Mirror-by-mirror state as of the latest probe cycle, with per-mirror p50 / p95 latency and the current rotation timestamp.
All three production mirrors are responding. The headline mirror leads p50 latency at the lowest figure across the rotation; backup A sits second; backup B is in the elevated-latency band that defines its degraded-mode classification. None of the three is offline; the elevated-latency state on backup B is consistent with its role as the failover mirror.
The full per-mirror table with onion addresses, latencies, and current rotation timestamps is on the homepage. Use the Copy buttons there. The mirror addresses on this status page rotate on the standard 24-hour cadence; do not bookmark a mirror — bookmark this page and re-resolve the rotation on each visit.
A hidden service reachable at exactly one address stops being reachable the moment that address is flooded, and flooding one is cheap. Several running together means pressure has to be spread across all of them to have any effect. For you the benefit is smaller and more useful: when one is slow another usually is not.
Account, balance, open orders and message history sit behind all of them equally. Switching costs nothing beyond a fresh connection, and an order that looks missing after a switch almost always means you are signed out rather than looking at different data.
Start with whichever you used last. If it stalls, give it a full minute before moving on, then work down the list rather than reloading. Once one answers, stay on it for the session, because switching halfway throws away a connection you already paid for.
Each address builds its own circuits, so the fastest one this hour may be the slowest one tonight. That is why no permanent ranking is published here. A measurement from yesterday describes yesterday's path and not yours.
Component states, the 90 day uptime history and the incident log live on the status dashboard. This page covers one question in depth rather than repeating the board.