“Strict rotation” and “random” strategies ignore separation rules altogether. That’s the sole point here.
“Random” picks just any song from the folder.
“Strict rotation” always uses the next song in the folder, i.e. the one that hasn’t been scheduled for the longest time - please note that this may only be accurate for the hour it had been last scheduled in, so two songs from the same hour may be treated equally. That’s because the scheduler does not keep any “play count” or “next in rotation” info anywhere, but always reconstructs the info from the past scheduled playlists when you start scheduling the next time.
The “smart” strategy is what has been the default/only strategy until now. So with all the separation rules and so on.
To understand the “variance” setting, remember how the scheduler works when picking a song: It sorts all songs by the “last scheduled” timestamp, oldest to newest. Then walks through the list, filters out songs that don’t satisfy the attribute filters, and also calculates the penalty from the separation rules. Only the songs with the lowest penalty (ideally 0) are considered. And from that list of “perfectly matching songs”, it uses either the first one (with a probability of 50%), or the second one (with a probability of 25%), or the third one (12.5%) and so on (up to the 8th or so).
Until now, the 50% probability was hardcoded. Now you can adjust it with the “variance” setting. For example, 100 would mean 100% = always use the first song.