Oh that's interesting. That guys says the cancel number the game gives is the frame on which the move you cancel into starts coming out. In that case the calculation you're doing sort of makes sense. I always interpreted it as the opposite - the number of frames that got cancelled off of the recovery(I think this is traditionally what cancel refers to - cancelling recovery - but I guess I'm not sure). I always look for a high cancel number, but that means you want a low one. In either case I'd have to look at a bunch of frame data to prove or disprove.
Here's the swipe breakdown -
f4, f412, 4 - jail, reversals don't come out against a bot
34, f21,212, 12, 3, f3 - beat reversal flipkick, which should mean they can be backdashed/armored out of but beat all normals
f41, b1d4, 14, 1, 2, b3, d3, d4 - lose to both reversal flipkick and reversal spin, which means most characters can at least poke out
21~swipe loses to flipkick but beats spin. This probably isn't useful since 212~swipe beats flipkick.
The biggest take away from this is that you can block confirm f412 into swipe and it's safer, does more chip, and builds more meter. Doing it off the string with gaps is probably pretty safe, for now at least. Even when people learn the gaps you can probably turn it into a guessing game
I looked at the math you did and it looks right, so either the method you used or the frame data we're given is wrong. I tried using that method for some other strings I know that do(and don't) have holes in them and couldn't get consistent results nor could I really see any method or pattern that worked. So that method is useless for whatever reason. Personally I think there's just a lot of data we don't have. I can't think of any obvious examples in MKX but in previous NRS games moves would have different recovery on hit/block/whiff(that was how Ares guaranteed Godsmack resets on some characters worked, for example). This game looks like it has some sort of thing where sometimes things visually slow down slightly on block(I thinks that's why some of the option selects worked?) so who knows how that affects things. There's probably different cancel advantage on hit and on block and the cancel number it gives is something completely different and that combined with all the wrong frame data just makes everything impossible to calculate cleanly. As much as I wish you could nerd out and just do a bunch of math with the numbers they gave us you really just have to test everything. Luckily NRS gave us a pretty robust training mode to do that with