Tuesday, November 29, 2016

Season 9 superfinal, games 63-66

After 66 games Stockfish leads by 11-5 with 50 draws. Stockfish keeps its +6 lead while Houdini collects another win, with the help of the opening this time.

The opening of game 63 was a sharp position from the Velimirovic line in the Sicillian. When the initial storm quieted down Stockfish was up a pawn with connected passers and an exposed king on the queen side, and Houdini had a passer in the center. Stockfish had an eval advantage that went up to 1, but after exchanging a pair of rooks the evals started to go down. In a QRB vs QRB position Houdini's passer was blocked on the 3rd rank and couldn't advance while Stockfish couldn't safely capture it. The game ended in a 3-fold repetition.
In game 64 the queens were exchanged early, Houdini tried to keep its advantage by giving a rook for a knight and pawns. After a series of exchanges only B vs BN remained and Houdini was 3 pawns up, connected passers on the queen side. As the game progressed it became clear that Stockfish's extra knight was not strong enough to stop the white pawns, but with opposite color bishops Houdini could only reach the 7th file with a pawn. In the end Stockfish could sacrifice its pieces for the remaining pawns to get a pawnless draw.

Game 65 started with an eval advantage for Stockfish that gradually increased to over 1. After a series of exchanges only QRB vs QRB remained in what seemed to be a balanced position.



Material was equal and the pawn structures were identical. However, the white rook was much better placed and it could immediately block the black rook from defending the b pawn. In the long run the black b pawn was doomed. Another weakness was the black king, too far from the action and under threat of a potential back rank mate. To exploit these weaknesses Stockfish exchanged queens and captured the black b pawn. 



After exchanging the bishops the white b pawn became too strong and there was no hope of a king side counterplay. Stockfish won a few moves later.
Game 66 started similarly, with an eval advantage for Houdini of about 1. The exchanges started earlier in this game, the result of which was again a QRB vs QRB position.


There were more pawns in this position, material was equal but Houdini had a passer. The black c pawn was hard to defend and the black king was a weakness as before. After Houdini captured the c pawn and exchanged bishops the action was around the white passer. Houdini's evals were already high in the following beautiful position.

Stockfish is almost in Zugzwang: the king can't move, the queen can't move without losing the rook, moving the rook allows Qf5. Houdini just waited while Stockfish made a few pawn moves on the king side, exposing its king. The game was over very quickly. This opening was too imbalanced, both engines found ways to win as white.

 

Monday, November 28, 2016

Season 9 superfinal, games 55-62

After 62 games Stockfish leads 10-4 with 48 draws. Houdini, with another win, is showing that it can beat Stockfish in some positions. It is just that Stockfish beats Houdini more often, it is +6 with 38 games to go in the match.

The opening for game 55 gave white a rook advantage and a trapped knight. The starting eval was in white's favor but not overwhelming, and as the game progressed evals went down a bit. A series of exchanges left RRB vs RBB on the board, a rook pair vs a bishop pair, can Houdini hold?

Stockfish made very slow progress, the evals climbed slowly as the rooks and bishops danced on the board. After 20 moves Stockfish exchanged a pair of rooks to get a RB vs BB ending.



Evals started to climb faster, Stockfish was getting sure it would win. To me it seemed the bishop and rook dance continued as before, but now Stockfish was able to get its queen side pawns moving, capture a pawn and create a passer. From there the path to a win was clearer. Slow and beautiful RB vs BB endgame win for Stockfish.
Game 56 started with the same material advantage for white, and evals started at 1 and went down gradually. This time when most pieces were exchanged Stockfish got the material back and Houdini's advantage was a bishop pair in a RBB vs RBN position. After exchanging bishops Houdini was able to capture a pawn and create a passer, still only good for a draw. Another demonstration of Stockfish's power in an imbalanced opening.

Game 57 started with the center pawns locked and evals close to 0. The engines mostly shuffled their pieces, exchanging all the knights and stablizing their pawns. After a few pawn exchanges the engines exchanged all rooks and Houdini's queen infiltrated behind the white pawns, leading to a perpetual check draw.
In game 58 it was the bishops that were exchanged off, and from move 20 the engines started to shuffle pieces with a few pawn moves to complete the pawns lines.



After a long shuffle there were a few pawn moves and the engines continued to shuffle. Stockfish's eval was at 0, but Houdini's remained positive. When Houdini saw the 50 move rule approaching its eval gradually decreased, reaching 0 and even negative values (as a result the game was not adjudicated). To avoid a draw Houdini exchanged the knights and opened the h-file. The game passed 100 moves and the engines had under 2 minutes to play, a blitz ! Houdini used the open file to exchange a pair of rooks and suddenly it saw something in the position and its eval started to jump.



