Top 15 round robin (balanced books) 89% Drawfest

[Modern Times]

Some of you will be familiar with my Chess324 Top 15 round-robin, that uses an SPCC unbalanced book.
Chess324xxl_big+090_+119.pgn

https://e4e6.com/324/

White wins: 7'896 (37.6%)
Black wins: 271 (1.3%)
Draws: 12'833 (61.1%)
White score: 68.2%
Elo Spread top to bottom (Ordo) 213 Elo

I decided to run the Top 15 under the same conditions, except standard chess with:
SPCC 2400+_6mvs_40k.pgn

Classical human openings, balanced endpositions. Queens still on the board.
All lines filtered out of MegaBase 2020. No double endpositions.
KomodoDragon 1.0 analyzed each endposition on a QuadcorePC with 15 seconds thinking-time.
Allowed Eval [-0.39;+0.59]
2400+_6mvs_40k = 40720 different endpositions. Both players 2400 Elo or better

That is, the "traditional" approach.

Well, no surprises I guess. The result:

White wins: 1'036 (8.1%)
Black wins: 318 (2.5%)
Draws: 11'456 (89.4%)
White score: 52.8%
Elo Spread top to bottom 87 Elo (Ordo) and 50 Elo (bayeselo)

With the huge draw rate comes significant Elo compression, and in my view completely pointless running the top engines in this way. Hence the increasing popularity of SPCC UHO books for example.
.

Code: Select all

# PLAYER               :  RATING  ERROR  POINTS  PLAYED   (%)    W     D    L  D(%)  OppAvg  CFS(%)
   1 Stockfish 17.1       :    3859      6   991.5    1708  58.1  277  1429    2  83.7    3802     100
   2 Reckless 0.8.0       :    3835      5   929.0    1708  54.4  168  1522   18  89.1    3803      83
   3 Obsidian 16          :    3831      5   920.0    1708  53.9  149  1542   17  90.3    3804     100
   4 PlentyChess 6.0.0    :    3820      5   891.0    1708  52.2  110  1562   36  91.5    3805      67
   5 Alexandria 8.1.0     :    3818      5   887.0    1708  51.9  105  1564   39  91.6    3805     100
   6 Berserk 13           :    3804      5   849.0    1708  49.7   65  1568   75  91.8    3806      55
   7 Clover 9.1           :    3803      4   848.0    1708  49.6   62  1572   74  92.0    3806      63
   8 Viridithas 18.0.0    :    3802      5   845.0    1708  49.5   80  1530   98  89.6    3806      86
   9 Integral V7          :    3798      5   835.5    1708  48.9   65  1541  102  90.2    3806      52
  10 Caissa 1.23          :    3798      5   835.0    1708  48.9   63  1544  101  90.4    3806      88
  11 Stormphrax 7.0.0     :    3794      5   823.5    1708  48.2   57  1533  118  89.8    3806      73
  12 Horsie 1.1           :    3792      5   818.0    1708  47.9   56  1524  128  89.2    3807      98
  13 Motor 0.9.0          :    3784      5   799.5    1708  46.8   46  1507  155  88.2    3807     100
  14 pawnocchio 1.8       :    3773      5   769.5    1708  45.1   27  1485  196  86.9    3808      54
  15 Halogen 15           :    3772      5   768.5    1708  45.0   24  1489  195  87.2    3808     ---

  White advantage = 19.84 +/- 0.99
  Draw rate (equal opponents) = 95.08 % +/- 0.31

Total: 12'810 games played by 15 programs

White wins: 1'036 (8.1%)
Black wins: 318 (2.5%)
Draws: 11'456 (89.4%)
White score: 52.8%

[Graham Banks]

With the huge draw rate comes significant Elo compression, and in my view completely pointless running the top engines in this way. Hence the increasing popularity of SPCC UHO books for example.

Well, seeing more wins is always enjoyable, but that's about all it's good for.

Still, enthusiasts are always free to do as they wish.

[Modern Times]

With the huge draw rate comes significant Elo compression, and in my view completely pointless running the top engines in this way. Hence the increasing popularity of SPCC UHO books for example.

Well, seeing more wins is always enjoyable, but that's about all it's good for.

Still, enthusiasts are always free to do as they wish.

Testing these engines with traditional balanced books is good if you find draws enjoyable, but that's about all it is good for.

[Ajedrecista]

Hello:

I find interesting these types of tournaments, more like 'real' conditions where top human players try their best and not going into one-sided games from the very beginning.

An idea could be play the top 15 (or top N) once yearly, for example, with these conditions... a kind of unofficial World Computer Chess Championship. We would see the evolution of statistics at the very top: Elo span between engines, overall draw rate, overall white performance, average number of moves (overall, drawn games and decisive games) and so on.

Special mention goes to SF 17.1 and only two games lost over 1708, given the strong pool of players. Absolutely outstanding!

Regards from Spain.

Ajedrecista.