Wednesday, November 29, 2017

Season 10 superfinal, games 41-50

After 50 games Houdini is in the lead, 7-2 and 41 draws. Komodo got its second win of the match, this time drawing the reverse. We are at the halfway point of the match, Houdini is the favorite to win but Komodo made it more interesting, not letting Houdini get too far ahead.

The Komodo development team found a compilation problem that caused the version playing in the superfinal to be slower than the previous version that played in stage 2. After a short discussion it was decided that the current Komodo version will continue playing. A full description of the subject can be found in on chessdom.

Evals were very low in game 41, both engines spent a lot of moves developing their pieces with little contact between them. The queen side pawns were removed and the engines exchanged pieces through the open files, until only N vs B were left and a draw.
Game 42 started with low evals but with a slightly different pawn structure. The queen side pawns stayed on the board and on move 23 Komodo created a central passer and gave Houdini a pawn majority on the queen side.



Houdini pushed its queen side pawns, with all major pieces supporting and defending. Evals gradually increased until the tension was released with exchanges, leaving only RN vs RB with one queen side passer for Houdini.

The knight was better than the bishop in this endgame. Komodo's pieces were committed to stopping the b passer, leaving Houdini to adjust the pawn structure to its favor. Houdini diffused the black central pawn pair with one exchange, captured one of the g pawns and finally gave up its passer for another central pawn. The next pawn capture was brilliant and simple.



Komodo can't retake because it loses the bishop! With an advantage of two pawns it was easy for Houdini to convert the win.

Game 43 started with a series of minor piece exchanges as well as queens, that left Komodo with a pair of knights and a pawn to Houdini's bishop pair. Evals were at 0.5 and increasing. Komodo then gave a rook for a bishop and opened the queen side. After the rooks cleared away some of the pawns Komodo had a pair of connected passers, two pawns up in a RNN vs RRB position. Evals were close to 1 but got no higher. Houdini blocked the passers, captured one and gave a rook for a knight to stop the other. In the end the game reached a drawn rook ending.
The first 16 moves of game 44 followed game 43. Houdini kept its rooks and the position remained closed. The evals reached 1 like in game 43 but the engines could only shuffle. Komodo exchanged all the minor pieces and got a double rook ending. Evals came down to 0 and after taking a few pawns the engines agreed on a draw.

Komodo had a small eval advantage after the start of game 45. Then the engines started to exchange everything, leaving only QR vs QB. Evals came down and in an open board both engines had mating threats and the possibility of perpetual check. A draw was agreed after a while. Game 46 also had extensive exchanges after the start, this time the queens were first to go and the remaining pieces were RBN vs RNN. Evals were close to 0 but it took 40 more moves to get a draw.

Evals were low in game 47 though both engines were active. The kings castled in opposite directions and both faced pawn storms. The attacks on both sides turned out to be non-effective, and after the queens were exchanged and only RRN vs RRN remained the kings were safe enough. The game reached a drawn rook ending.  In game 48 Komodo did not castle and evals slowly increased  to over 0.5. Houdini gave a knight for 4 pawns leaving the black king unguarded in the center. Evals remained low while Komodo exchanged pieces, its king out of danger. The white pawns looked intimidating, Komodo just gave the knight back to get a double rook ending with Houdini a pawn up, enough to get a draw.

Komodo had a small eval advantage after the start of game 49, with two knights against Houdini's bishop pair in a mostly closed position.



Komodo considered Houdini's 22nd move as too slow and immediately attacked the king side. As a result Houdini's king side opened up and its king tried to hide behind a white pawn. After exchanging queens the attack was over, Komodo was two pawns up with evals close to 2. Houdini still had the bishop pair advantage, but was it enough?

Komodo's central pawns started to march forward and Houdini was forced to give a bishop to get rid of them. Houdini had a pair of connected passers on the queen side, but with a piece down and the wrong bishop it couldn't support them.

The queen side black pawns were doomed, as was the white h pawn. However the remaining white pawns were invisible to the bishop, and Komodo had the time and the extra piece to hit the last black pawn. The game was adjudicated before we could see this play out. A second win for Komodo, this time the opening was quite balanced. Can Houdini strike back immediately?
Evals stayed low in game 50. There was a complicated struggle on the queen side, full of pins and offered pieces not taken. Komodo made a real knight sacrifice to get a pair of connected passers in the center. After a series of exchanges Houdini countered with a passer on the queen side while its king was exposed. The action was over with more exchanges, Houdini gave its knight to get rid of Komodo's passers and a tablebase draw. No comeback for Houdini this time.

Tuesday, November 28, 2017

Season 10 superfinal, games 31-40

After 40 games Houdini is in the lead, 6-1 and 33 draws. Komodo finally got its first win of the match. Houdini won the reverse game so the opening may have been too biased. Getting closer to the halfway point it is very likely that Houdini will win the match.

Houdini's eval was negative right from the start of game 31 despite its uncastled king that was being attacked, possibly because of the space advantage it had on the queen side. After a few moves Komodo took over, evals climbing above 0.5. Houdini defended well and in the end all Komodo got out of the attack was an exposed black king and the possibility of repetition or perpetual check. In game 32 Houdini did not attack the king side, the position closed up and the engines were shuffling from move 31. After 30 moves the engines started to exchange pieces until they reached a drawn rook ending.

Game 33 started with an eval advantage for Komodo. After playing a few moves Komodo gave a rook for BN, and Komodo is considered a master of imbalances. After the queens were gone the board looked too empty for creating any advantage. However evals increased to over 1 when Komodo exchanged down to a RBN vs RR position with Houdini two pawns up and one of them doubled. Unfortunately for Komodo it only managed to get two more pawns. Houdini's rooks found an unbreakable defense and the rest of the game was just waiting for the draw rule. A rare chance for Komodo, still a draw. In game 34 the material stayed balanced and evals close to 0. The center was blocked and Houdini created connected passers on the queen side. Komodo responded by exchanging everything until reaching a drawn bishop ending.

Game 35 started with an eval advantage for white, Komodo pushed pawns on the king side and Houdini kept its king uncastled for a while. Eventually Houdini found it safe enough to castle, evals came down gradually and with only QRB vs QRB left the engines started to shuffle. The engines agreed to a draw after exchanging queens. Queens were off early in game 36 and Komodo felt safe to castle the king. Houdini's eval was over 0.5 for a while, after exchanging down to a RN vs RB position the evals came back down. In the end both engines queened pawns, but the game still ended in a draw.

Game 37 started with black two pawns up with its queen in white territory. Komodo used attacks on the queen to gain a development advantage and equalize. Komodo opened a file on the king side and stablized the queen side, turning the evals to over 0.5 in its favor. With a king side attack Komodo captured pawns and material was equal again, it even managed to be a pawn up. However Houdini defended effectively and was never in any real danger. After exchanging queens only RN vs RN were left and eventually the evals came down and a draw was agreed. In game 38 Komodo turned the two pawns it was down into a rook for bishop imbalance. Houdini had a knight trapped in the corner and Komodo had an advanced passer that got to the 2nd rank. Evals remained low even with all the action on the board. Eventually the engines removed all threats and balanced the material, reaching a draw in a QRB vs QRB position.

Game 39 started with an eval advantage for Komodo. The center was blocked with pawns, there was a small opening on the queen side and the king side could be opened up whenever the engines decided.

The engines shuffled for a while, many knight and king moves on both sides. Komodo applied some pressure on the open queen side, other than that nothing seemed to be happening. Yet the evals increased gradually to over 1, something subtle was going on. When the position opened with a few exchanges Komodo was a pawn up. Could Komodo find a win in this position?

At first Houdini seemed to be holding, Komodo did not seem to have a plan. Evals jumped over 2 when Komodo placed a knight on e6 after bishops were exchanged. The pieces continued to dance on the board, Komodo exchanged rooks and then queens with evals over 5.


Houdini couldn't escape this position. Komodo soon captured another pawn and with 2 pawns up the win was just a matter of time. First win for Komodo in the superfinal after 39 games, finally the Komodo fans have something to smile about.

In game 40 Houdini's evals climbed right from the start, over 1 already on move 16. Although material was equal Komodo's pieces could hardly move and Houdini had time to plan when and where to attack.