Houdini combined threats against the black pawns and a potential mate with the queen on b7, Stockfish defended as much as it could but Houdini found the winning moves. It forced the black pieces back, then brought the queen and rook to the h file forcing a rook exchange. After that the a7 pawn fell and Houdini's pawn on a6 was too strong. All this playing on increments. Houdini proved again it is better in shorter games, recall it won the rapid tournament.

In game 59 Houdini sent a bishop deep into white territory where it got trapped for 8 moves while Stockfish was unable to capture it. The bishop managed to get out of the trap unharmed. After exchanging queens only RBN vs RBN remained on move 27. Evals fell gradually to 0 as the engines exchanged a few pawns, and the game ended in a 3-fold check repetition.
Game 60 reached a rook ending by move 27, evals were at zero a few moves later. The rooks continued to capture pawns until the game reached a tablebase draw.

After the opening in game 61 Stockfish was a pawn up with a queen side passer while Houdini had a supported passer in the center. Stockfish used its knights to block the central black pawns, and together with its own pawns formed a line across the board and started to shuffle pieces behind it. Houdini's line was broken, its pieces defended the entry points dynamically. The standoff continued until reaching a 50-move draw.
In the reverse game Stockfish's eval dropped to 0 immediately. After exchanging all the rooks the black queen moved forward and threatened Houdini's king on the back rank. Houdini was forced to counter with a perpetual check draw.

Saturday, November 26, 2016

Season 9 superfinal, games 45-54

After 54 games Stockfish is leading 9-3 with 42 draws. Draw rate is going up again at 78%. Houdini is playing well, but with more than half the match over it's quite clear that Stockfish is better at finding wins. Stockfish will win the match, Houdini needs a miracle to recover from 6 games down. 

In game 45 Stockfish had an eval advantage out of the opening, a few pieces were exchanged early including the queens, both engines had passers that could be dangerous later. Houdini's bishop was trapped and could only defend pawns.



The evals started to jump when Stockfish sacrificed its passer on the d file. Through the open file Stockfish exchanged all rooks, then it regained the pawn while exchanging the bishops. Only knights remained, and Stockfish was very happy with the position.


Black's c pawn could not be defended in time, and this created a 3 to 1 pawn advantage on the queen side. Stockfish blocked the black passer with its king and Houdini could not stop the white pawns.
In the reverse game Houdini had an eval advantage that grew to about 1, with similar material composition but different pawn structure. Stockfish simplified to a RB vs RB ending with opposite color bishops, and though Houdini was a pawn up it wasn't enough. The endgame was played for almost 90 moves before both engines agreed to a draw. Houdini couldn't find a win with the help af this unbalanced opening.

In game 47 Stockfish had a small eval advantage after the opening. When pieces and pawns started to be exchanged and the position opened the evals dropped to 0. The engines exchanged down to a BN vs BB position with one pawn each, resulting in a draw.
In game 48 the evals reached 0 even faster. By move 27 on RR vs RR remained, both engines had passers but neither engine could get an advantage and the game was adjudicated when the pawns stopped moving.

In game 49 the engines castled in opposite directions, Stockfish started to attack the king side with a pawn storm and the evals started to climb. Houdini reacted by exchanging everything and after move 24 only RB vs RB remained with Stockfish a pawn up and evals at 1. Evals continued to climb for a short while, Stockfish with a passer in the center, but after exchanging rooks the evals dropped back to 0. Houdini regained the pawn and exchanged the bishops to get a draw.
In the reverse game Stockfish reacted to the pawn storm on the king side by a counter attack on the queen side. The evals fell to 0 and the engines gradually exchanged pieces until reaching a QB vs QB position. The game continued for 25 more moves but the draw was never in doubt.

In game 51 after a few piece exchanges the engines kept their remaining pieces behind their pawn lines and started to shuffle. It took almost 100 moves for the game to finally end, after a long shuffle the engines broke the pawn lines and exchanged down to a drawn queen ending.
In game 52 the engines did not form pawn lines. Through an exchange Stockfish created an advanced passer in the center and gave a rook for a bishop to keep it. Evals were close to 0 even when the pawn reached the 2nd rank. Houdini gave the material back and exchanged pieces, capturing the black passer in an opposite color bishops ending, another clear draw. Third drawn opening in a row.

