This is basically my understanding as well, with the addition that, iirc,
by default the final output is a summary over the (default 3?) "best"
iterations of the baseline and candidate, respectively. The idea is to
allow each version to "put its best foot forward".
Michael
On Fri, Jan 7, 2022 at 1
My understanding is that, 1) there isn't any specific relationship
between the iterations, and 2) the final output is a summary over all
iterations. The idea is that randomness might affect results on any
particular iteration, but by running multiple times (20 I think?) and
then aggregating the sta