The queen side was blocked and for a short while Komodo dreamt it could hold a fortress. Houdini's response was to sacrifice a knight for two pawns and open up the king side. Komodo's position collapsed very quickly, with so little room to move its pieces were stuck on the wrong side of the board.



The game was adjudicated with material almost balanced, before Komodo had to lose material to avoid being mated. This was the first opening of the season (including stage 2) with two wins for white. This time Houdini fans could claim that the previous loss was due to the opening, not only Komodo.

Sunday, November 26, 2017

Season 10 superfinal, games 21-30

After 30 games Houdini leads 5-0 with 25 draws.
Houdini continues to dominate the superfinal, Komodo has not been close to winning a game yet. There is still a long way to go but the trend is getting clear and it is hard to view this as statistical noise.

Game 21 was balanced with a very small advantage for black. The evals stayed close to 0 throughout the game. All the minor pieces were off the board by move 38 and the game ended in a repetition. In game 22 Houdini equalized immediately and most of the game the engines had positive evals that got as high as 0.5. The queen side was open, for a long time the engines shuffled without making contact. Then the exchanges started and soon the game reached a rook ending and another draw.

Houdini willingly walked into a trap in game 23, giving a rook for a bishop and pawns. Evals were low, and Houdini used an attack on the white king to exchange most of the pieces. In the end only a white knight remained, black's two extra pawns secured a draw.
There were almost no exchanges in the opening of game 24. Houdini's eval fluctuated within the 0.4-0.8 range, Komodo was more conservative. On move 26 Komodo chose to exchange pieces, winning a rook for a knight while exposing its king to attacks. 



Komodo defended its king at the price of losing all the king side pawns. Houdini's pawns looked intimidating as they started marching forward. The evals continued to increase above 1.5.



Komodo had to give a rook to stop the white king side pawns, and the RBN vs RB ending was easy for Houdini to convert. The remaining black pawns were on dark squares where the bishop couldn't assist, and they were soon gone. A knight and two pawns up was enough for a win.

In game 25 there were only a few pawn exchanges in the first 25 moves. Evals stayed low as the engines started to exchange pieces. The game reached an opposite color bishops ending and a draw. In game 26 the queens were exchanged early and the engines closed the center. The engines exchanged pieces through the open flanks, and only RBB vs RBN were left on move 31. The rest of the game was mostly shuffling until the draw 60 moves later.

Queens were exchanged early in game 27, there were many pieces exchanged and many open files. By move 29 only RNN vs RBN remained, the game ended in a draw after most pawns were taken. In game 28 Houdini as white kept the game alive longer with evals of about 0.5. However, after a few exchanges the evals came down, the advantage was not realized on the board. The game reached a king and pawns ending and another draw, despite black queening a pawn.

In game 29 the center was blocked with pawns, Komodo had a small eval advantage though its king remained uncastled. Houdini started a series of exchanges that led to a QN vs RRN position, and then used irs rooks to force a perpetual check draw. In game 30 there was more action along the queen side. Komodo's eval were at 0 right after the start, Houdini agreed not long after. The engines exchanged all the queen side pawns and most pieces. The game ended in a perpetual check in a QN vs QN position.

Friday, November 24, 2017

Season 10 superfinal, games 11-20

After 20 games Houdini leads 4-0 with 16 draws. So far Houdini is dominating the match, Komodo does not seem to find winning lines at all.

In game 11 the center and queen side were blocked by pawns, Komodo made an attempt to break the wall on the king side. Evals climbed to 0.8 for a short while, however after several exchanges only RRN vs RRB remained and the position remained mostly blocked. The evals slowly came down in an endless shuffle. Komodo sacrificed a knight for two pawns, only extending the shuffle.
Game 12 started in a similar manner, this time pawns blocked the entire board by move 22. Houdini's eval was over 1, apparently due to a space advantage.




The engines started to shuffle. Essentially nothing happened for the next 65 moves except a few exchanges, evals came down and the engines had very little time left. Around move 100 Houdini sacrificed a knight for two pawns and opened up the center. Komodo saw no reason for alarm and evals remained low, but now Houdini had two connected passers that looked dangerous.



By this time the game was in blitz mode and Houdini showed its superiority. It exchanged knights and was able to advance the pawn to e7. Komodo's pieces had to stop the pawn, leaving Houdini's queen to capture the black pawns. When the white pawns started to march Komodo was defenseless. A third win for Houdini, this one was quite unexpected. This was the kind of win that Komodo usually finds, not this time.

Komodo started with a small eval advantage in game 13, about 0.5 when on move 22 it had a bishop pair to Houdini's knight pair. A massive exchange in the center left only RBB vs RNN and evals started to come down. After exchanging a pair of rooks the engines agreed on a draw.
Game 14 developed very differently. Houdini's eval started to increase from the start, reaching 1 already on move 12. Komodo had an under developed queen side, and on the king side Houdini pushed pawns threatening the black king. Houdini placed its queen on the long diagonal and opened the center, giving its queen a direct line to the black king.



The combination of queen and bishop on the long diagonal was deadly. Komodo lost two pawns to escape being mated, evals above 2 and continuing to increase.



After exchanging queens Houdini used its king side pawns as a threat, capturing pawns on the queen side. The game was over quickly. Another win for Houdini in white, playing the opening much better than Komodo did in the previous game.

In game 15 Houdini exchanged queens early and the evals remained low. The game reached a RRN vs RRN position on move 26. Houdini's path to a draw was to give its knight for two pawns, leading to a RN vs R ending with no pawns for white. In game 16 Houdini created an eval advantage after the opening, but a long series of exchanges left only QR vs QN on the board with Komodo 2 pawns up, and the evals came down to 0. It took another 30 moves for the draw rule to stop the game.

Game 17 started with opposite side castling and a small advantage for black. Komodo pushed pawns on the king side and Houdini on the queen side. The position was open and both engines had attacking opportunities, but the evals stayed low. The game ended with a series of exchanges exposing both kings, a queening and finally a perpetual check. I had a live blog for game 18. Here Komodo went for a R vs BN imbalance early in the game. Houdini's king hid in the corner behind a black pawn, and the rest of the pieces were concentrated on the opposite corner with Komodo defending. Evals turned to Houdini's favor but it couldn't find a way to break Komodo's defense. Houdini managed to win a knight, and a series of exchanges led to a QR vs QB with Komodo 2 pawns up, another draw.

Game 19 started with an eval advantage for Komodo which came down almost immediately. Komodo won a pawn on the king side while Houdini attacked the queen side with all its pieces (both kings were on the queen side). After queens were exchanged evals were close to 0, Komodo's extra pawn offset by Houdini's bishop pair. The engines exchanged down to a drawn rook ending. In game 20 Houdini was quicker in the attack and evals grew quickly. Komodo had to defend the king in an open queen side and Houdini had a passer on the king side. Komodo exchanged down to a RB vs RB position with Houdini two pawns up. Evals were over 1.5 with passers for Houdini on both sides, yet somehow Komodo managed to set up its remaining pieces so Houdini couldn't advance (the opposite color bishops helped). It took 80 more moves to get Houdini to take the draw. Komodo magically escaped this game.

Thursday, November 23, 2017

Season 10 live blogging, Thursday Nov 23

Today I'm looking at game 18 of the superfinal between Komodo and Houdini. Komodo is struggling in this match so far, Houdini is 4-0 which to me is a surprise. Not only the result but the way that Houdini finds winning lines and Komodo doesn't in the same openings.

