Transparency Report
See for
yourself.
How we operate, what we collect, and what we don’t.
The data boundary.
Collection varies by product and by whether you use an account or checkout. Read the full privacy policy for product scope and exceptions.
Data Collected
- Connection success/failure rates
- Route performance metrics
- Node health telemetry
- Censorship event detection
- Country-level statistics (aggregate)
- Optional account contact details and wallet addresses
- Marketplace listings, job terms, orders, submissions, and payment evidence
- Security and abuse records when a request triggers review
Not collected in these flows
- Browsing-history logs kept by VPN nodes
- Per-user DNS-query histories in VPN measurements
- Card or bank account numbers entered in Voidpay wallet checkout
Voidpay wallet checkout asks you to approve a wallet payment or supported payment link, not to enter card or bank account numbers. Account identities, wallet addresses, order records, and payment evidence can still identify people; public-chain transactions can be visible to others; network privacy varies.
The privacy policy also discloses request and abuse-log exceptions, retention, and product-specific collection. These exclusions are not a promise that every Voidly service stores no personal data.
Published Data
Evidence you
can inspect.
Explore the underlying records, formats and methods.
- Country censorship scores and risk tiers
- Verified incident reports with evidence chains
- Platform and ISP blocking metrics (aggregated)
- Historical archive (10 years, 130+ countries)
- ML model performance metrics
Licensing and scope
Public censorship datasets explicitly marked CC BY 4.0 use that license.
Use the data hub’s source and licensing details to distinguish Voidly’s original outputs from upstream materials with their own terms. A report in the dataset is not automatically a confirmed censorship finding.
A number needs a source.
Keep published snapshots, counters and current service checks distinct.
Reported figures / Sources and datesInspect the numbers
Network Stats
Data as of March 2026
The published report carries that date. Its fixed uptime and latency figures are reproduced below as report values, not a current service check. Counters and model snapshots have separate sources.
- Nodes
- Read the network source ↗The original configured node figure remains in the original report; it is not a live fleet count.
- Reported coverage
- Global (Europe, Americas, Asia, Oceania)Published report description; not a new coverage measurement.
- Reported uptime
- 99.8% (30d rolling)Original report value. Its “Data as of March 2026” label supplies no measurement source. Check current services.
- Reported average latency
- 47msOriginal report value under the March 2026 label; not verified as current.
- Countries monitored
- 130Bundled country-index coverage; not the number of countries contributing to every dataset.
- Network measurements
- 38,779,164Site counter: bundled OONI measurements plus available API updates. A corpus total, not a live-service check.
- Incident records
- UnavailableWaiting for a source response. Missing or invalid counts remain unavailable. The original report calls this total “Verified Incidents”; it also includes suspected and disruption records.
- Citable censorship records
- UnavailableThe source’s separate citable-censorship field; missing or inconsistent values remain unavailable. Inspect incident statistics.
- Censorship classifier
- 0.71 F1 (internal eval)LOCO mean across 127 countries — 0.63 on the 61 countries with n≥30. The 0.87 median is inflated by 46 small-sample perfect scores. 794 of 1,116 positive training labels (71.1%) are country-days whose only incident is an IODA `disruption` row — network outages, not confirmed censorship. The same class was excluded from forecast labels in 2026-05 as ~94% noise. So this model is substantially trained to detect DISRUPTION, and its F1 should be read as such. Corpus is frozen at 2026-05-21; relabelling is a pending decision, not an oversight.. Snapshot generated 2026-10-03T22:42:07.537Z. Model source.
- Shutdown forecast
- 0.86 AUC (internal eval)LOCO median on the training holdout; time-based AUC is 0.55. Rolling-30d precision is 0.07; the endpoint flags the model DEGRADED. Read the evaluation and dates.
Open Source / Published tools
Read the work.
Published tools, probe software, and specifications at: github.com/voidly-ai
Review published tools and methodology. Verify claims against our open data.
Privacy requires transparency. Trust requires verification.
Published code can have different license terms. Check the repository and Terms of Service.
Contact
- Security issues
- security@voidly.ai
- Privacy concerns
- privacy@voidly.ai
- General inquiries
- hello@voidly.ai