Thursday, June 30, 2016

Season 9, stage 2, rounds 10-12

Table after 12 rounds:

The draw rate is 56/96 (58%), still dropping. There were many decisive games in the last three rounds, even most of the draws are interesting. 
We have seen two Chiron crashes in rounds 10-12, one in an obvious winning position. Fire was on fire with 3 wins, it is now in second place. Houdini suffered two losses and is now level with Stockfish.
The table suggests that the engines below Nirvana are less likely to qualify in the top 8 places. On the other hand there are still many games to play and anything can happen.

Some of the game highlights

Raptor had an advantage over Stockfish, and to get out of trouble Stockfish gave a rook for a bishop. This got Stockfish's eval to zero but Raptor was confident it was better, its eval stayed high as the pieces were exchanged. In the R vs B ending Raptor's eval jumped above 4, with no tablebase support it wasn't aware that the 6-man position was a draw. We had to wait for the 50-move rule.

Houdini got into trouble against Fire, a bit unexpected. It had a trapped bishop and an exposed king, and the pressure and evals were rising. Fire just directed its pieces at the white king, and at the right moment attacked. It came out a rook ahead, winning the game. It looked too easy, especially against Houdini. The rematch will be interesting.

Ginkgo was a pawn up early against Vajolet2, as well as a strong bishop pair. Ginkgo exchanged down to RBB vs RBN and won another pawn. Winning became straightforward, it took some time but Ginkgo made it look easy.

Komodo - Jonny was very interesting since Jonny had good results so far but playing Komodo can be frightening. Komodo had an eval advantage in a closed position, and the evals were over 1 when the position opened through a series of exchanges. The game reached a R vs B ending with two extra pawns for black, I expected Komodo to win with such an eval. However, Jonny was able to hold the position for a long time, and the game reached a drawn 7-man position. Jonny is now officially an engine to look out for this season.

Protector held off Rybka's attack and equalized the game. Unfortunately for the viewers the RB vs R ending was misjudged by both engines as a big advantage for white, and we had to wait for the 50-move rule to prove them wrong.

In Chiron - Gull Chiron had the advantage but Gull defended well. They reached an opposite bishop ending with white a pawn up, a sure draw which Chiron misevaluated, and then for no reason Chiron crashed. A similar thing happened in stage 1, Gull gets a bonus 0.5 point.

Hannibal - Andscacs started with a series of exchanges that left Hannibal down a bishop for a rook but ahead on eval. Andscacs' queen side was undeveloped and its queen was harassed by white's pieces. Andscacs gave the material back in an attempt to relieve the pressure, but Hannibal was in a won position, one of the shortest wins in the stage.

Andscacs bounced right back in the game against Raptor. Andscacs had an eval advantage and was a pawn up, but the crucial moment came when it gave a rook for a knight and a strangle hold of the black king. After a series of exchanges the position was a BN vs N endgame, and the game was stopped in a zugzwang leading to the collapse of the black pawn line.

Rybka was a pawn up against Chiron, with connected passers on the queen side. Chiron tried to block the pawns and keep the position closed, but Rybka was able to exchange pieces and reach a winning endgame. That was Rybka's first win of the stage.

Fire was a pawn up against Ginkgo, with connected passers on the queen side. After many exchanges Ginkgo chose to give a rook for two knights, leading to a RR vs RNN ending. The pawns started to march and Ginkgo coulnd't stop them, another win for Fire (second in a row).

Raptor - Houdini got into a RR vs RB ending and seemed to be heading for a draw. The only advantage Houdini had was a pair of connected passers in the center. Raptor saw an opportunity to take the black bishop, but in return Houdini got another passer and moved its king and rook forward. The game reached a 6-man tablebase win for black, Raptor not realizing this with no tablebase support.

Playing black against Naum, Fire attacked the white king with pawns on the queen side. As a result Fire took a rook for a bishop and also boxed the king in the corner. This became a serious back rank mating threat which Fire used to win material and the game. Three wins in a row for Fire, wow.

Komodo was happy when Vajolet2 offered RB for its queen. Vajolet2 thought it was equalizing but in fact this led to a won pawn endgame for Komodo. Don't mess with the lizzard !!

Protector held an eval advantage over Nirvana from the opening but somehow (Protector inaccuracies?) Nirvana found a way to turn the game around to its favor. It attacked the white king with all available pieces until Protector couldn't move at all. Adding one pawn to the attack was all it took to break the white defence, reaching a won RR vs RB endgame.