There were almost no exchanges after the opening of game 53, Stockfish maintained an eval advantage as high as 0.7. The position opened a little with a few exchanges, Houdini sacrificed two pawns but the evals fell to 0. After Houdini sacrificed a third pawn the evals started to turn to its favor as it developed a dangerous attack on the white king. Stockfish gave material to stop the attack, exchanging pieces and giving a rook for a knight so that only QBN vs QRB remained. Houdini pressed on getting evals as high as 0.5, but after queens were exchanged Stockfish's king was no longer in danger. The game continued for another 25 moves before a draw was agreed. This is the first time in the match that Houdini had a serious attack playing with the black pieces.
In game 54 Houdini's eval advantage lasted longer. By move 24 it was a pawn up and had two passers, though there were still many pieces on the board and the strength of the passers was not clear yet. The evals gradually decreased as the engines fought for control of the passers. Eventually Stockfish captured one but Houdini held on to the second and it got to the 7th rank. By this time only RRN vs RRB remained and Stockfish's eval was at 0. Houdini was forced to let the passer go and the game ended in a tablebase draw. Houdini played this opening better than Stockfish in both games, still the result is two draws. Four drawn openings in a row.

Friday, November 25, 2016

Season 9 superfinal, games 37-44

After 44 games Stockfish leads 8-3 with 33 draws. Since the shocking 4-0 start of the match after 16 games the result has been 4-3 (including one contested win for Stockfish). Houdini needs a winning streak of its own to have a chance of winning this match, doesn't seem likely though.

Game 37 started with an exchange-fest, only RRB vs RRB remained by move 20. The engines shuffled for about 30 moves with a few pawn moves, Stockfish kept a small eval advantage. When the engines started to exchange pawns and pieces the evals dropped to 0, and the game reached a drawn rook ending.
In game 38 the pieces stayed on the board longer. Houdini had a small eval advantage that gradually decreased, until after about 30 moves it jumped up close to 0.8. Indeed after simplification Houdini was able to win a pawn in a RRN vs RRB position. It had two queen side passers while Stockfish had one in the center. Houdini tried hard but couldn't improve its position, one pawn was not enough. Stockfish exchanged pawns and a pair of rooks to get a drawn RN vs RB ending. The draws continue, 5 openings in a row.

Evals stayed very low in game 39, the engines were ready to declare a draw from move 23. After the queen side pawns were locked the game almost ended in 3-fold repetition, then it was almost adjudicated. The second 3-fold attempt was successful.
In game 40 the evals also stayed low. The pawn structure was different, the queen side pawns became static early in the game but Houdini had a passer in the center. After exchanging queens both engines focused their attention on this passer, eventually Houdini didn't have enough support and the pawn was captured. A few moves after that a draw was declared.

Game 41 was a miniature for this match, a 3-fold repetition draw after only 29 moves. After locking the king side pawns the engines ran out of ideas and shook hands. Game 42 was completely different. There were only a few exchanges, each engine kept its pieces behind its pawns. The evals stayed low and after the king side was blocked the game seemed to be heading for a long shuffle and a draw. Instead Houdini's eval jumped over 1 and Stockfish's eval also moved up.


Houdini had a space advantage and could plan how to break the pawn line while Stockfish could only wait. Houdini took some moves to arrange its pieces and then launched the attack. As a result the queen side opened up, all the minor pieces were gone and Houdini was a pawn up.


Evals started to climb as Houdini focused its efforts on the weak queen side pawns. For the price of two king side pawns Houdini cleared the queen side and created two passers. To finish the game Houdini exchanged two rooks for the black queen and the pawns started to march, leading to a won Q vs RR ending. Finally a win after 14 straight draws.

Game 43 started with an eval advantage for white, the engines castled in opposite directions and Stockfish had a pawn advantage on the king side. After queens were exchanged not many pieces were left on the board, Stockfish was a pawn up and the evals were at 0.8.


Houdini's king moved forward to help in stopping the white pawns. After several exchanges only RB vs RN were left. While the black king was in the king side Stockfish moved its pieces to the queen side, creating a passer.


Houdini finally took out the king side pawn that was a threat from the start, but its king was too far when Stockfish's queen side pawns started to march. It took some time but eventually Stockfish got one pawn to the 7th rank, enough for a win. Stockfish immediately strikes back after its previous loss.
The reverse game developed similarly, white had a pawn advantage on the king side and an eval advantage. However, in this game Stockfish opened up the queen side so that when the game reached a RN vs RB ending Houdini did not have any pawns there. Stockfish managed to get a draw without difficulty.

Tuesday, November 22, 2016

Season 9 superfinal, games 27-36

After 36 games Stockfish leads 7-2 with 27 draws. There have been 9 draws in the last 10 games, draw rate is increasing to 75%. At this level this is to be expected even with careful opening choices, we wouldn't want really one-sided openings that cannot be defended.
We are still waiting for a decision on Game 17.

Game 27 started with a small eval advantage for Stockfish. There were many early exchanges and at move 23 only RB vs RN left. Stockfish was up a pawn with an ugly looking pawn structure. Evals were at 1, can Stockfish get something out of the ending?