Game 17 just finished and there appears to be a small advantage for black in the opening. As white Komodo managed to get a draw without too much trouble. Can Komodo find a better line as black?

move 11: Komodo diverges with a5, evals are negative but lower than in game 17. Better?

move 12: Houdini is pushing pawns on the king side and Komodo on the queen side. Evals still negative but really low.

move 13: Komodo's eval is at 0! It took the pawn on g4, opening a file on the king side could be a problem here. Komodo sees a repetition draw in the distance in its PV, I never trust these predictions.

move 14: Houdini follows Komodo's PV and takes the h pawn, its eval goes the other way so not a blunder by Komodo yet.

move 15: Houdini moves Ng3 and not Nf4 as was expected, and it is still not happy with -0.27. Komodo replies with b4 on the queen side, with eval -0.34, seeing an R vs BN imbalance.

move 17: Imbalance achieved, now what? Both engines hiding their PVs. Earlier Komodo predicted exchanges in the middle, Houdini is thinking about it.

move 19: Houdini thought Komodo's Nh6 wasn't good, but Komodo responded with eval -0.61. Houdini's PV looked like a transposition of Komodo's but then Komodo changed its mind. Some action ahead? A long think by Houdini.

move 21: Houdini continued with its plan, its eval stays low. And then Komodo's eval goes down too, trying to open up the queen side with another pawn sac?

move 22: Houdini didn't take the pawn and will hide its king behind it. Evals are back up for both engines, can't tell what is going on.

move 23: Houdini's eval jumped to positive, really strange. Komodo is thinking, if it also doesn't believe its own eval then I guess it won't win here. OK, Komodo changed its move and eval down but still negative, it still has some hope.

move 25: Komodo is trying to activate its pieces, still holding on to a small eval. It's hard to believe we're going to see fireworks in this game.

move 26: again a long think by Houdini, eval jumping to positive again. Komodo still not impressed, moves the queen, still needs to activate more pieces.

move 27: positive or negative, who has the edge? The engines keep playing moves not expected by their opponent. Which of them sees more clearly?

move 31: Houdini refusing to exchange queens, its eval up to 0.28.

move 36: Komodo's eval is now also positive, both engines agree that Houdini is better now. Can Houdini pull another rabbit from its hat?

move 40: engines' PV agree on a long series of moves that simplify the position a little. Evals stay positive but not increasing just yet.

move 47: Houdini left the PV at move 46, preferring to grab a piece. The imbalance is now R vs B and Komodo 2 pawns up. More simplifications ahead.

move 54: we're down to QR vs QB, and Houdini's eval is negative again, very close to 0. This will be a draw, unless we have a crash...

move 58: draw adjudication. The evals gave us something to look at in the game, on the board the game wasn't all that interesting.

Season 10 superfinal, games 1-10

The season 10 superfinal has started, a 100 game match between Houdini and Komodo. Stockfish did not qualify for this season's superfinal, its fans disappointed. The chat is much quieter, this is probably related - Stockfish fans can be unpleasant when it is not winning. The time control for the superfinal is 120+15, so the games are 4+ hours long.

After 10 games Houdini is in the lead, 2-0 with 8 draws. Komodo has had one game with a significant chance of winning so far, but couldn't convert to a win. Still too early to draw conclusions.

In the first two games there was no book, the engines had freedom to play as they liked. Game 1 was a Bogo-Indian opening. The king side was closed and the action was focused on the queen side and the open c file. After exchanging a pair of rooks on move 28 the evals became close to 0, the engines started to shuffle. A few more exchanges led to a QN vs QB position. Houdini's queen side pawns looked dangerous but Komodo forced a perpetual check draw.
Game 2 was a French opening. Komodo as black was two pawns up on move 15, with Houdini's queen side free of pawns. Komodo's king remained uncastled though it later marched to the queen side. Evals were in black's favor due to the material advantage, while Houdini applied a lot of pressure on the black king and Komodo had to defend. The engines started to shuffle from move 30, with a few pawn moves that reset the draw counter. It took a very long time for Komodo to give up its pawn and eval advantage, draw was reached only on move 185.

In game 3 Komodo had a small eval advantage, Houdini's king stayed uncastled and its pieces had less room to move. Komodo castled long and Houdini pushed its queen side pawns and applied pressure with its major pieces. Komodo gave a pawn to open the king side but didn't get any meaningful attack. Houdini equalized and the game ended in an early repetition draw.
In game 4 Komodo castled its king and did not push pawns on the queen side. Houdini pushed its king side pawns and the evals increased to over 1. The engines exchanged minor pieces until only the majors were left on move 30.

Komodo's pawns on the queen side were weak and easy targets, Houdini's pieces were very active and made multiple threats. After exchanging queens Komodo couldn't hold on to all the pawns and Houdini was able to capture one.

Houdini now had a passer, and with Komodo busy stopping it Houdini soon captured more pawns, and the game was over. First win of the superfinal goes to Houdini.

The position was balanced after the opening in game 5, evals close to 0. Houdini had better pawns on the queen side, Komodo concentrated its pieces there to stop them. Both engines had passers, all were gone together with most pieces and the game reached a drawn rook ending.
In game 6 Houdini was much more aggressive from the start as white. It attacked the king side with pawns and pieces, exposing its own king in the process. The evals fluctuated for a while, eventually they reached 1 and stayed there.



Houdini's advantage was not clear on the board. It seemed the engines did not have a plan as they shuffled for a while, yet the evals kept increasing. After exchanging queens Houdini won a pawn on the queen side with evals over 2.7.



It was only a matter of time until Komodo's remaining queen side pawn was gone. With a passer and two pawns up Houdini's win was guaranteed. Two wins in a row in white for Houdini.


Game 7 started in a closed position and a small eval advantage for Komodo. By move 24 all the queen side pawns were gone and the engines were mostly shuffling. Most pieces were exchanged after a while and evals came down to 0, nothing exciting happened until a draw was agreed. In game 8 the pawn structure was slightly different, the a and b files were clear and the center was blocked by pawns. It took longer for the engines to exchange pieces, Komodo's knights were active on the king side for a while. However this game also had a shuffling phase and it ended in a QN vs QN draw.

Komodo started with an eval advantage in game 9, which increased to over 1 despite being a pawn down. On move 24 Komodo surprised Houdini and started a series of exchanges that left only Q vs RB on the board. Komodo evaluated the imbalance at almost 2, but it turned out that Houdini had an unbreakable fortress. The white king was trapped in the corner and the queen could not win on its own. Still, the first game where Komodo made a significant attempt to win a game. In game 10 Houdini did not get any advantage at all, Komodo equalized very early and evals were close to 0 by move 23. The game was adjudicated in a drawn rook ending.

Monday, November 20, 2017

Season 10, stage 2 stats

Draw rate, wins



Final draw rate was 67%, better that expected for the stage.



Game termination

The three most common game termination causes were:
37.5% - TCEC draw rule
31.3% - TCEC win rule
15.2% - TB position

There were two crashes by Andscacs and Booot. As a result of Andscacs' crash in a winning position it was left out of the top 6 while Ginkgo is in.

Moves per game



Median= 59.3
Average= 63.6

The distribution is similar to stage 1, about 9% of the games were longer than 100 moves.

Time per game (hours)


Median= 2:55
Average= 2:49

The engines in stage 2 had half an hour more time per game, compared to stage 1. The games were slightly less than an hour longer.
 
Openings

There were variable length book openings in this stage chosen by Cato.
The first letter of the ECO codes was distributed as follows:


An almost uniform distribution.

If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:

Sicillian - 12 times
King's Indian - 10 times
Spanish - 10 times

Some of the book lines were short enough to allow different ECO codes in reverse games, this happened in 10/56 paired games.

Reverse pairs, wins


There were no biased openings in the stage, an excellent result though Cato stated that this may be indicate that the openings chosen were not biased enough. There were decisive results in 61% of the openings.


Reverse pairs, same moves 



Reverse games diverged very quickly, 41% diverged immediately after book, 68% diverged at most after two plys.

Season 10, stage 2, final results

Final results of stage 2:



Houdini and Komodo move on to the superfinal of season 10. Stockfish finishes a close 3rd only half a point down. Andscacs and Booot finish in 7-8th, will therefore be playing in division 1 in season 11. Fire, Chiron and Ginkgo will join the top 3 and will play directly in the premier league, together with the two qualifiers from division 1 (The new format for season 11 was published on Chessdom).

Komodo finished strong with 5/7 in the last RR. Houdini had 7 straight draws, enough to hold off Stockfish with only 4.5/7. Stockfish is the only engine with no loss in stage 2 (as well as stage 1). It managed to beat Komodo once and Fire twice but failed to beat Chiron in four games. This is the first TCEC season that I've watched (since season 4) where Stockfish has not qualified to the superfinal. Better luck next season.