Chiron outplayed Jonny, with some help from the opening. The game reached an opposite colour bishop ending, Chiron with three extra pawns. This was enough for a win, but then Chiron crashed again. Such a disappointment for the viewers, Jonny lucked out of a first loss in the stage.

Andscacs had a better pawn structure than Gull, and after several piece exchanges this became an eval advantage and an extra pawn in a QR vs QR position. After queens were also exchanged Andscacs had a three pawn advantage in a rook ending, easy win.

Featured game: Stockfish - Houdini
Stage 2, round 11
Link to game on TCEC

This was the first game of the stage featuring two top-3 ranked engines (Komodo hasn't played either of these engines yet). Houdini had just lost its first game of the stage, while Stockfish was still unbeaten but was not performing as well as was expected.

After 30 moves the position was not clear. Material was equal, Houdini seemed to be defending and it had a worse pawn structure. Could Stockfish find a way to improve its position?

Ten moves later Stockfish had a solid eval advantage, a pawn up with two potential passers. There were still many pieces on the board and Houdini is a tough defender, the white king was also exposed to attacks. Tension was running high. The game continued with exchanges that simplified the position until only RB vs RB were left.

The evals were up but the opposite color bishops raised doubts about whether Stockfish could really pull this off. It was far from easy, but the evals gradually grew until the game was adjudicated.

Houdini could hardly move and its bishop could not see the white pieces. The win was not obvious but both engines saw a queening in their PV. Well done Stockfish.

Sunday, June 26, 2016

Season 9, stage 2, rounds 7-9

Table after 9 rounds:

The draw rate is 46/72 (64%), continuing to drop. We have had 11 decisive games in the last three rounds, almost half. Remember starting with 12 draws in a row?
Komodo is still dominating at +5. Houdini is a surprise for me in second place, it wasn't performing so well in the last two seasons. Perhaps the return of its developer (R. Houdart) has given it a new strength, perhaps it likes the new TCEC hardware. Stockfish is continuing to disappoint, only at +2 after 9 rounds. Other than these three, only Jonny and Fire are still unbeaten in the stage.

Some of the game highlights

Chiron - Raptor got into a more or less static position, with many pawns and pieces on the board. After a few exchanges the position opened up, with only QRR vs QRR left. Chiron was up by two pawns but the evals favored Raptor. Suddenly Raptor saw something and its eval jumped to near 6. After a series of checks and forced moves Raptor was able to force an exchange of a rook for the white queen, winning the game. Surprising how Chiron didn't see this coming, well done Raptor.

Protector had very little room to move playing black, and Hannibal took advantage of the situation to get a favorable QN vs QB ending. After queens were exchanged Hannibal was a pawn up with an advanced passer winning the game.

Gull - Naum got into a blocked center position, with Gull a pawn ahead. Gull's pieces had more room to move and its knight was better than a bishop in tight spaces. The knight was exchanged for a bishop and another pawn, and when a third pawn was captured the game was won for Gull. There were still many pieces on the board but Naum couldn't avoid material loss.

Jonny thought it had a big advantage against Houdini. After a series of of exchanges Jonny had a BN for R imbalance with an advanced passer, and its eval was climbing while Houdini remained indifferent. After exchanging pieces to get rid of the passer the only piece left was Jonny's knight, but it couldn't control all of Houdini's pawns and defend its own, leading to a draw.

Houdini had a better position against Nirvana, but Nirvana managed to hold the line for a long while, hiding its king behind a white pawn. At some point Houdini won a rook for a knight and the evals jumped, but still Nirvana fought to stay in the game. After about 30 moves Houdini found a way to break the line, exchange queens and win. Did Nirvana blunder in the end?

Ginkgo - Jonny was over after 18 moves with repetition. They really didn't want to play...

In Komodo - Gull it looked like Gull was doing fine for 30 moves. Then Komodo started to gain eval gradually, advancing pawns on the king side where it had a 4-3 advantage. When two pawns reached the 6th rank Gull had very little room to move. Together with a rook and support from behind Komodo started winning material and the game was over.

Hannibal - Raptor reached a QRB vs QRR position where white had three pawns on the queen side. Hannibal steadily moved them forward, supporting with the king and its pieces. Raptor could not stop them all, the game was over quickly.

Andscacs was up a rook to bishop exchange early in the game against Chiron. When the position opened up a little Andscacs had a strong attack on the black king and the advantage became apparent. After queen exchange a white rook infiltrated the black defense attacking pawns from behind, and the game was over.