After exchanging rooks the central group of doubled pawns became strong, it guaranteed a pawn majority in the center and it was an obstacle for Houdini's pieces. Evals were gradually rising, Stockfish thought it was doing well.



At this point Houdini was having trouble finding moves to play. Its king side pawns could not advance, its king and knight had to keep an eye on the white pieces. Stockfish attempted to get its king to g4 to help its pawns advance. Houdini countered with the c pawn and forced the white king back to the center. The c pawn became another target to defend and Houdini's pieces were already overworked. Houdini was forced to allow Stockfish to play f5 but then its position collapsed. After capturing the c pawn Stockfish's king could infiltrate the king side winning the game. Well played endgame by Stockfish.
In the reverse game more pieces remained on the board. Houdini had a bishop pair against Stockfish's knight pair, both kings were a little exposed to attack. Houdini kept a small eval advantage, Stockfish's eval dropped to 0 when a series of exchanges simplified to a RBB vs RR position. When Stockfish captured the last white pawn and gave a rook for a bishop the RB vs R ending was a draw.

In game 29 the engines exchanged almost all pieces after the opening, reaching a rook endgame on move 24. Stockfish had a small eval advantage for another 10 moves, but there was no real advantage on the board resulting in a short draw.
In the reverse game the engines kept a few more pieces on the board, reaching a QRN vs QRN position on move 25.  After queens were exchanged Stockfish's eval dropped to 0. Houdini remained optimistic much longer, it traded 3 pawns for the black knight but with no pawns left it couldn't win.

Stockfish did not castle in game 31, a locked pawn line in the center made its king relatively safe. Houdini's king side pawns were far from the king and Stockfish used an open g file to make attack threats. In the end both Stockfish's threats on the king side nor Houdini's counter on the queen side were effective, the engines exchanged rooks and simplified to a QBN vs QBB position. Evals dropped nearly to 0 and a draw was agreed.
In game 32 the engines exchanged almost all the queen side pawns, Houdini castled late to find safety for its king. When the f file opened the engines placed their rooks there, leading to the rooks being exchanged. Then the game reached a BB vs BN ending with evals close to 0 and another draw.

Game 33 reached a RRB vs RRB position by move 24. Stockfish had a small eval advantage for a few more moves, but after exchanging a pair of rooks and with opposite color bishops the evals fell quickly to 0. Stockfish had to give the bishop to stop a black pawn, still a draw.
In game 34 Stockfish's eval reached 0 even faster on move 16, though there seemed to be a lot of potential in the game. Houdini lost a rook for a knight but got an active attack on Stockfish's exposed king. The game was heading for a 3-fold check repetition when Stockfish changed its mind. The draw was achieved after the queens were exchanged, Houdini won back the material reaching a drawn rook ending.

There were many exchanges after the opening of game 35, and at move 30 only QB vs QN remained. Stockfish had a small eval advantage, yet it was Houdini that had a passer in the center. Houdini's king was more exposed and Stockfish used checks to drive it to the queen side and then capture pawns on the king side. Houdini had two queen side passers to counter Stockfish's 3 to 1 pawn advantage on the king side. Here Houdini played for a draw, it blocked the white pawns, found a protected spot for its knight and defended with its queen and king. Stockfish could not break this defense and after 30 moves it gave up and agreed to a draw.
Game 36 also started with many exchanges, after move 21 only QRB vs QRB remained. Houdini won a pawn while Stockfish continued to simplify, reaching an opposite color bishop ending. One pawn was not enough and Houdini couldn't break through the black line with all the black pawns invisible to its bishop. It took 70 moves for the engines to stop their shuffling and declare a draw.

That's 4 straight openings with only draw results. Is Stockfish playing defensively waiting for the match to end? Can Houdini close the gap? Still many games to go.

Monday, November 21, 2016

Season 9 superfinal, games 21-26

After 26 games Stockfish leads 6-2 with 18 draws. Game 17 is under review and the officials may change the result to a draw. 5-2 looks a lot better for Houdini than 4-0 did. Can Houdini make a comeback?

Game 21 started with a 7-ply book of the King's Gambit Accepted. As expected evals out of the opening favored black and the fans were anxious to see how Stockfish will handle being a pawn down with a king that can't castle. However, Houdini could not find a line that would improve its position and Stockfish equalized quickly. Unexpectedly, as the game progressed evals started to suggest Stockfish was better.



The advantage seemed to be due to having more space, and it was not clear how to proceed. Evals did not move for a while as the engines shuffled pieces.Then Stockfish made a small hole in Houdini's pawn line on the king side, with evals rising above 1. Stockfish used all its pieces to apply pressure on Houdini's position through the hole.

