Selection one jumps between the start and the end of the show repeatedly until you make it to the middle (selection two), after which point it'll move on to getting you to rank the best episodes, and then the worst episodes.
Definitely looking into seeing if I can come up with something better though! The problem is making sure that whatever algorithm is picked remains as close to O(n log n) as possible. Randomising options in a way that makes require a lot more comparisons would be far worse.