When we change how we test or report, or find that a published figure was wrong, it goes here with the date.
Last updated: 8 November 2026
A change is a new rule, test setting or feature. A correction means something we published was wrong or misleading. The rules themselves are on the methodology page.
Headline speeds and pings are now medians of the last 30 days of successful tests instead of means, with the middle half of results (25th to 75th percentile) shown on provider pages. A median is barely moved by a few unusually fast or slow tests. Rankings and head-to-head wins use the medians too, so some figures and a few location winners shifted slightly. Means remain in our raw data.
Added jitter to the comparison table and server tables, and a “last tested” time on every server row. Locations with no successful test in 48 hours are marked stale.
Published the data rules on the methodology page.
Comparison rows measured for only one provider named that provider “best here”. They now show “Insufficient data” and have no winner.
“As low as” ping figures quoted a single best test. They now quote the closest location's 30-day average.
Streaming answers named specific services. We measure speed, not whether a service works through a VPN, so they now say only whether the bandwidth is enough for 4K.
Provider and comparison pages now show a note when a provider's 7-day pass rate is below 90%.
Most NordVPN test connections had been failing in the weeks before 4 November. From 4 November more than 97% of NordVPN tests complete each day.
Verdicts and pass rates on provider pages now use the last 7 days.
Some pages used all-time results and others the last 30 days, so the same location could show two different speeds and either provider could appear to win. All rankings, averages and head-to-head counts now use the last 30 days, and locations with fewer than 5 successful tests are left out.
The methodology page gave the wrong number of locations tested per provider. It now shows the live count.