Some of the game highlights:

Komodo - Chiron: Komodo's eval gradually increased after the opening. Chiron tried to create a wall across the board, Komodo waited for the right moment to open the queen side and attack through the open files. Komodo used the fact that the black king was exposed to win two pawns, enough for a win. A good start for Komodo in the last RR. It needs all the wins it can get.

Andscacs - Stockfish: Andscacs had an eval advantage, the engines didn't seem to have a plan and Stockfish soon lowered its eval to 0. As the engines continued to shuffle mostly Stockfish started to imagine it had an advantage. The position remained closed for almost 60 moves, then after several exchanges Stockfish gave a rook for a bishop and pawns, and its eval was approaching 1. The pieces remaining were RRN vs RBB, and the engines exchanged down to a N vs B ending, black a pawn up. Despite having tablebases Stockfish could not find a winning line. The game continued for more than 100 moves before adjudication. The longest game of the stage so far. Stockfish now 0.5 points behind Komodo and in 3rd place.

Fire - Houdini: Fire had an eval advantage for a while, after a series of exchanges only RN vs Q were left. Both engines could not improve their position and the game was adjudicated a draw. At the end of round 22 Houdini is 1 point ahead of Komodo, Stockfish 0.5 points behind Komodo.

Fire - Komodo: Komodo let its queen get trapped in the center and traded it for BN and pawns. Despite the apparent material imbalance the evals were close to 0. Komodo's knights became very active on the king side making many fork threats. Fire was forced to give its queen for RN, leading to a RRN vs RBN drawn position. All round 23 games were draws, no changes in the table.

Komodo - Booot: There were a lot of fireworks after the opening. Komodo sacrificed a rook and opened the king side with a mating threat, Booot was barely able to defend. After many exchanges only QR vs QR were left with Komodo a pawn up and evals around 0.7. Komodo outplayed Booot in the endgame, eventually exchanging queens into a won rook endgame. A second win for Komodo in the RR, it is a point ahead of Stockfish with only 4 games each left.

Andscacs - Fire: Fire equalized quickly after the opening, and then evals became negative. Andscacs was two pawns up but its king side pieces were not developed and its king was exposed in the center. The game reached a BN vs RB position with Fire evals at around 1. Andscacs traded two advanced passers for a bishop, and then Fire's better endgame (using tablebases) was sufficient for a win. Andscacs' chances of being at least 6th are small.

Andscacs - Komodo: Andscacs thought it had a significant advantage for a long time, Komodo's eval stayed low. After Komodo created a passer in the center and put all its major pieces on a file on the queen side evals became negative. Afte exchanging queens with only RRN vs RRN remaining Komodo's pawn on the 2nd rank was much stronger than Andscacs' pawn advantage. It was only a matter of time for Komodo to convert a win. Third win for Komodo in the RR, it's getting closer to the superfinal. Two losses in a row for Andscacs, both in white pieces. 

Fire - Ginkgo: The opening was balanced with a small advantage for white. Fire played the opening better and its eval gradually increased. When the position opened up and many pieces were exchanged Fire won a bishop for two pawns and evals were over 2 in a RBB vs RN position. Ginkgo tried to advance pawns on the king side, Fire stopped the pawns and was quickly winning. 

Stockfish - Booot: Stockfish's eval climbed above 1 right after the opening. Before long it was a pawn up with an advanced passer on the 7th rank. Booot exchanged down to a RB vs RB position but the white passer was enough for a win. After round 25 Komodo and Houdini lead, Stockfish is 1 point behind. Booot in 7th is 0.5 point behind 6th place and 0.5 points ahead of Andscacs in 8th. Only 3 games per engine left.

Komodo - Stockfish: Stockfish equalized quickly and exchanged pieces until only QBB vs QBN remained. After exchanging queens the game reached an opposite color bishops ending and a draw. With two games left Stockfish needs a miracle to get to the superfinal. All the games in round 26 were draws, no change in the table and only 2 games left.

Fire - Booot: Fire had an eval advantage that gradually increased after the opening. After many exchanges only RRN vs RRN were left, Fire was a pawn up with connected passers on the queen side and evals over 1. Booot gave a knight to try to stop the white pawns, Fire had no problem winning the endgame. With this loss Andscacs and Booot will almost certainly be ranked 7-8. They are both automatic tiebreak losers because of crashes, so they need a crash in the Chiron - Ginkgo game to have a chance.

Komodo - Houdini: Stockfish was trailing by a point and needed a decisive result in this game to have some chance of qualifying to the superfinal. Komodo had an eval advantage from the opening. Houdini closed the position and avoided exchanges, keeping the evals low. There was only one exchange (B for N) in the first 40 moves of the game. Houdini had doubled pawns and Komodo had a better center, but nothing more. When the engines opened up the center the evals came down, and even turned negative for a while, but there was nothing in the position. The engines exchanged down to a RR vs RN position, and with both kings exposed to checks the game was soon adjudicated. As a result of this draw Stockfish is out of the superfinal, Komodo and Houdini are in.

Stockfish - Fire: The opening favored Stockfish and its eval reached 1 quickly. There was an eval jump to over 2 when Stockfish initiated a series of exchanges that led to a RBN vs RR position. Fire's rooks had little room to move with many pawns on the board. After a pair of rooks was exchanged Stockfish's eval was already over 5, it created a passer that would cost Fire its remaining rook and the game. The win comes too late for Stockfish, already with no chance of qualifying for the superfinal.

Chiron - Ginkgo: Chiron had an eval advantage after the opening. The position was static for a while with the engines shuffling. After all the knights were exchanged Chiron's bishops became very active on the king side, evals increasing over 1. Chiron opened up the king side, forcing a bishop for rook exchange. Then Chiron traded the material advantage to get a winning rook ending. This game closes the stage. 


Featured game: Houdini - Stockfish
Stage 2, round 27
Link to game on TCEC

Stockfish was trailing Houdini by a point with two games remaining. It needed at least a draw to keep some chance of qualifying. Houdini had a small eval advantage after the opening, Stockfish remained relaxed. Houdini castled long and advanced its king side pawns, Stockfish's king remained uncastled and it advanced pawns on the queen side. It wasn't clear where the black king could go, Stockfish chose to castle into the white pawn storm.




This looked very dangerous for black, and Houdini pushed the king side pawns closer to the black king. Stockfish remained calm, advancing on the queen side and careful not to move anything on the king side. It did not want to open files or create weaknesses around its king. Instead Stockfish began exchanging pieces. After queens were off a few files were opened and the evals increased a little. Eventually the position was simplified to a double rook ending, Houdini a pawn up.



Houdini still had an advantage, even one pawn can be significant with all the rooks on the board. However Stockfish was able to exchange one pair of rooks, still Houdini with a pawn advantage. For a short while Houdini's eval was higher than what it had been all game.



Houdini's excitement was short lived, its eval quickly dropped down to 0. Both engines raced their pawns forward, Stockfish was faster and Houdini had to trade pawns to avoid queening. The remaining 2 pawns against 1 rook ending was an easy draw. Stockfish managed to escape a difficult position, keeping its superfinal hopes alive for at least one more game.

Friday, November 17, 2017

Season 10, stage 2, RR 3

Table after RR 3:



An excellent RR for Houdini with 6/7 points, including wins against Fire and Komodo. It is firmly in the lead 1.5 points clear of Komodo and Stockfish. Stockfish also beat Komodo, and with a solid 4.5/7 is back to second place with good chances of getting to the superfinal. Komodo only got 3.5/7 in the RR, including two losses to its main rivals. These were its first losses of the season, and it will have to fight in the last RR to qualify. 

An interesting battle for 6th place is taking place, almost unnoticed. In the new format for season 11 the top 6 of stage 3 qualify directly to the top 8 stage (premier league?), while the bottom 2 will enter into the second league together with 6 other engines, only the two leaders of this league will continue to the top 8. Currently all four engines (Booot, Chiron, Ginkgo and Andscacs) are in danger, with Andscacs 0.5 points behind. Booot and Andscacs also have crashes that mean they automatically lose tiebreaks.


