Have you considered using a different metric for scoring? When all scores are going to be very low, nobody is going to be that far off. You should reward consistency more I think


Right now, my preds are 29, 22, 15 and 34.. Relatively weak I think for Vaan and Rikku.
My absolute differences are 1.73, 4.73, 5.89, 2.88.


These numbers aren't good at all, they are actually a large percent of the score Rikku and Vaan are actually getting. I mean, being 5 or 6% off when the character is only getting 10 or 25% is a HUGE distance compared to being 5 or 6 off out of a 60% a character would score in previous years. And being 5 or 6 off then meant a score of 45 or 44 - here its even easier to be 5 or 6 off on a few and get an even stronger score than other years.


Also consider when somebody we KNOW will be fodder. Everybody says 8% for them, basically gets it right, and then is off by 10 each on the other 3. That is god awful, you just blew the final score for each character by like 40% of how much they're grabbing across the board, but got 43.3 points.


It's all relative of course, the best predictor will still win, but I like a formula that rewards being close everywhere a LOT over being close on a few and mediocre on the others. As it is, comebacks will be harder, swings will be small, protecting a lead easier, and the contest may end up decided much earlier if somebody gets a lead. I think having scores cover a wider range makes things much more interesting.


Anyway, that formula is just the distance formula. (acualA-predictedA)^2 for each of the characters. Add it, then take the square root.


Current scoring: ~46.2
Square differences, sum, square root (standard distance formula): ~42.6