What level is a 300 rating?

[Uri Blass]

I do not play with clocks when I give queen and 2 rooks material handicap but I think one of the problems is that the student I teach tends to play too fast.
The rating of my student is based on 10+0 games in chess.com
I guess your student can improve her rapid rating if she play 15+10 time control in chess.com because many of the players are not using their time in the low level.

Note that I also do not know if one of her problem is losing games on time or resigning too early and a mistake of stalemate is clearly a practical chance in these levels.

[Ovyron]

So it seems that something causes severe compression of engine vs engine ratings (relative to human ratings) at these very low levels

It's because there's too many gaps, if top engines had to play 1600 rating engines only leaving the big gap, they wouldn't be able to reach their current ratings and you'd observe the same compression at the top.

For these I suggest the following:

Build a special version of Stockfish that plays a random move some of the time, and the rest plays the strongest moves. You have one that plays random moves 99% of the time, another 98% of the time, another 97% of the time... and you use these entities to fill those gaps, decompressing the bottom of the list. The one that plays 100% random moves doesn't play against the bottom engine, it plays against the one that plays random most of the time, but the strongest on the other ones.

People could measure their engine's progress by seeing at what percentage of random moves it beats, and in your case, student's improvements could be measured against it (say, a student that could only beat 26% random moves now being able to beat 25% random moves.)

[Uri Blass]

The main problem with random movers is that there can be an improvement based on a better strategy against it and not based on improvement in playing strength against humans because no chance that your human opponent is going to make that mistake.

Chess programs should try to simulate humans even when they do mistakes.
For example a problem that I see with weak humans who are not beginners is not that they generally do not see captures but simply that they forget about one of their pieces (usually a piece that did not move for many moves) and do not look at it so do not see the capture that this piece can make.

The same player usually can see a direct threat of a piece against another piece and escape from it but may do mistakes because the player ignore pieces that did not move recently and allow them to capture or ignore indirect threats.

Another common problem with weak players is not seeing backward captures or long captures.

Another problem is when there are 2 direct threats when they see one threat and protect against it and do not see the second threat that is more important.

Another problem is thinking about one good move that cause to ignore a better move.

Here is an example:
https://www.chess.com/puzzles/problem/1817944

My student failed in this problem and replied Rxe6 by Kxe6 and I am sure that she could find the mate in case of telling her it is a mate problem.
I told her already to look for mates first in problems but she does not think that way.

This is another example when she played Rf8 in reply to Qxg7 because of thinking that if the opponent attack she needs to escape(of course she could solve the mate in 1 problem).

https://www.chess.com/puzzles/problem/1598082

[lkaufman]

So it seems that something causes severe compression of engine vs engine ratings (relative to human ratings) at these very low levels

It's because there's too many gaps, if top engines had to play 1600 rating engines only leaving the big gap, they wouldn't be able to reach their current ratings and you'd observe the same compression at the top.

For these I suggest the following:

Build a special version of Stockfish that plays a random move some of the time, and the rest plays the strongest moves. You have one that plays random moves 99% of the time, another 98% of the time, another 97% of the time... and you use these entities to fill those gaps, decompressing the bottom of the list. The one that plays 100% random moves doesn't play against the bottom engine, it plays against the one that plays random most of the time, but the strongest on the other ones.

People could measure their engine's progress by seeing at what percentage of random moves it beats, and in your case, student's improvements could be measured against it (say, a student that could only beat 26% random moves now being able to beat 25% random moves.)

I don't believe in this explanation. I run tests like 300 elo plays Skill 2 of old Komodo 12, 400 elo does so, then 500, once dragon comes out plus raise the skill level by one, etc. No big gaps. The elo system works fine when the gap is below about 300 elo or so. This is much more sensible than using versions that randomly vary from negative elo to 3500, that's not a simulation of anyone or anything. But my current method doesn't seem to work very well either, so I welcome other suggestions, but this one makes no sense to me. I suspect that the problem may be related to weak engines (doing one or two ply searches) being unable to checkmate or perhaps to avoid simple draws by repetition or stalemate. Humans resign, flag, or just stop thinking once they blunder a major piece, but engines don't care, they just keep playing, sometimes saving the game if both are just doing one ply searches. At least this is my current theory.

