We do not have the same tournament environment, not a single component is identical in terms of software, hardware, time control, rounds, etc. etc.
For interest, the tournament manager used here is cutechess-cli v1.0 which is the same as used by the Stockfish testing. This is probably the most accurate tournament manager. If in doubt, you can reproduce the result by running it yourself.
Additional tournament conditions:
1. Time control is 1 minute + 1 second which is longer than Stockfish testing LTC of 1 minute + 0.6 second.
2. The tournament was ran under Linux (PCLinuxOS 4.9.13), AMD 4 cores, 8GB ram, cutechess-cli v1.0.
3. Used single core only with 10 simultaneous games.
4. Book used is customized which are selected from the games between Brainfish and Stockfish/Sugar/Raubfisch with average of 120 seconfs per move which is longer than the 70 seconds of Brainfish. All of the games were draws up to 15 moves which is truncated to 5 moves for the tournament with a threshold of 30 centipawn. Moves selection are random from a set of 700 drawn games and each game is repeated to make sure each engine gets equal white and black color.
drj4759 wrote:Additional tournament conditions:
[...]
2. The tournament was ran under Linux (PCLinuxOS 4.9.13), AMD 4 cores, 8GB ram, cutechess-cli v1.0.
3. Used single core only with 10 simultaneous games. [...]
2.5 simultaneous games per core? Quite a novel approach.