How we measure the mirrors
Every site in this space publishes latency numbers. Almost none of them say how the numbers were produced, so here is ours.
Published reachability figures are worth exactly as much as the methodology behind them, and most sites in this space publish a number with no methodology at all. Some of those numbers are invented. This page is the disclosure that lets you judge ours.
What the probe does
A probe builds a fresh Tor circuit, requests the login screen of a mirror, and records whether it completed and how long it took end to end. It runs on an interval, from a single vantage point, over the ordinary Tor network with no special routing. There is nothing clever about it, which is the point.
Latency is reported as the median across recent probes rather than the mean, because the distribution has a long tail and the mean is dominated by a handful of bad circuits. The tail is real and matters to your experience, but a single number that swings wildly is less useful than a stable one.
What the numbers do not mean
- Not your latency. Our circuit is not your circuit. A mirror we measure at a good number can be unusable for you at the same moment because of where your path happened to land, and the reverse happens too.
- Not an uptime guarantee. A probe that succeeded says the mirror answered us once. It says nothing about the minute after.
- Not a health measure of the platform. Most of what these numbers move on is Tor network conditions rather than anything about the market.
- Not comparable across sites. Another site's figures come from a different vantage point with a different method. Comparing their number to ours is meaningless.
Where it is unreliable
Single vantage point is the honest weakness. Everything we publish is what one machine on one network sees, and a problem local to that machine looks identical to a problem at the mirror. We flag figures as unreliable when all three mirrors degrade at once, because in practice that pattern is usually us rather than them.
The other weakness is that a probe only fetches the login screen. It confirms that something answered at that address and rendered. It does not confirm that the thing answering is the real platform, which is a question no probe can settle and only a signature can.
Why publish at all
Because the alternative is readers guessing. Somebody whose connection to one mirror is failing wants to know whether it is them or the address, and a number produced by a disclosed method from an independent vantage point answers that better than nothing does. It just has to come with the caveats, and the caveats are the reason this page exists rather than a chart with no text under it.
Verified working Nexus Market mirrors
Three v3 onion addresses currently serving the production market, signed under PGP fingerprint 0x7F2A0A9D. Use the Copy buttons.