Still Houdini's position looked solid enough, it wasn't going to fall through a direct attack on the king. Instead Stockfish attacked the center of the board causing a series of exchanges that resulted in a knight for rook trade and a QRB vs QBN position.



Using the rook and queen Stockfish could check the black king and go after the weak black pawns. It took many more moves and was far from easy but Stockfish managed to eliminate the black queen side pawns, and after queens were exchanged its passer was unstoppable. Quite a surprising win for Stockfish. Does this game show that the King's Gambit is good for white?
In the reverse game Stockfish as black had an eval advantage for a longer time and its pieces stayed active. However, it too could not find a winning line and as pieces were being exchanged the evals drifted towards 0. After exchanging queens with only RBN vs RBN left the position became static and a draw was reached. A second good result for white in the King's Gambit, I vote for having more games with this opening in TCEC.

In game 23 Stockfish seemed to sacrifice a rook and then a knight, when in fact the knight could not be taken due to a mate, and a few captures later Stockfish got the material back. It was a pawn up with evals above 1 in a QBN vs QBN position from move 28. After exchanging queens Houdini created an advanced passer, but it couldn't hold it after bishops were exchanged as well. The remaining knight ending was a draw but it took almost 50 moves for the engines to realize this.
The reverse game 24 took a different path. Houdini had a supported advanced passer in the center and Stockfish had a 3 to 1 pawn advantage on the queen side. Houdini's eval was above 1 but it did not increase, Houdini couldn't improve its pawns position. Stockfish created two passers on the queen side, one of them advanced. When the king side pawns became static both engines started to shuffle and the evals started to go down. With evals close to 0 the ending was spectacular, Stockfish sacrificed a bishop to eliminate the white pawns, then with queening threats and an attack on the white king got the material back and reached a drawn RB vs R ending.

Houdini's king remained uncastled in game 25 but Stockfish was not able to get a useful advantage. Evals were at 0 from move 27 even though Stockfish had a bishop pair and a passer. The game reached a RB vs RB ending which led to a drawn tablebase position.
Game 26 was nothing like game 25. Both engines castled quickly and there were many exchanges after the opening. After queens were exchanged Houdini had an eval advantage, with better pawns on the queen side.

Here Stockfish chose to take the d pawn and abandon its a pawn. This looked dangerous because Houdini's b pawn became a passer. After some simplification only RB vs RB were left. Did Stockfish have the power to stop the passer?

Stockfish saw it was in trouble, evals jumping high. Houdini took longer to realize it was winning, it managed to queen the pawn for the price of a bishop, forcing Stockfish to lose its rook and reaching a won R vs B ending. Houdini's second win, it is not going down without a fight.


Saturday, November 19, 2016

Season 9 superfinal, games 17-20

After 20 games Stockfish leads 5-1 with 14 draws. Or is it 4-1?. Game 17's surprising outcome has caused the first heated fan argument of the superfinal, see details below. Houdini finally managed to win a game and Stockfish is no longer unbeaten in the season.

Game 17 started with a 13 move book of the Sicilian Dragon, Yugoslav line. Stockfish had an eval advantage and a strong king side attack, and it was willing to sacrifice a bishop to put pressure on Houdini's king.


In an amazing sequence of moves that both engines saw in advance Houdini sacrificed its queen and gained a rook, then Stockfish captured a rook, and then Houdini combined its rook and two bishops to force the white king to a8 (!!) and win back a rook.


As the game continued Houdini allowed Stockfish to capture its pawns while taking the king side pawn. Stockfish slowly advanced its queen side pawns and to prevent queening Houdini traded a rook for the two. Through all of this the evals gradually decreased, and for the last 10 moves both evals were 0. Then came the shock: the game ended in a table Q vs BB position and Stockfish was given the win. Inconceivable !! Both engines used 6-man tablebases and both evaluated the position at 0 in a 5-man tablebase win ???
It turned out that the reason was simple. The win is 72 moves long, with no pawns and no captures for more than 50 moves. Thus it was a tablebase win but a draw under conventional chess rules. TCEC rules use tablebase decisions without taking into account 50 move draws (so called 'cursed wins'). Both engines were playing with the 50 move rule logic and were unaware of this rule in TCEC.

What was to be done about this? Either keep the win for Stockfish or decide it was a draw after all. A replay was also suggested but that's a different question, since any game replayed will develop in new directions, so this should be done only if there was some technical problem with the game, not really the case here. Everyone had an opinion and for days people argued about this in the TCEC chat. Some of the arguments I saw:

Stockfish win:
a) Rules are rules. This was Martin's argument and it is very strong, especially when you consider the next argument.
b) This is a known TCEC incosistency. There were previous TCEC games that ended similarly and decided by tablebase despite violating the 50-move rule.
c) The 50-move rule should be ignored altogether in engine game endings, tablebases are the real truth and should always decide.
Draw:
a) Conventional chess and TCEC use the 50 move draw rule, and this should also apply to tablebase decisions.
b) Both engines considered the position a draw since they were using the 50 move rule in their logic. It is likely that the game would have continued otherwise if they knew they were heading for a white win. It is very probable that other drawing lines exist without using cursed wins.

