latency measurements for NRCheck (https://nrcheck.tigerville.no) should be a lot more accurate now.
before, i was including the time to connect in the measurement, but the number of workers (and the type of web browser) affected it too much, so now i ignore the time it takes to connect and look only at the response time to a query after connecting, which seems to be a lot more consistent.
the only disadvantage of this is that i'm no longer measuring how long it actually takes to connect, and some relays are a lot slower at this than actually responding to queries once connected.