There's one important caveat here: you need to start with a random state, then find a solution to that that doesn't start with R* and doesn't end with F*. If you instead generate random states until the scramble sequence satisfies those constraints, this introduces a minor bias to the scrambles. (This is already mentioned in the GH issue, but it's worth highlighting separately; generating random-state scrambles comes with a whole bunch of subtleties that might be missed on first glance.)So the initial scramble needs to have nothing that cancels or partially cancels an F at the beginning or an R' at the end.