Personally I think this should be declared a draw. I can't argue with 'rules are rules' though. In any case the match result will probably not be affected by the results of this one game.

In game 18 Houdini played with more caution, instead of an all out attack it used slow strangulation. Stockfish's pieces hardly had room to move while Houdini could plan the best way to proceed.



Houdini exchanged a few pieces and went a pawn up. The evals gradually increased to over 2. After exchanging queens only RRN vs RRN remained, Houdini's central passer looked strong.

Stockfish's position deteriorated fast. Houdini used a nice combination to fork a knight and a rook. Stockfish gave the rook for a knight and together with the advanced passer this was enough to give the win to Houdini.

There were many exchanges after the opening of game 19 and after 26 moves only QRN vs QRN were left. The engines continued to capture pawns, evals were 0 after queens were exchanged. The game ended in a tablebase draw. In game 20 the engines kept their pieces on the board and started to shuffle behind pawn lines on move 27. Stockfish had a bishop trapped in Houdini territory, but it had enough defenders so that Houdini couldn't capture it. The evals dropped to 0 when Houdini broke the blockade to avoid a 50 move draw, and the game reached a draw in a RB vs RB ending.

Thursday, November 17, 2016

Season 9 superfinal, games 9-16

After 16 games Stockfish is leading 4-0 over Houdini with 12 draws. Houdini has not had a game where it was close to a win yet, even with biased openings. Stockfish is very much in control of this match.

In game 9 Stockfish appeared to sacrifice a rook, but Houdini was careful not to accept. Instead the engines started a series of captures that resulted in the following position on move 20:

Material was even but the white pieces were much better placed and Houdini needed time to free its rook and pinned bishop. By the time the rook got out of its corner the evals were above 1 in a RB vs RN endgame. Stockfish played better, advancing its queen side pawns and eliminating the ones on the king side.


When Stockfish blocked the black king away from the pawns it was game over, nothing could stop the white pawns from queening. A third straight win for Stockfish in white.
In game 10 the engines exchanged many pieces as well, this time Stockfish gave a rook for a knight and pawns and this led to a RB vs NN on move 24. Houdini kept a small eval advantage for a few moves more before the evals collapsed to 0. The engines continued to exchange down until reaching a drawn tablebase position. Stockfish is 3-0 up after 5 openings, the match looks one-sided so far.

Stockfish had a small eval advantage after the opening of game 11. There were many early exchanges and only RRB vs RRB were left at move 26. Evals fell to 0 soon after and the game ended in a drawn tablebase position. In the reverse game the exchanges were different and the queens stayed on the board. Houdini had an eval advantage until it completely locked the king side pawns. Stockfish's eval dropped to 0 immediately and the engines started to shuffle. It took another 60 moves for the engines to agree on a draw.

Game 13 also started with a small advantage for white and many exchanges. The game reached a rook endgame on move 31 with evals at 0, and a drawn tablebase position 20 moves later. In game 14 the engines castled in opposite directions. Stockfish's eval dropped to 0 very quickly as it built a strong defense for its king. Houdini's pieces were scattered and its king a little exposed and the evals turned in Stockfish's favor. With only rooks and knights on the board each engine had 3 connected passers on opposite sides. For a short while Stockfish's eval rose above 1 when it went a pawn up, but the advancing pawns forced exchanges and the game reached a drawn rook ending.

There were almost no exchanges in the opening of game 15, each engine kept its pieces behind the pawn line. Stockfish had more space while almost all of Houdini's pieces stayed on the first 3 ranks. For almost 15 moves Houdini just waited while Stockfish slowly advanced its pawns across the board. It appeared that Houdini was safe enough, the evals got very close to 0.




Suddenly Stockfish's eval jumped, it considered Houdini's 33rd move an inaccuracy. After a series of exchanges the king side was open and Stockfish had a Q for RB imbalance. A few more exchanges opened up the queen side as well, the evals remained under 1.

The white queen was very effective in making threats and applying pressure on the black pieces, Houdini was forced to defend and gave a rook for a knight. The evals increased very quickly, and although Stockfish lost its pawns its strong pieces surrounded the black king and Houdini didn't have the power to stop them. Stockfish outplayed Houdini in this game, the result is now 4-0.
Game 16 also had pieces behind pawn lines and Houdini had space advantage. Here the king side pawns were locked and there were many exchanges in the center resulting in a RRB vs RRB position on move 28. After a pair of rooks was exchanged Stockfish's eval fell to 0, and Houdini's eval did the same a few moves later. The game ended in a tablebase draw.


