Further update:
I changed two other affected relays to guard last night.
The consensus weight didn't budge at all and bandwidth usage was negligible, so I've deleted all my onion keys for those two relays.
So these relays are starting fresh. We'll see how it goes. I'm going to wait until I get some flags and see a measured speed of anything above 160b/s, as I'll assume this means regardless of actual accuracy, the bwauths have been able to determine *something* about me, then I'll switch the flip back to exit. This should take a while, as they have to go through the different stages again.
My first relay's consensus weight hasn't budged yet, but it was in the thousands before I changed from guard to exit, so I'm not concerned - it hasn't dropped is the point I'm trying to make.
The first relay is now finally being used appropriately, at 13Mb/s in and out each. Since last evening I've done over 250GB in/out.
Arm is still reporting the measured speed as 45Kb/s.
Relay that is now working https://atlas.torproject.org/#details/D60B02A13F5D9CAFD6EC27A5332C5FEF5B7691...
You can see from March to near the end of April, as an exit relay, damn near nothing was going on. Around April 20th I changed it to guard, and activity finally started happening. Yesterday I switch back to exit, and you can clearly see the exit probability jump. :)
For comparison: https://atlas.torproject.org/#details/034A61A65419B5A115A7B6B253CAC4C9D5731E...
Like the other relay, it's consensus never went above 20. However, when I switched the flip to guard, the middle and guard probability increase was technically present, but negligible. 0.0002% and 0.0001%, respectively. Because the first relay's consensus weight responded almost immediately to the change, I arbitrarily decided the problem for the other two relays was not fixed. So I deleted the keys and am starting over to see if that helps any.
And here we are.
Matt Speak Freely