Humanized Engine Rating List

[lkaufman]

Yes, these results for Dragon 3 Elo 2300, Komodo level 21 and Komodo level 20 seem about right to me, since I am competitive with them at classical, so I could test them myself. Other engines (or levels) I have not tested, but generally it seems that values are correct for higher rated ones, but maybe too high for lower ones. I suppose it's because the rating difference between them and Leela is too big to be reliable.

Yes, I am inclined to agree that the low end ratings are too high. Using Lc0 as the opponent causes a rather severe contraction in the rating differences, perhaps more than is justified for human opponents. What I'm seeing with the Dragon Elo ratings is that they seem reasonably accurate (for the specified Rapid TC) at around 2000 and higher, but when you go too much lower than 2000 they are a bit too generous, quite a bit too high when you get down around 1000. I still don't understand exactly what causes this, but I think it has something to do with the way in which the engine is weakened once it drops below a simple one ply search. I think that the simulation of human play is better around and above 2000 level than it is around and below 1500 level. Perhaps in future versions we'll be able to improve the simulation of low-level human play.