@mtreinish reported a case where increasing the number of trials in the StochasticSwap pass resulted in longer circuits. This should not happen with a fixed seed, because the trials should just pick the best one. Unless there is something going on with how individual layers get better with increased trials, but those layer choices combine to make the overall circuit worse. I would like to understand what's going on here, and whether we can ensure that increasing the trials gives us better circuits (at the cost of longer transpilation time).

I don't have a particular example to reproduce here, I think the original example was a quantum volume circuit. But really a variety of examples should be investigated.