1 ply search experiment vs human opponents

[Martin Hertz]

We are usually interested in the highest possible playing strength, on a specific hardware.
A few year ago I've made some experiments with the pretty weak DOS program CP2150 on Lichess
at 3.3 GHz (200 times faster than a 80486 @ 33 MHz) with 30 min and after 300 games it reached
around 2450 ELO without being banned. Surely you will being banned if using better DOS programs.

So I was curious about the rating of a strong program with a handicap of 1 ply search without
book on Lichess. I choosed MChess 5.0 for this experiment also with 30 min thinking time for the
opponent that should be more than enough for a human player against 1 ply only.

What is your estimation for the rating on Lichess with this handicap? I'll post the result soon.

[jarek]

Hi Martin,
that's very interesting.
I guess the 1-ply search is limited to non-captures and in case of captures the algorithm will go as "deep" as it is needed (i.e. evaluate the position after captures/recaptures). Am I right?
best regards,
Jarek

[Murat]

Something is not right with the 1 ply search getting up to 2450 Elo.
It should not get a win over anyone higher then 1400.

Murat

[Martin Hertz]

@Murat:

The Chess Player 2150 reached 2450 ELO at full strength, that means:

- 3.3 GHz real DOS (not emulated)
- Book on
- Ponder on
- Up to 7 plies in the middle game (only 2 plies on a 8086)
- Up to 10 plies in the end game (10 plies is the maximum of the CP2150)
- 30 min + 10 sec increment for the CP2150 and the human opponent

I think one reason for not being banned is, that Lichess has difficulties
to recognize this program as an engine, because it dosn't play like a typical
chess program, due to the type B strategy and it's pretty weak. Starting the
test, I didn't expect such a high rating for this slow searcher.


@jarek:

Yes, you are right. MChess was restricted to 1 ply with quiescence search and no book.
I expected a rating of 1300-1400 ELO, but the evaluation of MChess seems very strong,
because it reached a rating of 1957 ELO that was really surprising to me.

The Lichess Chess Insights after 27 games and 995 moves:

- Average centipawnloss: 56
- Accuracy: 72%
- Rating: 1957 ELO (better than 92% of the classical player)

A very unexpected result.

[Martin Hertz]

Due to the unexpected high result of both programs, I compared the Lichess ratings of some top classical
players with their FIDE rating and it turns out that the FIDE rating is around 300 Elo lower.

So we have aound 2150 FIDE Elo for the CP2150 after 300 games and around 1650 Elo for the handicapped MChess.