk-tier floor) for chronic-blocker country pages.

How to use it

Pull GET /v1/atlas/voidly-score for the full ranked board, or GET /v1/atlas/voidly-score/{cc} for one country with the full per-component point breakdown. The smoothed_score is the headline number; delta_vs_yesterday is the movement. Rebuilt every day at 04:35 UTC. For "where is censorship the worst" pages, reach for Atlas Score v2 instead — the two scores are complements, not rivals.

The source record.

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