Example: http://warlight.net/LadderGames?ID=0&LadderTeamID=1233

but in your example doggy beat quit a few good players. none after the expirings, but also not the same as winning first 10 games and nothing signature

terra: 1608
The General: 1784
Mr. Stein: 1258
negpue: 1617
jasdanmoo: 1398
nagromo: 1245
Honey Badger: 691
manunuma: 953
{101st}Blowfly: 1076
skunk940: 1208
Stephan: 1071

He has certainly beaten plenty of great players in the past. But those 11 are not great 1v1 players. Maybe one of them is good. If you are ranked based on 15 games and 11 of them are relatively easy, that makes going 15-0 easier. 20-0 would require amazing luck (without stalling), even if 11 of the 20 are relatively easy games.

"20-0 would require amazing luck (without stalling), even if 11 of the 20 are relatively easy games" Eh, I could do that.

go do it then

I got 16, and they were all hard opponents. If I had gone from rating 0 I dont doubt the first 4 (minimum) would have been easy wins, I didnt because wins against very low rated players reduce your rating.

the alt has more confidence. the main account has excuses.

The thing about Doggy is that he can win a single game against a high rated player (even just above 1900) and end up with a 2000+ rating, at the pace he is going. He's not likely to reach 2100, let alone a record, but it's a relatively easy way to gain a higher rating than you might otherwise deserve. His previous high point was 1956, with wins against good players and a normal distribution of losses. If he takes a loss, even to a higher rated player, his rating will be much more accurate than if he takes no losses, even with only a single win against a good player.

People with more skill in math: Is there a fair way to weight ratings downward when there are no losses counted yet? I'm thinking of a system auto-correct that assumes some level of losses even when there are no actual losses, to help prevent gaming a low sample size.

Duke of Ben: some statistical analysis methods disregard a certain highest and lowest percentile of results to level things down around the mean result. i think some sort of variant could work better here than to simply penalize lack of losses.

i guess what i am suggesting is an algoryhtm that ignores the top and bottom games, a higher number of them when the number of games is lower. and a lower number of them when the number of games is higher. i guess the higher and lower fluctuating as a percentile of the average number of games that active players have in the ladder.

so if you have 20 games and the average is, lets's say 35. the algo ignores the two top and two bottom games of the player.

and if you have something like 30 games you only get ignored one game from top and one from bottom.

in practical terms it would imply that if you're new or returning you must beat more higher rank players to debut on top.

a simpler way to implement it could also be achieved with a co-efficient (or whatever it's called) multiplier, of the number of games relative to the average of the ladder. so if you have 20 games and the average is 35, you get your rating multiplied by 20/35. only when you reach the average number of games do you have your real 1:1 ranking. and if you have more games then average you would also get a slight boost.