After a series of exchanges in the opening of Gull - Protector white was a bishop up for three pawns. Before long the only pieces left were BB vs B, and the bishop pair was stronger than the extra pawns. An easy win for Gull.

Jonny won a pawn right from the start against Naum. Many exchanges later Naum was able to get the pawn back, but in the rook ending it had a weak pawn structure which turned into a 6-man winning position for white.

Vajolet2 was ahead on eval for a long time against Houdini, even though Houdini was a pawn up from the opening. Vajolet2 didn't make anything out of its advantage, and after 30 moves Houdini started to dominate. It was not clear how Houdini could improve its position, but its eval suddenly jumped (Vajolet2 blunder?). The black rook and queen found a way behind the white pawns, and after the queens were exchanged white's position collapsed.

Featured game:Stockfish - Vajolet2
Stage 2, round 8
Link to game on TCEC

After 27 moves Stockfish was ahead on eval, a pawn up, and Vajolet2's king was a little exposed. With the way Stockfish has performed so far it was not clear whether this was enough for a win.

Then Stockfish sacrificed a knight, we haven't seen this for a while. The black king became even more exposed but the evals didn't jump immediately. It did a few moves later when Stockfish concentrated forces on the king side, and Vajolet2 was a bit careless defending.

A few moves later there were just too many threats on the king for Vajolet2 to defend. It was still up a piece but had too many of its pieces on the wrong side of the board and unable to help.

The game was over soon, Vajolet2 had to give material to get out of the immediate dangers, the PVs show that it was about to be down a full rook. Stockfish still 'has it' if you get it in the right mood, just hasn't happened a lot this stage.

Wednesday, June 22, 2016

Season 9, stage 2, rounds 4-6

Table after 6 rounds:

The draw rate is 33/48 (69%), much better than what we had in the first two rounds.
Komodo is dominating the start of this stage, other than that the differences between the engines are quite small. If this trend continues the qualification race may be very close, still a very long way to go.

Some of the game highlights

Vajolet2 - Jonny reached a drawn 6-man rook ending, with two white pawns on the g and h files. Vajolet2 did not have tablebase support so it thought it was winning for about 15 moves before it realised that there was no way to win.

Fire had a rook for a knight in the game against Rybka, and had an eval advantage of about 1. In the end it had to give the exchange back and the RB vs RN endgame was drawn.

Stockfish was a pawn up against Gull, and ahead on eval. However, after most pieces were exchanged one pawn was not enough in the rook endgame, with 4-3 pawns on the king side. It took quite a while for both engines to lower their evals for adjudication.

Houdini started the game against Andscacs with an eval advantage. It was able to convert threats on the black king and control over the dark squares into a winning BN vs BN endgame with an advanced passer. Andscacs had to give a piece to avoid queening, and Houdini won comfortably.

Ginkgo - Hannibal reached a blocked BB vs BN position, where black had a passed pawn but there was no way to break the pawn line. Hannibal's eval kept climbing to over 5, with nothing to back it up, going back down reluctantly for the 50-move draw. Someone should look at that eval function.

Komodo exchanged RPP for BN against Protector. The position became blocked with a black pawn line (Protector had all pawns on the board), Komodo had a space and eval advantage. When Komodo challenged the pawns on the queen side Protector chose to lock pawns completely. This allowed Komodo to focus only on the king side, and it found a way to force black to move a pawn and weaken the line. This was enough for Komodo to convert the space advantage to material and a winning position, the game ended just before the wall broke down.

Protector was a pawn up early in the game against Raptor, with two passers on the queen side. After exchanging most pieces the position was a rook endgame with a passer for Protector, it was a bit tricky but Protector found the way to win.

Andscacs sacrificed a knight in the opening against Ginkgo, to get a strong attack against the king and to win 3 pawns. Later in the game with less pieces on the board the advancing pawns were much stronger than the knight leading to an easy win. Will Ginkgo be able to return the favor in the rematch?

Nirvana had an eval advantage in the opening against Vajolet2. After a while the engines started exchanging pieces, reaching a queen ending with white a pawn up. The advancing passer with queen support was enough for a win.

Fire was up a pawn out of the opening against Nirvana, offset by 3 pawns in one file. In the middlegame Fire outplayed Nirvana and won a rook for a bishop. In the RB vs BN endgame Nirvana had connected advanced passers but with no support from the king, Fire had a dangerous passer as well. Both engines gave pieces to get rid of the passers but this left a winning R vs B position for Fire.

