Change the 1v1 Ladder starting rating to some value other than 0 (900-1300 would be great).
Much of the problems that get blamed on Bayeselo are in fact simply because the initial rating on the 1v1 Ladder is 0. 0, unfortunately, does not mean "no rating." It means you start from the bottom and set off a vicious garbage-in-garbage-out cycle of volatile, exploitable, slow-to-converge Bayeselo ratings, because Bayeselo is conservative and will keep matching you at the bottom. This probably also depresses ladder participation, since players tend to leave at a very high rate when they win 4/5 of their first 5 games.
Note that:
- Bayeselo is relative so 0 isn't any more special a value than 100 or -300. On the 1v1 ladder, it just means "1500 points worse than the average player."
- Every single new ladder player is significantly better than a 0 rating would imply. Players on a long boot streak hit 600-700. (In practice, an initial rating of 0 implies "worst player on the ladder" as the null hypothesis, hence the vicious cycle)
- By my math the actual skill level at the ~25% percentile for new (first-time) ladder entrants is somewhere between 900 and 1300. You could also do 1500 because...
- ... starting at the center (or close to it) on the rating distribution will help Bayeselo converge faster, so this won't just mean newbies start on boot streaks (especially since Bayeselo has an uncertainty parameter). Think binary search.
for more details: https://www.warzone.com/Forum/641242?Offset=0
-
l4v.r0v commented
whoops, loss streaks, not boot streaks