Well, bears aside, I had an App for S60 that did the same, except for a little trick I'd like to see. I don't hold much hope, since it seems like it needs a rewrite.

Still, throwing it out there: The old app allowed me to "correct" each cycle as it happened. It kept a list of "events" and tried to guess the next event by the average of each cycle. It also had the added advantage of listing a bar graph of days per cycle.

Basically, I set it on 1st, then it predicted 28th. I correct it to 29th, so the next month instead of 26th (assuming 30 days/ month), it guessed 27, effectively guessing 29 days instead of 28. And so on, averaging the last few months for a better prediction.

Again, as a coder, it seems like it's a completely different approach. It would instead keep a database of events, and linear time, and tried to guess the next event by average. Best thing was that if you skipped an event or two, correcting the third also corrected the ones before it. E.g., after a missed update, if the next event was after 56 days, it auto-set the previous missed event to 28, as half of the said interval.