I found myself (~1600 classical) playing a player rated 1500? this morning that was rated ~2200 in blitz/bullet. Obviously this person should be rated much higher and odds are I'm going to lose. That isn't a great feeling. Maybe I'll learn something =)
Anyway, I expect the experience can be improved. Simply running a regression of known ratings across the player pools would allow you to estimate a person's classical rating from their blitz rating with some degree of precision to set a starting rating and avoid some of these less than great match ups.
This makes sense. Unless the provisional rating is categorically different than a rating with a high deviation, it seems pretty easy to adjust provisional ratings based on other time control rating.
After one game this opponent's rating is now 1874. Changing the rating system will encourage others to try to "game" it.
I'm curious, how are they going to game the ratings? It seems that they could just as easily game provisional ratings right now.
The point that the player is now rated 1874 is sort of irrelevant. You are likely generating millions of bad matchups. And if you find yourself hovering around the 1500 rating you'll be getting matched up with a large number of players with provisional ratings which is annoying.
#4 The trade-off here is: how many bad match-ups are there? As indicated a player gained 374 points from 1 game, so for there to be millions of bad matchups it means we're registering millions of untitled but titled-strength players.
My own take about the nuisance value isn't so much the provisional-ness of the opponent's rating, it's that it's possible to know whether the opponent's rating is provisional (while also staff jokingly call such players "dirty question-markers"):
github.com/ornicar/lila/issues/6864Alternatively I proposed that (?) is less informative than showing 2*RD as Prof. Glickman recommends
http://glicko.net/glicko/glicko2.pdf but Lichess apparently prefers the (?) indicator and the tension it causes (instead of showing RDs and letting players see that even established players can have relatively large RDs, and that having a provisional rating is nothing to worry about except for leaderboard purposes).
For argument's sake let's assume you are 1500 rated and you only play 1500? rated players. If you play 1000 games with those players on average ~63% (630) of your opponents will not be within 150 pts of your own rating. Mind you that 63% is the minimum because it is in the center of the distribution. To that end to generate a million bad match ups you don't need to register millions of titled players. At maximum you just need to register a few hundred thousand players (500K players * 0.63 x 3 different time controls).
Therefore I fail to see why setting a provisional rating based on player ratings in other time controls doesn't make sense. Terrible analogy time, it's akin to telling Usain Bolt that he has to run the 200M in JV at the high school level because while we've seen him crush the 100M we don't know how he'll do at the 200M. The converse could be true as well just replace Usain with a turtle or something.
Edit: if anyone is curious where I got the data I pulled it from the classical ratings distribution plot and I assume the ratings distributions will be fairly consistent across time controls.
#6 I'm 2183 at bullet so let's just assume that I'm 2400 at ultrabullet.
But if you're serious about this proposal, comment on
github.com/ornicar/lila/issues so people who aren't categorically opposed to your suggestion can see it.
Some would say ... Play the chess position, not the players rating.
Chess experience is not a chess rating.
Time spent playing equals experience !!
Look at a players profile and add up all the games played vs time spent playing on site.
Rather than changing the rating system, it would be better to create a category formula to give the players a category letter.
Sort the players by category, then sort each category by rating and finally pair by won games.
If their are 4 groups of letters, it would be like 4 tournaments in one tournament.
Category A players should not be paired with category C players. If they get paired, players could nearly bet on who would win.
A provisional rating of 1500, should only be playing in provisional permitted open tournaments. Once their rating has stabilized, then they can join other semi-closed tournaments that best fits their category.
Make a 24/7 constantly running tournament open only for provisionally rated players could be useful. Include the bots in that tournament that have a predetermined rating. Once the provisional players have played enough games, their rating will have been somewhat stabilized and will be ready to play in other human tournaments.
You don't get an exam mark and then do the exam. You do the exam to get a mark.
New chess players need to go to an AI provisional chess tournament to get tested, before they get a provisional rating.
With their provisional rating from engines they can then play against humans in the lobby to fine tune their rating.
The last step is playing in human tournaments only once the profile rating has stabilized.