Some of the game highlights:

Stockfish - Andscacs: Stockfish had an eval advantage from the start. It then decided to sacrifice a knight in the center, leading to a long forced line that ended in a RB vs RB position 15 moves later. Stockfish was a pawn up with eval of around 1, but the bishops were of opposite color and the position was very drawish. After the rooks were exchanged it was obviously a draw, however the engines played on for more than 130 moves until they recognized that. 

Houdini - Fire: Houdini started with an eval advantage that grew to over 1 after a few moves. Houdini had much more space and a strong king side attack. More and more white pieces joined the king side and Fire had very little room to defend, the evals getting higher with every move. The game was adjudicated with tension at its highest, the PVs showed that Houdini was about to win material and that the black king would be in a mating net. Second loss in a row for Fire, Houdini 0.5 points ahead of Stockfish in second.

Komodo - Fire: Komodo outplayed Fire in the opening, getting an eval advantage of about 1. Fire countered through the open queen side, and after exchanging pieces the position was RB vs RB with opposite color bishops. The engines started shuffling, Komodo was a pawn up but it was a doubled pawn and the game was heading towards a 50-move draw. Komodo pushed pawns on the king side and Fire captured one, Komodo thought it was the wrong pawn and its eval started to jump. Now Komodo had a passer and its eval said it was clearly winning, but Fire was still holding. The 50-move counter was getting low and Komodo's eval came back down to 1 when Komodo found the right plan, bringing its king around to the queen side, sacrificing a pawn to reset the counter and attacking the f7 pawn. This time Fire could not defend and Komodo finally won the game. Incredible eval fluctuation, again Komodo manages to win in a drawn situation.

Ginkgo - Stockfish: The opening was balanced but Ginkgo played inaccurately, allowing Stockfish to take the initiative with evals increasing rapidly. The white king was vulnerable and under a black bishop threat. Ginkgo exchanged as many pieces as it could while the evals increased over 2. Stockfish's PV showed opposite color bishop endings, giving fans hope/fear of a draw. The game reached a BB vs BN position, and though Stockfish was two pawns up the evals remained constant for a while. Finally Stockfish found a winning line, though it was an opposite color bishops ending. Stockfish and Houdini are again with equal points.

Booot - Komodo: A fast repetition draw, only 24 moves.

Stockfish - Chiron: Another repetition draw, this time in 30 moves.

Houdini - Ginkgo: Houdini had an eval advantage after the opening. Ginkgo held on for a while, a pawn up, exchanging pieces and supporting a passer on the queen side. Houdini had a strong center. Ginkgo tried to open the king side and this allowed Houdini to block the center and put pressure on the black king. Houdini's eval increased rapidly, its major pieces very active on both sides. Within a few moves Houdini removed the black passer threat and was a pawn up in a winning QRN vs QRB position, with the black bishop trapped on the king side. Houdini ahead of Stockfish again.

Komodo - Andscacs: Komodo had a significant eval advantage after the opening, climbing to over 1. Andscacs' pieces were awkwardly placed, it had to accept a rook for knight exchange with evals over 2, and a RRB vs RBN position by move 26. Komodo then simply exchanged everything to get a winning king and pawns position. Easy win for Komodo, Andscacs has a chance in the reverse game.

Chiron - Houdini: The opening was balanced, after a while Houdini took the initiative and the negative evals started to increase slowly. Houdini opened a hole on the king side and put pressure with all its pieces. Chiron made a weak counter on the queen side, but Houdini's attack was strong with evals above 2. Houdini shifted to the center and was able to capture two pawns, exchanging down to a won double rook ending. At the end of round 18 Houdini is 1 point ahead of Stockfish.

Houdini - Booot: Houdini got a favorable opening and had an eval advantage. Booot tried to close the position and this worked for a while. Eventually Houdini found a way in, attacking first from the king side and then from the queen side. After a series of exchanges Houdini won a piece and reached a winning QB vs Q ending. Houdini is leading the table at the end of round 19.

Fire - Chiron: The opening was more or less balanced, a small eval advantage for Fire. Both engines developed pieces quietly with little contact between the pawns. When Fire started advancing pawns on the king side the evals began to increase. Chiron then chose a Q for RB imbalance, and with an exposed king had a hard time defending against Fire's queen and rook. Fire exchanged down to a QR vs RRB ending one pawn up and soon a second pawn, win for Fire.

Stockfish - Houdini: For a short while Stockfish thought it had a small advantage. Houdini equalized comfortably, both engines did not develop a serious threat. They exchanged pieces until only the kings were left in a drawn position. Four draws in round 20, no change in the table.

Houdini - Komodo: Houdini thought it had an advantage. For a while Komodo's eval remained low, but it started to increase when its king side became weak. After exchanging a pair of bishops Houdini controlled the king side. Evals climbed very fast as Komodo's king side crumbled, its king had to walk away. The game was adjudicated before Houdini started to win material. With this win Houdini extends its lead, will Komodo be able to win the reverse game?

Fire - Stockfish: Fire got a favorable opening, and evals gradually increased to 0.7 for Fire and 1 for Stockfish. Stockfish fans became very angry on the chat. They calmed down when the evals stayed at these levels and the engines began shuffling. Stockfish exchanged a pair of rooks and gave a pawn, leaving a QRB vs QRB position. All the pawns were on the king side and the extra white pawn was doubled. It took many moves and more exchanges but it was clear that Fire could not find a way to improve. The game ended in a drawn rook ending. At the end of the RR Stockfish and Komodo are level on points, 1.5 points behind Houdini.

Ginkgo - Chiron: Ginkgo had a significant eval advantage after the opening, which was eventually realized as a R for N advantage. However even with evals over 2 Ginkgo was unable to convert. It reached a R vs N endgame with only one pawn remaining, a 6-man tablebase draw. Ginkgo's lack of tablebase support was again its downfall.

Featured game: Stockfish - Komodo
Stage 2, round 19
Link to game on TCEC

This game was incredible for the fans, with a chess level that was way over my head - and I assume over most/all people's heads as well. After all these are chess giants using 43 cores to play.

Stockfish's eval started to increase after move 16, but as Komodo's eval remained low the feeling was that Komodo was confident in its defence. Stockfish's eval also came down a little, so probably false alarm.



The engines exchanged a few pieces and pawns, Komodo's king side became exposed and Stockfish had two connected passers. Komodo's central pawns looked strong, its eval stayed low. Who was right?




Both engines advanced their pawns, two pairs of connected passers marching in opposite directions. Queens were exchanged and it seemed the game was heading for simplifications. The game reached a RN vs RB position with Stockfish a pawn up. Evals were still at 1, and after two occasions of almost repetition draws, were there enough pieces on the board to convert?




Here Stockfish showed that it too has incredible endgame skills. The plan was to give the advanced h pawn for one of the queen side pawns, and then to use the fact that the bishop cannot help and that the black king is far away to get the second pawn. The maneuver was quite difficult to execute, it took almost 30 moves to complete, but when it was over the remaining white pawns secured a win for Stockfish. An amazing game which kept the fans wondering about the outcome for a long time. The first loss for Komodo, and without too much help from the opening. The reverse game will be very interesting to watch.

Tuesday, November 14, 2017

Season 10, stage 2, RR 2

Table after RR 2:


The current draw rate is 38/56 (68%), which is excellent considering the high level of the engines playing. In the first double RR with 28 reverse pairs there have been 14 1.5-0.5 results and two 2-0 results, and no biased 1-1.

Komodo leads the table with 6 wins, Stockfish and Houdini 1 point behind. Stockfish trailed Komodo and Houdini until the last round, where Houdini's first loss of the season and Stockfish's win over Fire allowed Stockfish to catch up. This shows how close the top 3 are in the table and how any decisive game they play can have a dramatic effect on the ranking.

The 4 engines in the bottom half are very close to each other. Andscacs is last only because it crashed in the first RR while winning against Ginkgo. Without the crash they would have swapped places in the table.


Some of the game highlights:

Komodo - Chiron: The evals were in Komodo's favor at around 0.5 for a while, then jumped to 1 when Komodo placed a protected knight in the center of the black territory. Evals continued to increase as Komodo captured two pawns while exchanging down to a BN vs BB position, an easy win.