[Ovyron]

but this one makes no sense to me.

At least it makes more sense to me than queen+rook handicaps and things that don't reflect real chess. Even you could use it against your students, playing a normal game, but the engine takes over your side and plays a random move with some % chance, they'll play at your level at some % chance, as they improve you can decrease this %.

(note I'm not making this up out of thin air, I also teach chess to people, and found the best way to make them improve is to make them sit on their hands and continue calculating despite having found their move, looking for a better one, and some 700 elo is achieved rather quickly)

[Uri Blass]

but this one makes no sense to me.

At least it makes more sense to me than queen+rook handicaps and things that don't reflect real chess. Even you could use it against your students, playing a normal game, but the engine takes over your side and plays a random move with some % chance, they'll play at your level at some % chance, as they improve you can decrease this %.

(note I'm not making this up out of thin air, I also teach chess to people, and found the best way to make them improve is to make them sit on their hands and continue calculating despite having found their move, looking for a better one, and some 700 elo is achieved rather quickly)

I do not claim that odd games is the best way to improve but same is for playing against a player who play a random mover in part of the times.
My student played against bots who play often random moves and the problem is she learned from it to play blunders inspite of knowing that they are blunders because of knowing there is a good chance that the opponent is going to make a more serious blunder.

For example RxN that lose the rook for a knight may be a good idea if there is a probability of 40% that the opponent is going to miss QxR and maybe even allow the rook to capture the queen in the next move.

[Uri Blass]

I can add that developing the queen for plans like
e4 Qh5 Bc4 Qxf7 mate is a good idea even against random movers with 1500 elo against humans
because even if the probability for a random move is only 3% there is a significant chance that the opponent is going to blunder when it is not a good idea against humans at this level.

[Ovyron]

My student played against bots who play often random moves and the problem is she learned from it to play blunders inspite of knowing that they are blunders

It seems students playing bots is a really bad idea in general, if the bot will eventually make a random blunder the student can learn to be passive and wait for that blunder, which won't work on a human because you need to be active to cause the blunder.

[Peter Berger]

Another problem is thinking about one good move that cause to ignore a better move.

Here is an example:
https://www.chess.com/puzzles/problem/1817944

My student failed in this problem and replied Rxe6 by Kxe6 and I am sure that she could find the mate in case of telling her it is a mate problem.
I told her already to look for mates first in problems but she does not think that way.

This is another example when she played Rf8 in reply to Qxg7 because of thinking that if the opponent attack she needs to escape(of course she could solve the mate in 1 problem).

https://www.chess.com/puzzles/problem/1598082

To me Rf8 looks forgiveable, while missing Rd1 in your first example is not.

For me it is very difficult to imagine a player who even does training having an ELO of 300 after say a month.

There is very little you have to know to reach ELO 1000 . You have to be able to deliver mate with two rooks and a bare queen , and eventually with one rook against the bare king - this one everyone can learn easily ( besides people who are seriously handicapped). Then someone has to show you a few very basic patterns, like the backrank mate, the concept of a fork, the mate on f7/f2 - that's about it.
Then some basic ideas on what the opening is about ( getting pieces out, control center, castle) - for reasons unknown few beginners seem to be able to understand this without being told/taught.

I used to do a lot of training with kids involving heavy handicap games - when they know the above already a queen handicap is usually enough for them to be competitive if you allow them one or two takebacks per game.

Intuitively I think that random movers are useless for training - as even extremely weak players have all kind of "plans" - and be it to keep all their pieces and steal those of their opponents to deliver the above endgame mates they have learned.

Peter

[Uri Blass]

Extremely weak players cannot keep all their pieces otherwise they could be better than extremely weak players.

300 elo rating is not the weakest level of extremely weak players and even 400 elo or 500 elo players sometimes blunder their pieces and allow them to be captured in the next move.

There are players who can improve to 1000 elo without a trainer but there are also players who inspite of training do not get close to this level.

Different players have different abilities.