Stockfish held an eval advantage over Jonny for a while, but after a rook exchange the evals came down. Stockfish couldn't find a win in the RBN vs RBN position, Jonny gave the knight to get rid of the white pawns and the game ended in a repetition draw.

Ginkgo kept a small eval advantage against Gull, material being equal. Evals started going up near the endgame where Ginkgo had a passer in a B vs N position. Both engines played without tablebases, and when they reached a winning 7-man position it was not clear Ginkgo could find the win. There were no serious blunders though, and the win became clear for both engines after a while.

Andscacs blockaded effectively playing black against Naum. After more than 70 moves Naum started to open up the position (out of frustration?), and even got a passer. However, it turned out that the central pawns Andscacs had in the center were more effective with less pieces on the board. The RN vs RB ending with connected passers was a win for black.

Featured game: Komodo - Hannibal
Stage 2, round 6
Link to game on TCEC

The opening looked quiet, Komodo with a small advantage. The kings castled in opposite directions and Komodo immediately started a pawn attack on the king side. Hannibal was happy to get a two pawn advantage but this opened up files and the queen was the only defender left.

Suddenly Komodo's eval jumped to almost 4, BOOM. Hannibal didn't see it because of the rook sacrifice that was coming.

After the rook sacrifice Hannibal had a long think and realised it had a problem. The black king was in a mating net, with no pawns and the rooks too far to help. Hannibal had to lose the queen to avoid an immediate loss.

The position was lost anyway. Komodo forced a knight for rook exchange to reach a won Q vs R ending. What an attack for Komodo, beautiful.

Saturday, June 18, 2016

Season 9, stage 2, rounds 1-3

Now that the preliminary stage 1 is over, there are no more weak engines that hand out easy points. The competition in stage 2 is going to be much harder, who will finish in the top 8? This stage will have 8-move openings so there will be less opening blunders. This is a double RR, so each match will be played twice reversing the colors. This stage is then less about the openings and more about engine strength.

Table after 3 rounds:

The draw rate is18/24 (75%). The stage started with 12 draws in a row, and the fans were getting restless. There have been decisive games since then, but it looks like it's going to be a hard stage for all the engines.

Some of the game highlights

Komodo was a pawn up against Ginkgo, but Ginkgo's bishop pair was stronger than Komodo's pair of knights in the RNN vs RBB endgame. The evals jumped when Komodo won a rook for a knight, but with very few pawns left Ginkgo was able to hold the draw with its bishops.

In Chiron - Stockfish black went for an exchange of BN for RPP. For a while the evals were in Stockfish's favor, but Chiron was able to equalize. Stockfish gave material to get strong passers but Chiron was able to force a perpetual check for the draw.

Andscacs tried a rook for bishop exchange against Vajolet2, in order to break the pawn blockade. In the RB vs RR ending the pawns were taken until black had none. The game reached a drawn 6-man KBPP vs KR, but since both engines did not use tablebases the game continued for 60 more moves before adjudication.

Raptor used a nice combination in the game against Jonny, and got a QR vs QR ending with a pawn advantage. The evals were over 1 for a while, but Jonny had a pawn on the 2nd rank and an active queen, and that was enough to force a draw.

Nirvana gave a pawn against Rybka but couldn't get enough compensation for it. After most pieces were exchanged the position was a BN vs R endgame, Rybka a pawn up. The evals indicated Rybka had an advantage, with a passer guarded by the king and rook. But Nirvana managed to create a threat with its own passer, and when the two passers were exchanged the position became drawn. Did Rybka blunder a victory?

Vajolet2 and Gull reached a 3-fold repetition after 31 moves. This was the eleventh straight draw, frustrating.

Stockfish - Hannibal, first decisive game of the stage,  see featured game.

In Houdini - Chiron Houdini developed a strong attack on the black king that cost Chiron a knight for two pawns. From here it was a gradual increase in evals corresponding to pieces taken off the board and Houdini retaking pawns. Houdini was given the win in a RRN vs RR position with a pawn up. Two decisive games in a row, now the stage is officially open.

Ginkgo was a pawn up against Protector, in a position with a passer and with many pieces on the board. Protector defended well and refused to exchange pieces until very late in the game. When pieces started to come off the board the evals went down, the game ended in a tablebase draw.

