How Reachability Is Measured
Every status badge on this site comes from a documented probe, not a guess. This page explains how a sweep works, what each status word means, how the 30-day uptime figure is computed, and — just as important — what the monitor deliberately does not measure.
What Happens in One Sweep
A sweep is a fixed sequence run on a schedule. The same steps run every time so readings are comparable across days.
The probe pipeline
- From each probe region, build a circuit using that region's recommended transport.
- Request a known, fixed resource on each of the three onion mirrors.
- Record outcome and round-trip time, or a timeout, per mirror per transport.
- Repeat from the unfiltered control node for the same three mirrors.
- Write every result to the log with a UTC timestamp and recompute uptime.
Where probes run
Standing nodes sit inside four of the most aggressively filtered networks we have stable access to — East Asia, Central Asia, West Asia and Eastern Europe — plus a Western European control on an unfiltered connection.
The censored nodes show what a real user behind that filter would experience. The control shows what the onion service looks like when nothing is in the way. The gap between them is the whole point.
What Each Status Word Means
Four words appear in the tables. Each maps to a precise probe outcome so a reading is never ambiguous.
Reachable
The mirror answered within the timeout and round-trip time fell inside its published latency band. This is the healthy state.
Slow
The mirror answered, but round-trip time exceeded the band. Usable, often a sign of upstream congestion or CDN pressure on the meek path.
Intermittent
Some attempts in the sweep succeeded and others timed out. Typically transport-level instability such as snowflake broker churn or partial IP enumeration.
Blocked / Unreachable
Every attempt for that pairing timed out. If the control still succeeded, it is regional filtering; if the control also failed, it is a logged endpoint incident.
How the 30-Day Uptime Figure Is Built
A single honest number per mirror, recomputed after each sweep.
For each mirror we count the sweeps over the trailing 30 days in which the endpoint answered from at least one probe region within the timeout. That count divided by the total number of sweeps is the uptime percentage shown on the status table.
We deliberately use "reachable from somewhere" rather than "reachable from everywhere". A mirror that is blocked in one censored region but fine elsewhere is still serving users; counting it as fully down would overstate outages. Region-specific availability is what the matrix is for — the headline uptime answers the simpler question of whether the endpoint was alive at all.
"Slow" sweeps count as up, because the mirror responded. "Intermittent" sweeps count as up only if at least one attempt succeeded. This is a connectivity measure, not a quality-of-service guarantee, and we say so wherever the number appears.
Why a Control Probe Changes Everything
Without an unfiltered baseline, a status page cannot tell an outage from a block. With one, the answer is usually obvious.
Control up, region down
The endpoint is healthy; the failure is filtering in that region. Action is to switch transport, not to wait.
Control down too
A genuine endpoint incident. It is logged on Bulletins and users are pointed at the other two mirrors.
All green
If you still cannot connect, the block is between you and the network — local transport, ISP, or a stale address.
What This Monitor Does Not Measure
Being explicit about scope is part of being trustworthy.
- Authenticity. A probe proves an endpoint answered, never that a given copy of an address is the real one. That is the job of signature verification on the Verify page.
- Anything behind the login. We measure the entrance only. We do not operate, observe, or report on the marketplace, its escrow, or any vendor system.
- Your own connection. We cannot see your network. The control probe is the closest proxy, and it is why we publish it alongside the censored readings.
- Real-time truth. Readings are periodic snapshots with timestamps, not a live stream. Between sweeps, conditions can change; the timestamp tells you how fresh a figure is.
SEE THE READINGS IN CONTEXT
The mirror status table puts this methodology to work: per-endpoint reachability, the region matrix, latency bands, and the rolling uptime log.