Well it does mention that BBR uses "The RTT is the minimum of all RTT measurements over some time window that is described as “tens of seconds to minutes”." This is a much larger measuring window than previous RTT tracking TCP. At 10s of seconds it is also hopefully enough to tackle the most common forms of jitter and intermittent flows.

At truly bad flow characteristics you need to start some truly weird line coding methods with stream data split over multiple packets with redundant coding to rebuild dropped packets etc, but this kind of thing just kills latency.