Voidly open data
The internet.
On the record.
Explore censorship. Trace the evidence.
Take the data into your own work.
A closer look.
A clearer picture.
Start with a country or an incident.
Follow the evidence from there.
Find an incident
Documented events. Sources you can cite.
Explore a country
Compare scores, blocking and network context.
Take it with you.
From a spreadsheet to a research pipeline.
Choose the format that fits.
For your code
Structured country scores and context.
Open JSONFor your spreadsheet
A familiar format for your analysis.
Open CSVFor the longer view
Historical measurements on Hugging Face.
Open archiveVoidly-original data: CC BY 4.0.
OONI raw archive: CC BY-NC-SA 4.0.
The record.
In your workflow.
Query the API. Connect your assistant.
Bring evidence into what you make.
curl \
https://api.voidly.ai/data/censorship-index.jsonFetch the public JSON file. No API key is required for this download.
Open to inspection.
The sources, terms and detail behind the data.
Sources, licenses and citation
The incident and evidence records draw on OONI, IODA, CensoredPlanet and Voidly probes, with test-list context from Citizen Lab. Source coverage and timestamps differ; a measurement count is not an incident count.
Voidly-original incidents, scores and annotations are CC BY 4.0. Upstream measurements keep their own licenses. OONI raw data and Citizen Lab test lists carry CC BY-NC-SA 4.0; upstream-derived raw subsets retain those terms.
When citing a result, include its stable URL, source, date and relevant limitations. Open the citation tools.
Data distributions and endpoints
Country-index files include country scores, levels, sample counts, blocked services and scoring context. ISP detail is available separately. The archive includes dates, countries and aggregated measurement outcomes. Inspect each response for the current schema.
- Country index fields
country, score, level, trend, samples, blocked_count, block_rate, isps[]- Historical archive fields
date, country, anomaly_count, ok_count, confirmed_count, failure_count, measurement_count, anomaly_rate- ISP detail fields
country, asn, name, block_rate, threat_level, blocked_domains[]- Forecast response fields
forecast[], summary (max_risk, drivers), confidence
GET /data/meta- Versions and endpoint discovery
GET /data/censorship-index.json- Full public country index
GET /v1/vpn-accessibility- VPN reachability observations
GET /data/incidents- Classified incident records
GET /data/incidents/delta?since=…- Changes after a timestamp
GET /data/country/:code- A single country
GET /data/methodology- Machine-readable methodology
POST /hydra/v1/predict- Classifier interface; see model evaluation
GET /v1/probe/domain/{domain}- Probe observations for a domain
Specialist data and methods
Explore forecasts, platform scores, VPN reachability, service checks, block-page fingerprints and censorship-intent analysis. Model scores are estimates; coverage, confidence and validation belong beside every result.
Semantic incident search returns related records for a natural-language query. Open the existing search example.
Country rankings do not predict the exact day of a shutdown. Read the methodology and model evaluations before interpreting model output.
Block-page attribution
Fingerprints distinguish government-mandated block pages from ISP and filtering-vendor pages. The reference catalog includes Roskomnadzor (Russia), Iran TIC, Great Firewall (China), BTK (Turkey), CITC (Saudi Arabia), PTA (Pakistan), Kominfo (Indonesia), Fortinet and Netsweeper. An attribution is a source classification; inspect the evidence and date before treating it as a current finding.
Probe records can include dns-poison, tcp-reset and blockpage methods, alongside an attributed blocking entity. Domain coverage is a rotating subset of the Citizen Lab test list. The network and statistics endpoints above expose their own current scope.
Specialist endpoint reference
- Platform scores
/v1/platforms/scores · /v1/platforms/{platform} · /v1/platforms/country/{code}- Censorship intent
/v1/measurement/summary · /intent-profile · /category-leaders · /co-blocking · /network-depth- Service accessibility
/v1/accessibility/check · /v1/accessibility/batch · /v1/accessibility/service/{name}- Semantic incident search
GET/POST /v1/incidents/similar · country, incident_type, min_confidence filters
Semantic-search request example:
curl "https://intelligence.voidly.ai:8443/v1/incidents/similar?query=WhatsApp+blocked+in+Iran&k=5"VPN reachability reference tests include Voidly nodes, NordVPN, ProtonVPN and Mullvad (WireGuard UDP:51820). A listed test target does not prove that provider works on every network.
Packages, probes and embeds
Probe client
Read participation and privacy details first.
Probe overview and risksPython packageDocker imageAnother way in.
The public API is also available over Tor. Read the access details and limits.