Hi all, I'm running an exit relay and seeing a strange discrepancy between bandwidth authorities. Some measure my relay without any problem, while others fail every single attempt with stream errors. Relay: beacon Fingerprint: 85082CD18CF6C931968105DFFFDDE82EBC4C9AF8 Observations from 2026-07-22: moria1 (13:47 UTC) — success=3/3, error_stream=0, error_circ=0, bw=6400 tor26 (06:23 UTC) — success=1/3, error_stream=0, error_circ=2, bw=1 maatuska (14:21 UTC) — success=0/26, error_stream=26, error_circ=0, unmeasured=1, vote=0 longclaw (09:32 UTC) — success=0/7, error_stream=7, error_circ=0, unmeasured=1, vote=0 gabelmoo — no data for this relay What stands out: error_circ is 0 at both maatuska and longclaw, so circuits build fine. The failures happen at the stream stage, and they happen every single time. Meanwhile moria1 measures the same relay flawlessly on the same day. Configuration: ExitRelay 1, ReducedExitPolicy 1, IPv6Exit 1, RelayBandwidthRate 37 MBytes, RelayBandwidthBurst 42 MBytes What I have checked locally, without finding anything: - DNS: local Unbound resolver, 0 serverfailed, 0 refused, 0 truncated, 213 timeouts out of 4920 queries (~4%) - IPv6: globally routable, verified from outside - MTU: full 1500 bytes pass unfragmented on both v4 and v6, no packet loss, ~6ms latency - TCP: ListenOverflows 0, ListenDrops 0, PruneCalled 0, conntrack 710/65536 - File descriptors: 65536 limit, ~620 active connections - OOM: tor_relay_load_oom_bytes_total 0 across all subsystems - Onionskins: 0 dropped - Firewall: outbound policy ACCEPT, nothing blocked - Tor logs: no warnings or errors over the past several days - System clock: synchronized via NTP The relay is relatively young (relay_in_recent_consensus_count between 28 and 85 depending on the authority), so I'm also open to the possibility that this simply resolves itself with more measurement attempts over time. Thanks, beacon