I’ve had a non-exit relay (602D56DFD408530F7FE3056CAF1DA413A50ECAE4) that has been getting low weight votes from authorities [snip]
It might be a similar issue as https://gitlab.torproject.org/tpo/network-health/sbws/-/issues/40190.
I would also ask for the Tor developers’ reasoning for using median of authorities in consensus weighting rather than mean since in my mind, if no authorities are malicious/compromised to give false values, then mean would give a more accurate representation of actual measured bw performance from the authorities, IMO.
You're right, "**if** no authorities are malicious/compromised", but what if one or more become compromised?. Taking the median was specified at https://spec.torproject.org/dir-spec/computing-consensus.html Hope it helps, juga