Consensus health doesn't track direct connection timings
The directory authority munin graphs (ygzf7uqcusp4ayjs.onion) that track direct download timeouts show a significant problem with Faravahar. The timeouts are so bad that it was impossible to directly fetch network documents from Faravahar because it would fail almost all of the time and cause the graphs to be useless.
Sina was notified about this problem, and he pointed out that the connectivity was fine because https://consensus-health.torproject.org/#downloadstats shows that Faravahar is doing good there, has no timeouts and sometimes is even better than others.
It seems like consensus-health is only using client timings (where the client requests the consensus via a one-hop tor circuit tunneled connection). For the case of Faravahar, this works fine, and thus in consensus health we see no issues with it.
The problem is that Faravahar is dying during the direct connections. These direct connections are what all tor relays do (and not the tunneled connections).
consensus-health should track these direct connections in addition to the tunneled connections, so that these network issues can be exposed better.