After 10 games Stockfish leads 3-1 with 6 draws. The score was 1-1 after the first two game pairs, then Stockfish won two more game pairs and it has a small lead. In game 10 Leela made a long term blunder in what appeared to be a drawn position, playing without thinking.
Openings for the superfinal were prepared by Jeroen Noomen and GM Matthew Sadler, see the details here. GM Sadler is also commentating many of the games live on his Silicon Road stream.
Games 1-2 started with a 22-ply book in the Sicilian Richter-Rauzer, Neo-modern variation, with white castling long and the black king in the center uncastled. In game 1 Stockfish opened the queen side and threatened the white king. Leela played a few moves in a PV agreement without thinking, Stockfish's eval came down. Leela opened the king side, Stockfish gave a rook for a bishop, its king was stuck in the center but safe enough. The engines started to exchange pieces, eventually reaching a drawn queen endgame.
In game 2 Leela did not open the queen side immediately, Stockfish had more time to strengthen the defense of its king and its eval went over 2 quickly. Leela's eval went over 1.5 after Stockfish opened a file on the king side, the black king was stuck in the center. The engines exchanged queens and were down to 3 pawns each. Leela thought for 30 minutes before moving its king to the king side, trying to build a fortress around it with its pieces.
The engines seemed to be shuffling for a while, Stockfish's eval increased while Leela's eval was more static. The white bishops controlled a lot of space, the black king side rook was trapped and the black pieces became increasingly passive. On moves 35-36 Stockfish moved its king, apparently without purpose, and suddeny Leela's' eval jumped over 3. Stockfish doubled rooks on the e file and went a pawn up.
Stockfish threatened mate on the back rank. The engines exchanged pieces until reaching a rook ending. Stockfish was a pawn up and the black king was isolated on the king side, Stockfish promoted the c pawn to a queen and won. Stockfish wins the game pair, it leads 1-0.
There were no exchanges after the start of game 3, Stockfish's eval started over 1. Stockfish opened the h file and exposed the white king, though its own king was also at risk. Stockfish attacked with a queen and knight, the white king seemed to be in danger but Leela defended well and evals continued to increase. After exchanging queens and a pair of bishops the black attack was over, Leela grabbed a pawn on the queen side and created doubled passers.
Leela captured a second pawn, after exchanging a pair of rooks only RBN vs RNN remained. Stockfish managed to create a passer and to capture one of the white passers. Leela was still a pawn up and its eval slowly increased.
Both engines used a knight to make sure the passers can't move forward. After exchanging rooks Leela gradually captured black pawns, including the c passer. It was 3 pawns up and Stockfish could not stop all the advancing white pawns. Leela promoted to a queen and won.
In game 4 Stockfish's eval was over 1 from the start. There were no exchanges and the engines played behind their pawn lines. On move 26 the engines locked the king side and started to shuffle. Leela's eval dropped to 0, Stockfish did not find a way to break the pawn wall. A few pawn moves and piece exchanges extended the game, Stockfish gave up and lowered its eval for adjudication on move 115. Leela wins the game pair, score is now 1-1.
Games 5-6 started with a 26-ply book in the Semi-Slav defense, Meran, Wade variation. Only one pair of pawns was exchanged, white had a central passer, the black king was uncastled in the center. In game 5 the engines played out a long PV agreement, Stockfish gave a rook for a bishop and pawn and castled on move 26, its eval slowly coming down. After the pawn structure stabilized the engines started to shuffle, evals came down and the game was adjudicated early. In game 6 the white passer reached the 7th rank before Leela captured it. The engines again played out a long PV agreement, this time the black king moved and did not castle, Stockfish's eval stayed around 1. The engines started to exchange pieces and evals came down. The game reached a rook ending, white was a pawn up but the position was a draw.
Games 7-8 started in the Czech Benoni defense, in both games Stockfish's eval went over 1.5 after book. The engines exchanged DS bishops early, the black king side rook moved and white castled long. In game 7 Stockfish gave a pawn and opened the queen side. Leela pushed pawns on the king side while Stockfish increased the pressure on the white king. After exchanging queens only rooks and knights remained on move 23, evals were under 1. A series of exchanges reduced to a RN vs RN position, Leela was a pawn up but it could only shuffle. After 30 moves Stockfish gave the knight for two pawns, Leela's eval came down and the game was adjudicated.
In game 8 Leela kept the queen side closed, and the engines started to shuffle early. Stockfish's eval was over 1.5 but evals didn't change for a long time. There was one pawn exchange on the king side, and the engines reduced to rooks and knights as in the previous game. It appeared that Stockfish had no plan of attack or improvement. Then on move 40 Leela moved a pawn on the queen side and Stockfish's eval immediately jumped over 2. After doubling rooks on the f file Stockfish pushed a pawn forward and opened the file, its eval over 3.5. Its eval continued to increase after the engines exchanged a pair of rooks.
Stockfish gave the e pawn so that its d pawn became a passer. It moved its two knights forward while the black pieces were passive as Leela tried to block the white rook. Stockfish was very patient, it was hard to see progress but evals increased steadily. Stockfish regained the pawn and finally managed to exchange rooks on move 85, then it went a pawn up.
The double knight ending was slow. Stockfish captured a second pawn, then Leela gave a knight to stop the white d pawn. The win was a matter of time. Stockfish wins the game pair, it leads 2-1.
Games 9-10 started with a 20-ply book in the Sicililan Najdorf, Adams attack. The engines castled in opposite directions and all pieces were on the board, with one pawn exchange. In game 9 Leela pushed pawns on the king side and Stockfish pushed pawns on the queen side. A black pawn reached a3 and a white pawn reached h6. Both kings were weak on the dark squares and both were threatened by major pieces on open files. Stockfish lost its DS bishop in a trade, Leela used its DS bishop for defense. The position became stable on move 26, Leela's eval slowly came down until the game was adjudicated on move 66.
The start of game 10 was similar, though both engines stopped their advancing pawns earlier. The DS bishops were exchanged and so were the queens, the kings were not threatened. Evals slowly drifted down as the engines exchanged pieces, on move 34 only RN vs RN remained and evals went under 0.5. On move 37 Leela moved without thinking, as it sometimes does. Stockfish's eval immediately jumped over 3, blunder ?!
Leela spent more than 40 minutes on its next two moves, and its eval jumped as well, over 7 after move 40. The advantage was not evidently clear, more subtle and long term, but both engines had no doubt it was there. Leela did not try to defend its a pawn, both engines were more concerned about the king side. After exchanging knights Stockfish captured the f pawn and Leela captured the white a pawn. Both engines pushed passers on the queen side and both were a square away from promotion.
Leela's rook was on front of its passer, the white king was strong enough to capture the passer on its own. As the white king walked to support its passer Leela had enough time to create a passer on the king side, but not fast enough. Stockfish queened first, it was winning however Leela chose to continue. Leela tried to avoid a tablebase win, it only delayed for a few moves. Stockfish wins the game pair, it leads 3-1.
No comments:
Post a Comment