Season 9, stage 3 stats

Draw rate, wins

Final draw rate was 65.6%. There was a preference for white in the decisive games, 4 times more than black wins.

Game termination

The three most common game termination causes were:
39.3% - TCEC draw rule
30.8% - TCEC win rule
19.2% - TB position

There was one crash by Jonny, a bit disappointing for such a late stage. This was mentioned as a possibility in the stage 2 statistics report. At least it didn't affect the qualification outcome.

Moves per game

Median= 62.75
Average= 71.4

The distribution is similar to the one in stage 2.

Time per game (hours)

Median= 4:57
Average= 4:54

Games were longer than in stage 2 by about an hour, due to the longer time control - 30 minutes more per engine. I'm not sure what causes the double peak.

Openings

There were 8-move book openings in this stage chosen by Cato, so the engines had almost no say. The first letter of the ECO codes was distributed as follows:

Open games (C) were under-represented, and there were many flank openings (A), mainly Dutch and English, as well as Sicillians (B).

If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
Sicillian - 51 times
English - 22 times
Dutch - 20 times

The 8 predetermined moves usually made sure that the reverse pair games had the same opening definition. There were 6 cases where the full opening name was different, of these 5 had different ECO codes.

Reverse pairs, wins

Only 4/112 openings are suspect of a bias - each engine wins once, while 53/112 openings have a win+draw combination - a proof of engine strength, winning from one side and holding the other side. Cato's opening choices manage to keep the draw rate down with opening bias while letting strong engines find defenses the weaker engines struggle with.

Reverse pairs, same moves 

The reverse games diverged quickly but less so than in stage 2. About a half of the game pairs had the same first two plys. One game had the same move sequence for 25 plys (Houdini - Komodo, 1/2-1/2).