Normalize
We turn a full link into a clean website name. For example, a video link on YouTube becomes youtube.com.
How it works · version 2
We compare recent website checks by country and show the result in plain language.
You enter a website and choose a 7, 30, or 90 day window. Our server cleans up the address, checks recent evidence, and returns one report.
We turn a full link into a clean website name. For example, a video link on YouTube becomes youtube.com.
We group recorded website-connection tests by country and exact hostname. Known monitored sites include their explicitly listed main and www hosts; other subdomains remain separate. The displayed OK percentage is OK tests divided by all recorded tests, not the percentage of users with access.
Regional summaries add context. Different summaries and scheduled updates may reuse the same measurements; agreement does not establish independent confirmation.
The page gives a plain answer first, then lets you open the evidence, country rows, and optional technical failure details.
The headline, map, rankings, and alerts pages read the same saved hourly update. Opening those pages does not start another round of checks.
The scheduled overview reads public measurements for a bounded watchlist. It records its evidence window; opening this preview does not run a new probe inside China or another country.
Legacy incident history came from repeated score snapshots. A repeated snapshot may contain the same underlying checks, so this is not proof of independent networks or new measurements. New outcome-count reports do not confirm or resolve that legacy history; a rolling window can change without a new test.
The homepage, world map, rankings, and alerts use the same saved results, so their answers stay in sync.
Each headline includes its place, number of checks, and updated time.
The default period is 30 completed UTC days: the start midnight is included and the end midnight is excluded. The fetch timestamp is when we retrieved the report, not when every test ran. Repeated tests may come from the same network.
Worldwide rates are weighted by test counts, never averaged country percentages. Tests without a recorded country stay in the worldwide total. Cross-site rates are withheld when host scopes overlap or source windows differ.
Every page uses one shared evidence policy. Repeatedly downloading the same data does not make it independent confirmation.
Country summaries require at least 3 recorded blocking tests and blocking in at least half of all recorded tests before showing a sampled restriction. Access observed requires at least 3 OK tests and OK results in at least 80% of all recorded tests. Other results stay uncertain or need review. These are conservative display rules, not statistical confidence levels or estimates of access for the population. Separately dated, validated independent evidence may support its own conclusion.
Decisions use the exact integer counts before rounding. Displayed percentages describe recorded tests, with at most two decimal places; a nonzero share below 0.01% stays visible as <0.01%, and an incomplete share above 99.99% is shown as >99.99%. Only exact zero and complete fractions are shown as 0% and 100%.
Current reports show separate OK, unconfirmed anomaly, confirmed-blocking test, failed-test and unknown counts. These categories sum to the total. Failures and unknowns remain in the OK-percentage denominator. Confirmed blocking refers only to those recorded tests, not a countrywide result. Older LONI reports used a 0–1 upper-tail interference score; it is not the percentage of failed checks and cannot be converted into an OK percentage.
An OK result means access was observed in that test. A country with a few OK tests and many anomalies is not classified as generally accessible: the rate, counts and majority-outcome description remain visible. We do not apply a confidence cutoff or infer population availability from a small, non-random sample.
Missing or outdated evidence cannot establish a current answer. A small sample is displayed as a small sample; even one OK test cannot establish general availability.
Different networks, dates, and test methods can disagree. The page says so instead of averaging the conflict away.
Saved results keep popular reports quick and avoid repeating the same work.
Successful reports can be cached at the nearby edge for up to one hour, bounded by the remaining evidence lifetime. Every report shows when its evidence was checked; an outdated snapshot is not a current result.
A small list is refreshed every hour. Saved reports can be retained for up to 24 hours, but evidence older than 3 hours is shown as outdated or unknown. A fresh technical-detail check does not reuse an older result.
The map, rankings, and alerts use the same saved hourly update and show when it was made.
Normal checks and fresh technical checks have separate safety limits. If the service is busy, it asks you to wait one minute before trying again.
Coverage and ordinary network failures can affect results.
Our measurement data separates successful, anomalous, confirmed-blocking, failed and unknown measurements. A zero in the confirmed category does not mean the site is reachable. Connection failures and anomalies stay inconclusive unless the evidence supports blocking.
We validate the counts and dates directly, rather than relying on a separate country-summary field. Independent GreatFire observations can support a China result even when other network measurements have no confirmed records. GreatFire conclusions and network measurements are shown separately, with their own dates and windows. They are never added into one blocking percentage.
The report can use a dated GreatFire source snapshot for exact HTTPS websites. It is excluded after seven days without a refresh; an observation must be no older than 30 days and have at least three conclusions. GreatFire’s window can be up to 90 days, independently of the network measurement window you select. Mixed evidence is explained in the result.
GFWList is a maintained circumvention list, not a live test. A matching host rule adds context; it cannot by itself create a current blocked verdict. Derived aggregators do not count as an extra independent source.
GreatFire methodology · Website connectivity methodology · GFWList
Allow Google Analytics and Cloudflare Web Analytics on our information pages to help us understand readership? Google uses analytics cookies. Website checks and report pages are excluded.
Privacy details. You can change your choice at any time.