If it is a scheduled method, the algorithm is non-trivial, since it is based on the retention time scheduling windows which are not obvious to you. Though, you gain some insight into their distribution by looking at the View > Retention Times > Scheduling plot.
I think there is even a semi-random component to try to balance the number of transitions across multiple files, rather than simply dumping all possible transitions into each successive file, which would tend to put more transitions in early files. e.g. You might have a set of transitions for which only 1 peptide with 6 transitions exceeds the number you are willing to allow. This means you need 2 runs to monitor all of your transitions. A naive method would dump all other 734 transitions into the first method and 6 into the second method. Instead, Skyline uses random selection to partition the transitions which need to be measured as evenly as possible across the 2 necessary runs.
Not sure that makes it any clearer, but at least you maybe understand why the outcome is non-obvious.
Thanks for posting to the Skyline support board.