Andscacs - Stockfish: In the reverse game Stockfish had a crushing attack on the black king right out of book. Here Stockfish as black defended well and was never in danger at all, its eval down to 0 very quickly. Stockfish played this opening better than Andscacs, but it needs wins to improve its position in the table.

Houdini - Andscacs: Andscacs chose a different line of defence compared to the reverse game. Houdini's eval stayed at around 0.7 for a while, but jumped above 2 when Andscacs took two pawns on the queen side while abandoning its king to a crushing white attack. It had to get its queen back in a hurry, and as a result Houdini won a bishop. The resulting RRN vs RR position was an easy win for Houdini, though we didn't get to see it played out.

Stockfish - Ginkgo: Stockfish had an eval advantage of about 0.7 after the opening. Ginkgo made an inaccurate move and the eval jumped above 2. Stockfish gave a pawn and destroyed the black pawn structure, then exchanged down to a rook endgame with a passer on the 7th rank. After this win Stockfish is still in third place, 0.5 points behind Komodo and 1 point behind Houdini.

Komodo - Booot: Komodo played the opening better and had a better pawn structure and an eval advantage of over 1. Booot captured a rook for a knight but Komodo took over the center with its knights and had strong queen side pawns. Booot gave the exchange back but it was already lost, the game reached a RB vs RB position with Komodo two pawns up and winning.

Ginkgo - Houdini: Ginkgo's eval was around 1 for over 30 moves, Houdini's reached 0.8 yet it still managed to get a draw in the game. After many exchanges only RB vs RN remained with Ginkgo a pawn up, not enough. The fact that Ginkgo doesn't have tablebase support is probably the explanation.

Houdini - Chiron: Houdini had an eval advantage from the start. For a while it held at around 0.7, as the position opened and after a few exchanges the evals started to increase. It was not clear how Houdini could penetrate the pawn wall. After exchanging down to a QBN vs QBN position Houdini opened a file on the king side, and used the file to get its queen behind the pawns. Chiron's remaining pawns were easy targets and Houdini's passers were enough to get a win. Houdini wins this opening in white and in black, first 2-0 of the stage.

Stockfish - Booot: Booot crashed on its first move, disappointing for the tournament.

Komodo - Stockfish: Komodo forced the black king to move early in the game so Stockfish couldn't castle. Nevertheless Stockfish equalized the game, rapidly exchanging pieces until only rooks remained. The game ended in a perpetual check. Still only draws in the top 4.

Ginkgo - Komodo: The evals became negative early in the game and Komodo took over the initiative. After opening the queen side and exchanging all rooks the position became static, Komodo trying to find a hole in Ginkgo's defence without success for a long time. When Ginkgo moved a pawn, the draw counter was reset and Komodo's queen infiltrated behind the pawn line. Then one inaccurate move was enough for Komodo to bring down the whole fortress. The game reached a QBN vs QBN position with the white king exposed to attacks, after most pawns were gone Komodo went a pawn up creating a critical advanced passer. Almost 30 moves after the inaccuracy Komodo reached a winning 6-man B vs N position. Ginkgo missed this line with no tablebase support.

Fire - Booot: Fire outplayed Booot in the opening, winning a pawn and exposing the king to attacks. Most of Booot's pieces were busy on the king side defending its king, Fire used the opportunity to capture a second pawn on the queen side. Then it was only a matter of exchanging down to a RN vs RN position and a straightforward win.

Houdini - Stockfish: see my live blog for this game. Houdini kept a small eval advantage for a while, Stockfish's eval was at 0 much sooner and it kept seeing repetitions in its PV. After exchanging all the bishops the engines shuffled with the major pieces and evals became in Stockfish's favor, though always close to 0. A series of exchanges that started with a queen sacrifice led to a N vs R ending and a draw. Houdini and Komodo lead Stockdish by 1 point before the final round of this RR.

Stockfish - Fire: Stockfish had an eval advantage from the opening that gradually increased. All the pieces were concentrated on the king side, Fire's king was more exposed to attacks. A sequence of captures simplified the position and left a RR vs Q position. Stockfish's eval was over 8, Fire took longer to realize it was lost. First loss to Fire in the stage, Stockfish catches up with Houdini in joint 2nd place.


Featured game: Komodo - Houdini
Stage 2, round 14
Link to game on TCEC

This game produced one of the greatest 'Wow' moments of the season. Unfortunately I missed the live game, though the chat replay is available on the new twitch chat.

The start of the game was nothing special. Evals were low with a small advantage for white. Komodo gave a pawn to open up the black king side, Houdini was defending well with most pieces on the board.


There was some tension on the board with threats made by both engines. It appeared to clear with the minor pieces exchanged off the board. On move 48 Houdini took the last white bishop, evals still very low.



Komodo took back the bishop with eval 1.44! Where did that come from? Houdini continued to play, taking a second pawn and eval still low, yet Komodo insisted and even increased its eval a little. It took Houdini four moves to realize it was in big trouble. The d passer became a serious threat, and combined with threats on the black king this was impossible to defend. For a minute Houdini thought it could get out using perpetual checks.



Komodo was saying 'sure win', and Houdini was saying 'easy draw'. Unbelievable. It was Komodo who had it right, a few moves later the game was over in a queen ending, no way to avoid the d pawn promotion.

Sunday, November 12, 2017

Season 10 live blogging, Sunday Nov 12th

The first half of stage 2 will end tomorrow, and there are now three high profile games in a row: Houdini - Stockfish, Komodo - Houdini and Stockfish - Fire. I will only see the first live, the others are too late at night.

Houdini - Stockfish

Stockfish is a point behind Houdini in 3rd place, it needs wins. However I think that in this game it will be happy with a draw, Houdini is a very dangerous opponent and it has the white color advantage. 

move 6: Following game 24, though as white Stockfish expected Bxf3 and as black it played Bh5. Stockfish thinks there's almost nothing for white. So far no PVs so we have no idea what to expect.

move 7: Houdini is taking a long time to play. OK, still following game 24. Stockfish's eval down to 0.05.

move 8: Houdini diverges with 0 - 0. Stockfish is looking for exchanges and opening the center, what will Houdini do?

move 9 : another long think for Houdini, it had no problem playing fast as black.

move 10: Houdini's eval is quite high over 0.4. First PV for white, looks very different compared to what black has in mind.

move 11: Stockfish has less developed pieces and needs to castle. Eval now at 0, I never believe Stockfish when it starts seeing repetition draws in the distant future.

move 13: Houdini exchanges bishops, following Stockfish's PV. Still a significant eval divergence, white's king looks exposed, black still hasn't castled.

move 15: Stockfish develops the bishop, it's attacking the center and looking at the white king. Houdini's eval is dropping.

move 17: black finally castles. The black queen controls a long diagonal but is hiding at b8. This looks like a position where both engines would like to not move. Shuffling ahead?

move 20: the remaining bishops are looking at each other, could be exchanged off any move now. Will we see negative evals in this game? Seems unlikely.

move 22: Houdini's Qb5 was unexpected. Houdini lowers eval to 0.05, how will Stockfish respond? Nothing exciting, still eval at 0.

move 24: there go the bishops. Houdini's eval negative?? only -0.06 but still a surprise. Also a long PV from Houdini. Stockfish stays constant on 0.

move 26: Stockfish thought hard sbout this move, but stayed on its PV without moving the eval. It is still seeing a repetition draw.

move 29: there have been no knight moves for a while now. They are like the glue that holds the position together. Houdini's eval drops to -0.19 for no apparent reason.

move 32: will Stockfish go for the repetition in its PV? It is thinking hard, trying to find a better way to continue. Didn't find anything, now will Houdini play along?

move 34: Houdini makes different moves but there is still a repetition offer.

move 36: Houdini will not repeat, now exchanges pawns on the queen side, and Stockfish's eval goes slightly negative for the first time.

move 39: long PV agreement, seems that there is going to be a simplification soon.

move 52: The end of all the exchanges is a drawn N vs R ending, we'll either see a draw rule or a tablebase draw very soon.

move 60: tablebase draw. The maximum Stockfish got was a few moves with negative eval. Both engines were safe throughout the game, Stockfish will have to get wins elsewhere. The next game against Fire is a good place to start.


