2

I have a lot of questions about ratings, which are sort of connected, but I think I better split them. This is pair 1.

  1. I Googled a bit: By definition, ELO>100, and someone said (chess.com) he knows someone with 200. I'd prefer "official" FIDE data to hearsay, though. Does anyone have it? (Remember ELO began only at 2200 when it was introduced, with the floor dropping to 2000 later!) Note that I can easily search DWZ (German ELO equivalent), but only for a current year.
  2. Assume a chess program that knows the rules perfectly, but makes strictly random moves. I'm sure that experiment is so simple that it has been done yet in the computer chess league. What was the ELO result? (Note that you can't compare this with a human who has just learnt the rules. Those moves will be nonrandom; I'd bet on pawn moves.)
14
  • 1
    FIDE ratings date way back to 1970, they just weren't uploaded online. You can do way worse than random though: youtube.com/watch?v=DpXy041BIlA
    – David
    Mar 22, 2021 at 13:29
  • @David: Oh, definitely. The worst algorithm would try to force selfmate, but since by definition his opponent would try to do that too, a draw is the natural outcome. (To force selfmate, you need to have a very won position first...) Mar 22, 2021 at 17:26
  • Your second question is answered here: chess.stackexchange.com/questions/6508/… Mar 22, 2021 at 18:18
  • @BrianMcCutchon: And also in the paper that came with the epic video link: 400 and some, in contrast to 200-some in your link. (Not that much of a contradiction, my national and international rating also may differ by 200 due to different opponents.) Mar 22, 2021 at 18:54
  • 1
    @Evargalo: Now that you say it, I remember :-) You are correct. Feb 5 at 18:30

2 Answers 2

8

I googled a bit: By definition, ELO>100

Well, not really. 100 is the absolute floor for USCF ratings. If your performance would ordinarily set your rating below that, it is simply set to 100. For FIDE ratings, 1000 is listed as the minimum - if you go below that, you get de-listed entirely and have to earn a new rating. There's nothing in the definition of Elo that says 100 has to be the minimum.

If you go to the USCF player search page and search by rating, specifying 100, you will get pages and pages of results. It's fairly common for a kid in their first tournament to lose every game against weak opponents.

If you search FIDE, then you'll get hundreds of results if you specify a rating of 1001, and nothing below that. I don't know why nobody is listed as 1000; it seems statistically impossible to have that many 1001 ratings and no 1000 ratings if 1000 is allowed. Maybe 1001 is the real minimum for FIDE.

In both cases, this is an artificial floor, and if lower ratings were not specifically disallowed by the organizations, they would exist.

1
  1. One detail: the Elo rating, em 1st list (1971) started in 2200. About the people with the lowest rating, you can find this on the FIDE website, on the MegaDatabase, on the website Chess.com, LiChess and other ones. There are some errors in all of them. In MegaDatabase they are mixing BCF and FIDE ratings, so there are several players with 200 Elo and strength close to 2000. In LiChess and Chess.com there are many players using engines in some games, but not in other games, distorting the results. Despite these biases, you can find some useful information.

  2. Yes, there are several programs that play pseudo-randomly. There is an old version of Andscacs with the option of random_moves. Neg 3d, Pos 1.2, ACE 0.1, Brutus RND are some who are supposed to play randomly, but in reality it is not quite like that. In the CCRL ratings, there are several errors among the weakest engines. If you examine the individual games in the database, you will find many winning positions for White in which Black leaves due to an error, or that one program makes an impossible move and the other program loses the point. Another factor to consider is that by playing at random the rating will be higher in games with less time, because opponents will play worse, while the random-player will play at the same level. On the Lc0 website, you can download various rating levels from -1200 to over 3000. In tests with several weak engines and some that supposedly play at random, the worst ones had about 800 rating and the game reasonably similar to a random game. On the Lc0 website, the lowest rating is -1200, but in the tests I did with this engine, the performance was close to 835.

This page have a list of some very weak engines (some of them with errors): http://adamsccpages.blogspot.com/p/also-rans-rating-list.html

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.