Monday, November 14, 2016

Season 9 superfinal, games 1-8

After 8 games Stockfish is leading 2-0 with 6 draws. I wasn't expecting such a start, 2 decisive games in the first 8 and both Stockfish wins. After Houdini's strong performance in the rapid tournament I was expecting it be a tougher opponent, but perhaps in long time controls it hasn't improved a lot since stage 3.

This season Jeroen Noonen is in charge of the openings (Cato is still to blame, there's always a reason). The first four openings we've seen were all exciting and full of chances for both sides. We can hope that this trend continues, with as few boring draws as possible.

The engines have 3 hours each to think with increment but the games are usually shorter than 6 hours. We already saw this happen in the season 8 superfinal. At this level the engines recognize draws very quickly, evals converge to 0 and as a result we get short draws.

In the first game Stockfish sacrificed a knight and was sure it could get a 3-fold repetition. Houdini decided not to repeat, and instead traded its queen to get a Q vs RBN position. Stockfish's eval fell to 0 and Houdini's eval eventually did the same for a first draw.
In the reverse game Stockfish played a different defensive line and equalized very quickly. Houdini thought it was a little behind, perhaps because its king was a little exposed. Stockfish gave a rook for a bishop and pawns but the evals remained close to 0. The game ended in a 3-fold repetition.

Game 3 started with a small eval advantage for Stockfish but it didn't last long. There were many exchanges after the opening and by move 31 only RB vs RB remained. Houdini had an extra pawn but no passers. The game ended in a draw, as expected with opposite color bishops.
Houdini kept its advantage longer in the reverse game. It also kept its pieces and developed threats on Stockfish's king. The eval peaked at 1 (0.5 for Stockfish) but got no further. Stockfish hid its king behind a white pawn, and after sacrificing two pawns got a counterattack on Houdini's king. After massive exchanges only RB vs RB remained, and 25 moves later the engines agreed to a draw.

Game 5 started from move 18 of a well known Zaitsev line of the Ruy Lopez (deeply studied in Karpov-Kasparov in their world championship matches). Stockfish sacrificed a pawn on the queen side while shattering the king side pawns and exposing Houdini's king.


The evals rose to about 1 as the engines playes a series of exchanges, at the end of which Stockfish was up a rook for a knight and pawn. Houdini's king was not safe but had less pieces to worry about. Was the material enough for a white win?


Houdini managed to capture the white king side pawns leaving Stockfish with only one pawn in a QRN vs QBN position, and then Stockfish saw the win. After exchanging queens it could block the black king with the rook, allowing its own king to march to the center and eventually capturing the black pawns. After a few moves Houdini saw it was lost as well. First decisive game of the match, can Houdini win the reverse?
In the reverse game 6 Stockfish repeated Houdini's first move but surprisingly did not see its own attacking reply in the PV. Houdini continuation was less sharp and Stockfish shut down the attack very quickly. After several exchanges Stockfish created a queen side passer and Houdini countered with a rook sacrifice and perpetual check. Sadly it did not repeat and went on for 2.5 hours. Stockfish 1-0 after the third opening pair.

Game 7 started with another long book, 14 moves in a King's Indian opening. Stockfish started with an eval advantage of over 0.6, with pawns advancing on the queen side. Houdini's king side pawns were also advanced, threatening white's king while leaving its own king exposed.

Stockfish made a small hole in the pawn lines which allowed its queen to move closer to the black king. Houdini reacted by exchanging queens, evals rising to over 1.

Stockfish then used its queen side advantage, and the fact that the pawns blocked Houdini's pieces, to open up the queen side and create a passer.


Houdini tried to counter with its king side passers but they were easily blocked by the white king and knight. On the other hand Stockfish with a rook on the 7th rank put a lot of pressure on Houdini, the black king was too far for support and the rooks had little room to move. After exchanging a pair of rooks Houdini couldn't stop the pawn from queening without losing material. Second straight win for Stockfish playing white.
In the reverse game Houdini chose a different path, where it had no advantage on the queen side while Stockfish developed a dangerous attack on the king side. However, evals stayed close to 0 except for a few moves Houdini evaluated as a black advantage. In the end Stockfish's king side attack turned into a perpetual check and draw. That's the second opening in a row that Stockfish uses the opening advantage to find a win while Houdini plays with more caution and gets a drawn game.

Friday, November 11, 2016

Season 9 superfinal has started !!

