Season 22 superfinal games 1-10

After 10 games Stockfish leads 4-1 with 5 draws. Stockfish won the first 3 game pairs. KomodoDragon managed to convert a book advantage to a win only in its 5th attempt in game 9, but Stockfish won the reverse game and the game pair was a draw.

Openings for the superfinal were prepared by Jeroen Noomen and GM Matthew Sadler, see the details here.

In game 1 KomodoDragon diverged from the agreed PV and chose to exchange knight for bishop on the 3rd move after book. Evals went under 1 and kept going down slowly. There were a few more exchanges and some shuffling, evals came down to 0 and the game was adjudicated when the pawns stopped moving.

In game 2 Stockfish delayed the minor piece exchange and evals stayed above 1. KomodoDragon gave a rook for a bishop and a pawn and opened the king side with the white king uncastled. The engines followed Stockfish's PV, identical to KomodoDragon's with a few traspositions. Stockfish gave a second pawn and castled its king despite the open files. Evals increased as Stockfish doubled queen and rook on the f file, KomodoDragon's pieces were passive and it preferred to leave the PV and exchange down to a RN vs BN position.

Stockfish focused on the king side. After exchanging a pair of pawns the black h pawn was isolated and on a dark square, so the black bishop could not defend it. After some knight maneuvering Stockfish managed to capture the black h pawn while KomodoDragon captured the white d pawn. 

Both engines created passers, but the white passer on the king side was more dangerous. KomodoDragon couldn't support its d pawn and gave it away, it used all its pieces to try to prevent the white passer's advance. Stockfish exchanged knights after a while, and its rook captured pawns on the queen side. The game ended in a tablebase win. Stockfish wins the game pair, it leads 1-0.

In game 3 KomodoDragon spent a lot of time thinking in the first moves after book, there were no exchanges and both engines played mostly behind the pawn lines. The engines exchanged a pair of bishops on move 20 and a pair of pawns on move 32, there was a series of minor piece exchanges in the center and the engines started to shuffle. Evals came down, KomodoDragon moved a pawn to avoid a 50-move draw but after a few more exchanges the game was adjudicated. 

In game 4 evals stayed around 1 and there were no exchanges until move 25 when the engines exchanged a pair of pawns. The center pawns faced each other and the engines prepared to open files. In the middle of a PV agreement on move 27 Stockfish changed its mind about the next black move, it recommended fxe4 followed by a knight for pawns trade and more pieces exchanged.

KomodoDragon continued with the original plan and played exf4. Stockfish's eval jumped over 1.5, the engines played out a long PV agreement, opened the center and exchanged queens. Still according to plan KomodoDragon gave a rook for a knight and pawn, its eval still around 1. There was another long PV agreement, this time Stockfish's eval started to increase and as the engines exchanged pieces KomodoDragon's eval also started to react. The engines reduced to a RBN vs BBN position on move 48.

There was only one pawn left for each engine, this looked like an ending only the very best engines could win. Indeed Stockfish made it look easy, it exchanged its knight for the DS bishop, then somehow managed to trap the black king on the back rank. KomodoDragon had to give pieces to defend its king but couldn't avoid mate. Stockfish wins the game pair, it leads 2-0.

Games 5-6 started with a 16-ply book, a popular line in the Blumenfeld countergambit (without b5) played at the highest level in human matches. In game 5 there were no exchanges after the start, Stockfish placed a knight on b4 and trapped the white LS bishop on b1. Stockfish's eval slowly came down, the first pawn exchange was on move 30. There was one minor piece exchange and the engines started to shuffle. One pawn move extended the shuffle to move 121, KomodoDragon moved its trapped bishop and Stockfish traded it for the knight. The game was adjudicated not long afterwards.

In game 6 the pawn structure was a little different, KomodoDragon also placed its knight on b4 and trapped the white LS bishop on b1, though Stockfish reacted with an eval increase. The first exchange was a minor piece trade on move 23, Stockfish's eval continued to increase as its attention shifted to the king side. Stockfish gave a pawn and opened the g file.

The white king looked exposed but the black pieces had little room to move and KomodoDragon was more concerned with blocking the white pieces from moving forward. Stockfish moved its king to the center and its knight to f5, and eventually traded rooks and captured the black h pawn. Stockfish managed to move its queen to the back rank, KomodoDragon was almost in zugzwang trying to hold the position.

KomodoDragon moved its king to the corner and the white queen followed. After a long period of repeating checks Stockfish captured the black f, a and b pawns. KomodoDragon exchanged queens to protect its king, and then it lost a knight. The material advantage was more than enough for a win.  Stockfish wins another game pair, it leads 3-0.

Games 7-8 started with a variant of the Petrov, Cochrane gambit, where white sacrifices a knight for two pawns and the black king moves. Black has the advantage after book. In game 7 (negative) evals came down from the start and there were only a few exchanges. Evals reached 0 and the game was adjudicated on move 37 with most pieces on the board. The engines repeated the first 11 plys in game 8. A few moves later evals started to come down, the engines played long PV agreements and exchanged pieces gradually. Evals were low enough for the draw rule from move 40, but the game was adjudicated only on move 77 in a drawn 7-man position.

In game 9 there were no exchanges after the start except a pawn pair that was exchanged in the book sequence. KomodoDragon had a space advantage and evals stayed around 1.5. For a while the engines seemed to be shuffling, KomodoDragon waited for the right moment to push a pawn on the queen side.

Evals increased as the engines exchanged pieces and opened a file on the queen side. Stockfish moved pieces forward on the king side while KomodoDragon shifted to the king side and opened it with its pawns. The black king was threatened and KomodoDragon managed to capture a rook for a bishop.

KomodoDragon created a passer on the queen side and traded queens. Stockfish held on for a while but couldn't prevent a queening after KomodoDragon created a second passer on the king side. The game ended in mate.

In game 10 KomodoDragon gave an early pawn and opened the queen side, a totally different approach compared to the previous game. The queens were off early and the white king moved without castling. In a long PV agreement the were many exchanges and evals hovered around 1.5. Stockfish considered the black move 25 a mistake, as it led to a RN vs RB ending. KomodoDragon's eval also increased two moves later.

I don't really know why evals were so high at this point, the white advantage seems minimal with only a pawn up and no passers. Perhaps the black bishop doesn't have enough targets to attack or pawns it can defend, perhaps the black king is more passive. Evals increased very slowly, first Stockfish traded pawns on the king side and created a passer. Then the white knight moved to the queen side and the black king followed, perhaps to prevent an attack on the a pawn. When Stockfish shifted back to the king side the black pieces were too far, Stockfisn gave the white a pawn and captured the black h pawn.

KomodoDragon couldn't protect the black g pawn and Stockifsh captured it after a while. The remaining black pawn was not a serious threat, nothing could stop the advancing white pawns. The game ended in a tablebase win. There were two white wins in this game pair, score is 4-1.