Originally posted by Crowley
My suggestions on the previous pages rocks.
You all should go and read them - you'll see why that's the only true solution to this problem.
I rule.
I think you've got a nice idea, but the weighting you give doesn't work.
For myself, it works quite well as I have steadily increased and hovered around my highest rating. (Roughly) Applying your formula...
1200-1300 0 months
1300-1400
1400-1500
1500-1600 4 months 413.33
1600-1700 2 months 220
1700-1800 4 month 466.66
1800-1900 5 months 616.66
Total: 15 months
Banding: 1716
Which would put me in the 1700-1799 band which would be right, IMO.
But look at Checkmate187...
1200-1300 0 months
1300-1400
1400-1500
1500-1600 10 days 139
1600-1700 12 days 178
1700-1800 3 days 47
1800-1900 5 days 83
1900-2000 4 days 70
2000-2100 12 days 221
2100-2200 30 days 567
2200-2300 25 days 506
Total: 111 days
Band: 1811
He's obviously not an 1800-1899 player.
I'm not sure (yet) how to fix this, but I don't think time should be taken into account. Dustn's a 1900odd player. If he leaves for a few months, then these months shouldn't be used in the calculation due to his non-activity.
I think Ponderable's suggestion is probably the best. But the spike shouldn't be just one game. If you calculate the spike as the highest 30 day (maybe 10/30 game average seeing as time should be omitted from the calculation) average, then people wouldn't get screwed if some higher rated players got auto T/O'd.
Instead of banning players from banded tournies until 50 games have been played (even though I can see the merits of this), players who are obviously on the rise, for example
User 170139 upto September, their bands should be determined by the highest rating of their beaten opponents. If they beat 2 1600+ players while rated at 1400 while being unbeaten against 1500+ players, then their band should be 1600-1700. As soon as the angle of ascent levels off (I'm sure there are ways to figure this out mathematically), then the normal method of banding is employed again.
D