Friday, November 10, 2017

Season 10, stage 2, RR 1

Stage 2 has started almost immediately after stage 1 finished, with only a short break of a few hours. The games are now a little longer with 90 minutes for each engine. Increment is still 10 seconds, so games will not be much longer than 3 hours.

Table after RR 1:



The current draw rate is 20/28 (71%), quite high but to be expected with the conditions of TCEC stage 2 - high end hardware, long time controls, high quality engines. We will see how well Cato chose the openings after the reverse games of RR 2, the optimal result of a pair of games is one draw and one win.

Houdini is doing very well so far leading the table with 3 wins. Komodo is second with two wins, Stockfish is lagging behind with only one. Fire is the fourth unbeaten engine so far.

On the other side of the table the most surprising is Andscacs with 3 losses. Andscacs was unbeaten in stage 1 and there was hope it could do well in this stage as well. Andscacs was also involved in this season's most talked about game, the 'c7f7 event'. In the round 5 game against Ginkgo Andscacs got to a winning position but then threw the game away by making an illegal move c7-f7. This was the move it made earlier, c7 was empty at the time, and it was under a check with only one possible move to make. C7f7 became the most talked about subject on the chat.

The new chat should also be mentioned. Chatwing was replaced by Twitch, which takes a llittle getting used to.


Some of the game highlights:

Booot - Ginkgo: Booot had an eval advantage after the opening, with a bishop pair in an open position. Ginkgo offered a repetition draw that Booot avoided, but then Ginkgo took the initiative. It opened up the king side and forced exchanging pieces, until only RB vs RN were left and evals were 0.


Komodo - Fire: Komodo had an eval advantage after the opening, quite high at 0.8. Fire gave two pawns to open files, and this initiated a series of exchanges in the center. After the dust cleared only QBN vs QBB remained and evals were going down. After queens were exchange evals were at 0 until a draw was agreed.

Ginkgo - Stockfish: Stockfish found a repetition draw that included a queen sacrifice, in only 21 moves.

Houdini - Ginkgo: Houdini started in front with eval up to 1 at one point. As the game progressed more and more pieces were exchanged and evals slowly decreased, until at move 35 only BN vs BN remained with evals close to 0. It looked like Ginkgo secured a draw when Houdini's eval jumped followed by Ginkgo's. Apparently Ginkgo missed a maneuver that allowed Houdini's king to penetrate the queen side and eventually capture two pawns. This was enough for a win.

Fire - Andscacs: By move 22 only RRB vs RRB were left, Fire a pawn up and with a small eval advantage. The position became static with the white king side pawns free to move. The engines shuffled for a long time, the odd pawn move resetting the counter and time getting short. At move 80 there were a few pawn exchanges that opened the king side, and shuffling resumed. Just before the 50 move mark Fire forced a pawn capture which led to a B for R exchange and a RR vs RB endgame. Fire outplayed Andscacs in the endgame and reached a winning R vs B position. There was no clear blunder by Andscacs, was it time trouble or lack of tablebases?

Komodo - Andscacs: The center was blocked and most pieces were concentrated on the queen side. Both engines had an advanced passer on the queen side, Komodo's was better supported. Andscacs had a weak counter on the king side, then gave a rook for a knight in an attempt to block the queen side. With a few exchanges the queen side opened up, and Komodo's pieces started to dance. It is hard to describe the moves, combining threats to the king and to hanging pieces, pawn advances and making sure the king is safe. Komodo finally managed to safely exchange queens and then used its advanced pawn to reduce to a won bishop against pawns ending.

Chiron - Houdini: All the pieces remained on the board for a long time. The center and queen side were blocked by pawns and only the king side , where both kings were, had undeveloped pawns. Houdini moved its pieces to the king side, Chiron's pawns prevented its pieces from getting there and Houdini's evals increased rapidly. When Houdini opened a crack on the king side the game was over within a few moves. First win for black in this stage, second win for Houdini now in the lead.

Booot - Stockfish: Stockfish got a difficult opening but managed to draw by exchanging the minor pieces and blocking with pawns. Will it win in the reverse?

Stockfish - Komodo: Many exchanges after the opening and evals close to 0. The engines started to shuffle at move 19, it took almost 50 moves for the draw rule to stop the game.

Houdini - Booot: The position stabilized on move 26 and the engines started shuffling. It took Houdini more than 35 moves to start attacking on the king side, but then the evals started to increase quickly. Booot was able to defend, exchanging pieces when it could. In the process Houdini captured a pawn and the game reached a queen ending. Booot's king was vulnerable to check, and Houdini used this to capture a second pawn before queen exchange - at least in the PV since the game was adjudicated. Third win for Houdini, great first round so far.

Andscacs - Ginkgo: For a long time the evals stayed close to 0, Andscacs with a small advantage. Both engines had an advanced passer, Ginkgo supported its passer with its queen that was almost trapped in the white territory. When Andscacs started to attack the black queen its eval started to increase. It gave a pawn but got rid of the black passer, and after exchanging queens the rooks became strong. In a RRB vs RRN position Andscacs placed its rooks on the 7th and 8th ranks, its passer was also a big threat and evals jumped above 3. The win was close when Andscacs shocked everyone - perhaps itself too - by playing an illegal move. What ??? The move it outputted was c7-f7, this was the move it made before, now c7 was empty and in any case it was in check with only one possible move for the king Kh3-h4. Such a weird and tragic end to the game. C7f7 became a meme in the chat, probably for the whole season.

Komodo - Ginkgo: Komodo had an eval advantage of about 0.5 for a long time. The engines moved pieces around without a clear plan until Komodo started attacking on the king side. Ginkgo was tempted to take a pawn while Komodo took over the king side, evals jumping above 2. Ginkgo exchanged queens and gave a rook for a bishop, the remaining R vs B ending was won for Komodo. Komodo in second place with its second win.

Booot - Fire: Another miniature draw, only 18 moves.

Stockfish - Houdini: The queens were off the board early in the game. For a while Stockfish's eval increased over 0.5, Houdini thought it was nothing. After exchanging pieces the eval came back down in a RBN vs RNN position. It was just a matter of time to reach a draw.

Houdini - Komodo: see my live blog of this game. The game reached a RRB vs RRB position on move 17 and the engines started to shuffle. Later the engines reached an opposite color bishop ending. Really boring draw, good for both engines leading the table.

Ginkgo - Chiron: Ginkgo held on to a pawn lead and an eval advantage of about 1, exchanging down to a rook ending. Chiron had tablebase support and knew it was a draw, Ginkgo with no tablebases thought it was ahead all the way until reaching a tablebase draw.

Featured game: Stockfish - Andscacs
Stage 2, round 1
Link to game on TCEC

Stockfish started with an eval advantage in a closed French opening, and things got worse for Andscacs very fast. Stockfish developed a strong king side attack, Andscacs could defend with its pawns and very few pieces.


Andscacs tried to counter by opening the queen side and exchanging, the immediate threat was gone but the evals were approaching 4. Although Stockfish let one of its rooks get trapped in a corner, it was still able to take apart the black position.


Stockfish won a rook for a knight and freed its rook, Andscacs could no longer handle the king side threat and defend the queen side pawns.


The game was adjudicated with Andscacs a rook down. This was the first decisive game of the stage and the only Stockfish win in the first RR.