After a very long wait the season 9 superfinal is on its way. New versions of Stockfish and Houdini will play 100 games to decide the TCEC champion. Can Houdini make a magical comeback to the days it was the best engine in the world? It won the rapid tournament convincingly. Will Stockfish repeat its stage 3 performance and remain unbeaten for the whole season? After two losses to Komodo in the season 7 and season 8 superfinals it has a chance to win a second title, Komodo hasn't even qualified.

The season 9 superfinal was delayed for three weeks, and very little information was given to the fans about the reasons for the delay. Anton from chessdom did not post updates in the chat or on the facebook page and the fans were getting restless. Even Martin was not happy, it really felt bad to see the founder not happy with its creation. In the end a new server was obtained, this time with 44 cores, both engines submitted new versions and the games are back. The fans are happy, I'm happy, go TCEC.

The superfinal started at the exact same moment the Carlsen-Karjakin world championship started. Coincidence? I don't know but it is going to be hard to keep up with these two events, the best human and computer chess have to offer at this time.

Tuesday, November 1, 2016

Season 6 superfinal, games 41-48

This is part 8 of my looking back at the Komodo-Stockfish superfinal match of season 6. Previous parts can be found at
part 1, games 1-8
part 2, games 9-16
part 3, games 17-24 
part 4, games 25-28
part 5, games 29-32 
part 6, games 33-36 
part 7, games 37-40

After 40 games Stockfish led 9-3 with 28 draws.

Stockfish was very happy after the opening of game 41, it had a knight for two pawns advantage and a queen side passer, as well as an eval over 1. Komodo had strong central pawns and as they advanced they started to look dangerous. Komodo then allowed Stockfish's knight to make a QR fork and gave a rook for the knight. Stockfish's eval jumped over 2 but Komodo saw a way out. Starting from move 41 both engines agreed on a 10 move sequence that removed most threats and left a QB vs Q ending. The evals were high but with almost no pawns left Komodo's queen was sufficient defense and the engines shuffled until a draw was agreed.
The reverse game was a lot less eventful. Stockfish gave two pawns on the king side to Komodo's queen, and this allowed Stockfish to force a repetition draw by attacking the queen.

In game 43 Stockfish again had an eval advantage after the opening, with a pawn up and a queen side passer. Evals clmbed to about 1 and stayed there for a while, but after several piece exchanges evals came down again. The game reached a drawn queen ending.
In the reverse game 44 Komodo started with the same pawn advantage and queen side passer. Stockfish exchanged queens and equalized before move 20.

Stockfish kept the initiative, got rid of Komodo's passer and the evals climbed above 1. After several exchanges Komodo's pieces could only hope to defend and block against Stockfish's threats.


For more than 15 moves it seemed Komodo was holding on, the engines were shuffling and the evals remained constant. Then after what appeared to be two blunders Stockfish exchanged B for N, the evals shot up and Komodo got itself in a world of trouble.

The c3 pawn was pinned and could not be defended. A few moves later Komodo's position collapsed, it faced queening threats and was about to lose material. Did Komodo really blunder the ending or was it going to lose in any case?

There were many early exchanges in game 45, with only RRB vs RRB remaining after move 25. Stockfish was a pawn up but with opposite color bishops that was not enough. After the rooks were exchanged the position was a drawn bishop ending. In game 46 again Stockfish was ahead with evals over 1.5 after sacrificing a bishop for pawns in return for a strong queen side attack on the white king. Komodo gave its queen for a rook and got out of immediate danger. In the RRN vs QR position it seemed the engines were shuffling with no real progress and evals did not move much. It took the engines more than 50 moves to realize it was a draw and 10 more for adjudication.

After the opening of game 47 Komodo kept playing moves Stockfish didn't expect. Komodo's eval stayed close to 0 while Stockfish's was close to 1. Then on move 21 Stockfish had enough, this time its eval shot up to over 2.


By playing d5 Stockfish blocked the center, making it hard for Komodo to move pieces around. This move also chased away the knight and opened up the diagonal for the bishop on f2. A few moves later Komodo realized it was in trouble after Stockfish sacrificed the bishop and exposed the black king.


Komodo could not defend against all the white pieces converging on its king. The game was over by move 33 before Stockfish gained enough material to force a win. It's very unusual to see such a short decisive game at this level.
The reverse game 48 developed along very different lines. Both engines castled to the queen side, Stockfish opened the e file and exchanged rooks and queens through it. The evals were never far from 0 and the game continued until the engines reached a drawn tablebase position.

In the last 4 openings Stockfish got 2 wins and 2 more games where it had a significant advantage and Komodo managed to draw. Komodo could not achieve anything in these openings, it is now trailing 11-3 with 34 draws and has no practical chance to win the match.