Thursday, November 9, 2017

Season 10 live blogging, Thursday Nov 9th

Houdini - Komodo

This is round 7 of stage 2, Houdini leads the table with 3 wins, Komodo second with 2 wins.

move 8: Houdini is taking its time with the first move out of book. It is still tired after the last game against Stockfish. Over 5 minutes and we have a move, eval very low. Komodo answers almost immediately and with small negative eval, interesting.

move 9: Houdini immediately exchanges knights, and Komodo's queen is moving. Komodo's PV shows an early endgame is possible.

move 12: still following the previous Komodo PV, evals close to 0.

move 17: still the same PV, only RRB vs RRB remaining. Hard to see any outcome but a quick draw at this early stage.

move 24: Houdini still thinks it has a small edge. With so many pawns on the board it is hard for the pieces to move around, I think we'll see a lot of shuffling now. I hope Houdini lowers its eval so we can go home early.

move 31: shuffling in progress, Houdini could try to attack on the king side but Komodo is prepared.

move 35: Houdini's eval jumped a little, a change after more than 10 moves. Still not a lot seems to be differrent on the board. Now back to the previous 0.17.

move 37: now some exchanges, a pair of rooks and pawns on the king side, probably no change in eval. Perhaps Houdini will release us?

move 40: Houdini threatened mate, Komodo saw it though. Now the rooks are gone and we're left with opposite color bishops. Houdini still thinks it is ahead, long game ahead.

move 66: still nothing happening. Houdini is on increments and 30 moves left on the 50 move counter, though more exchanges are possible.

move 72: could it be?? Houdini lowered it eval and a draw adjudication is near.

move 80: finally over, really boring game. Houdini and Komodo are first and second at the end of the first RR.

Tuesday, November 7, 2017

Season 10, stage 1, rounds 22-23

Table after 23 rounds, final results:


The qualifiers are Komodo, Stockfish, Houdini, Andscacs, Booot, Chiron, Fire and Ginkgo.

The last two rounds were full of drama. Gull's two draws (after 3 in the previous rounds) gave it enough for a tiebreak, but then it had to watch as one by one the other contenders qualify with more points or better tiebreak scores. The race for qualification did not finish until the game before last, Chiron's win meant that it was in and Gull was out.

Komodo finished first with a margin of 1 point over Stockfish. Two draws in the final 3 rounds were not a good results for Stockfish. The crosstable for the top 4 engines is draws only, it is very hard to predict how these engines will be ranked in the next stage but I'm sure the competition will be harder.

Stage 2 started almost immediately after stage 1 was over. Good luck to all the qualifiers. To those which didn't I wish less bugs and better algorithms, hope to see improved versions next time.

Some of the game highlights:

Nirvana - Stockfish: The king side was empty while the action was on the queen side. The engines quickly exchanged pieces through the open files, reaching a drawn RN vs RB position. Stockfish with a second draw in a row, Komodo can extend its lead at the top. 

Chiron - Fruit: Chiron had an eval advantage from the opening, which increased rapidly with inaccurate play from the weaker Fruit. After many exchanges Chiron had an advanced passer in a RBN vs RBN position with eval over 2. Before long it was two pawns up, an easy win for Chiron. Chiron needs another win to have good chances of qualification, its SB score is high.

Hannibal - Booot: While most pieces were on the board Booot thought it had some advantage with evals at around 0.4. When the minor pieces were off and the engines exchanged queens the eval dropped to 0 in a double rook ending. Hannibal forced a perpetual check to get a draw. Booot has 14 points and one game to go against Ginkgo, may be heading towards a tiebreak.

Ginkgo - Houdini: Many pieces were exchanged early, Houdini had a bishop pair with a lot of space while Ginkgo was a pawn up with queen side majority. Neither engine could improve its position, and after exchanging a pair of rooks they settled for a repetition draw. Ginkgo is at 14.5 points with one last game against Booot. This means that one of these will have a final score of 14.5 (or less if Booot loses), so Chiron can force a tiebreak if it wins.

Gull - Bobcat: See the live blog for this game. Gull had a nice combination that started with a knight sacrifice, causing Bobcat's king to walk forward, eventually leading to being a pawn up. Evals were over 1 for a while, Gull first with a central passer and then with two connected passers. Bobcat was able to defend, blocking the passers and creating a wall that Gull could not penetrate. Evals came down as pieces were exchanged leading to a drawn RB vs RB position. It took a while for the game to end in a tablebase draw. Gull is now at risk of not qualifying, probably won't get more than a draw in the last game against Andscacs.

Andscacs - Gull: By move 21 only RBB vs RBN were left, evals close to 0. The game continued for a while but the drawn outcome was clear. With this draw Gull completes a run of 5 draws, it still has a chance to qualify but may need a tiebreak. Another consequence of this draw is that Ginkgo has qualified, since its final SB will be higher than Gull's in any case.

Texel - Fire: Texel won a rook for a knight, evals stayed close to 0. Both engines had advanced passers on opposite sides of the board in a RBB vs BBN position. Fire's pawns on the queen side were strong and Texel had to give back the exchange to try to stop them. The game reached an opposite color bishop ending, Fire was a pawn up. The evals were approaching 1 and Fire took another pawn, but still this type of ending is very drawish. Suddenly Fire's eval jumped above 6.5, Texel blundered in time trouble and Fire found a win, using pawns on both flanks to force Texel to lose material. With this win Fire qualifies for stage 2, two more spots available.

Booot - Ginkgo: Booot had an eval advantage after the opening, with more space than Ginkgo in a closed position. After opening the queen side Booot trapped a rook and Ginkgo exchanged it for a knight, evals were above 1. Ginkgo tried to close the king side, Booot opened it up as well, and more pieces were exchanged. Booot grabbed a pawn allowing Ginkgo's queen to capture a knight with a pretty check fork. After exchanging queens only R vs NN were left and Booot's connected passers on the queen side forced a win. Ginkgo's knights could not stop them without king support. This was a surprising win for Booot which now also qualifies for stage 2, Ginkgo dropping in the ranks but its place in the top 8 is already certain.

Gaviota - Chiron: This was a must-win game for Chiron. The game was more or less balanced with many pieces on the board and an eval advantage for Chiron. Gaviota was doing alright when most pieces were exchanged and only QRB vs QRN remained. With all the pieces on the king side Chiron's king found a hiding place behind a white pawn while Gaviota's king was under attack. Chiron's rook got to the 2nd rank and Gaviota could not defend its queen side pawns. After queens were exchanged Chiron was 3 pawns up, winning the game. Chiron is the last qualifier, Gull is out.


Featured game: Fire - Komodo
Stage 1, round 22
Link to game on TCEC

Fire with 14 points was still in the qualifying race, it would have been satisfied with a draw in this game. The position became static very early and the engines were shuffling from move 13, including a few pawn moves. The evals were slowly increasing for Komodo but there didn't seem to be a plan for improvment. This changed suddenly after Fire's move 30 Kh1.



Komodo saw an opportunity to attack the king with almost all the white pieces on the wrong side of the board, and with its d4 knight slowing down the defense. Fire became desperate very quickly, evals shooting up.



Fire tried exchanging pieces and giving its bishop, but it was not enough to stop the attack. The game was adjudicated a few moves later, PV showing that Komodo was about to win another piece.



Komodo shows it can attack with fury, and not just find small cracks in endgames. It will finish first in stage 1. With this result Fire was still in danger of not qualifying, it had to wait for the last round.

Season 10, stage 1 statistics

Draw rate, wins

Final draw rate was 40.2%, with a bias towards wins for white.


Game termination

The three most common game termination causes were:
57.6% - TCEC win rule
20.6% - TCEC draw rule
9.8% - TB position
9.8% - 3-fold repetition

There were 3 crashes (and no losses on time), Fizbo crashed twice and Arasan once. All in all stability was excellent.

Moves per game

median = 56.5
average = 63

There is a long right tail of long (and extremely boring) games. About 10% of the games were over 100 moves long, 6 games over 150 moves, the longest had 236.5 (Komodo-Chiron).

Time per game 

median = 2:01
average =2:01


The short added time per move, only 10 seconds, forced the games to end quickly even if many moves were played.

Openings


Using ECO codes there were 144 different openings played. The top ECO codes were:
B20 9 times - Sicilian defence
A80 8 times - Dutch
A01 7 times - Nimzovich - Larsen attack
B00 7 times - King's pawn opening
 
Using full opening name there were 235 different openings in 276 games. Reti: KIA was repeated 4 times, 8 more openings were repeated 3 times. 

Using only the first letter of the ECO codes we get the following distribution:
The openings were biased towards flank openings (e.g. Dutch, Reti and English), a lot less open games and almost no Indian defences.


If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
English - 33 times
Sicillian - 33 times
Dutch - 17 times
Reti - 17 times