Naum had an eval advantage out of the opening against Komodo, but Komodo was able to equalize and then to take the initiative. When the pieces were down to RBN vs RBN black's king side pawns were much better than white's queen side pawns. The game was adjudicated just before Naum started to lose material. Komodo with a first win, joining Stockfish and Houdini.

Komodo had an opening advantage against Raptor, but Raptor was able to equalize the position. Raptor was two pawns up (doubled passer) and Komodo had a bishop pair with evals close to zero. In the RBB vs RBN endgame Komodo was much better though and was soon clearly ahead. Raptor also failed to realize that the RP vs BP 6-man position reached was a win for Komodo due to the lack of tablebase support. Back to back wins for Komodo, best endgame expert in the field.

Rybka had a promising lead against Vajolet2, with two passed pawns on the queen side. Vajolet2 defended well, and was able to take one of the pawns. Rybka countered by exchanging the queen for two rooks, with evals getting closer to zero. With the second passer gone in a RRB vs QB position suddenly it was Rybka defending. The ending was a masterclass of how to use the queen's mobility in tight spaces. Vajolet2's queen almost singlehandedly won the game, taking out pawns while not allowing Rybka to move pieces. A surprising win for the new engine.

Jonny had an eval advantage out of the opening against Nirvana, and it was able to convert it into a winning RB vs RN endgame. It will be interesting to see whether Nirvana can win the rematch as easily.

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

Stockfish had an eval advantage after the opening, but after so many draws there were doubts that any engine could get a real win. Hannibal had a bishop pair against Stockfish's knights, but the knights were holding the center and much more active.

With a pawn attack Stockfish opened the king side, exchanging a knight for a bishop in the process. Now both kings were exposed but Stockfish had the initiative and was clearly ahead on eval.

After exchanging queens Stockfish reached a winning RN vs RB endgame, with the black pawns weak and vulnerable. Also notable is how all the white pieces were on dark squares, invisible to the black bishop.

Hannibal went for a final move combination that resulted in a 6-man rook ending with white a+b pawns. With no tablebase support it was unaware this was winning for white, and for 9 moves Hannibal's eval was a draw. Losing came as quite a shock...

Tuesday, June 14, 2016

Season 9, stage 1b stats

Draw rate, wins

The results are almost identical to the 1a group.

Game termination

The three most common game termination causes were:
66.7% - TCEC win rule
13.3% - TCEC draw rule
9.2% - TB position

There were 3 crashes (and no losses on time), slightly better than for stage 1a. All in all stability was excellent.

Moves per game

median = 55.25
average = 63.9

Time per game

median = 3:41
average = 3:41

Games of group 1b were about 20 minutes shorter than in group 1a.


Using ECO codes there were 79 different openings played. The top ECO codes were:
B22 6 times - Sicilian, Alapin's variation
A28 3 times - English, four knights
A43 3 times - Old Benoni defence
A45 3 times - Queen's pawn game
B30 3 times - Sicillian Defense
B50 3 times - Sicillian
C01 3 times - French, exchange variation
D02 3 times - Queen's pawn game
D30 3 times - QGD

Using full opening name there were 114 different openings, so almost no repeated opening. No opening was repeated more than twice.

Using only the first letter of the ECO codes we get the following distribution:

Compared to the 1a group, there were less Sicillians and closed games, and more open games and Indian defenses.

If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
Sicillian - 22 times
English - 16 times
QGD -  10 times

Season 9 stage 1b, rounds 13-15

Final standings,stage 1b:

Qualifiers are: Stockfish, Nirvana, Jonny, Fire, Naum, Chiron, Andscacs and Vajolet2. Like in stage 1a the final qualifiers were determined in the last round, Critter missing out on the very last match. Stockfish won the stage but it was a very close race. Crashes made an impact on qualification: without the extra points Naum and Andscacs would have had 9.5, same as Vajolet2 and Critter.

The final draw rate for stage 1b is 33/120 (27.5%). Almost identical to stage 1a.

Some of the game highlights: 

Jonny was very confident it had a big advantage playing black against Naum with evals getting high very early in the game. Jonny had a pawn advantage in the RN vs RB endgame with a 4 against 2 pawns situation on the king side. It took a while but this was enough for a win. Will Jonny win the stage? It still has to face Nirvana.

Fire and Vajolet2 had a long struggle on the queen side, where Fire had a 3-2 pawn advantage, later reduced to 2-1. Fire came out ahead with two passed pawns and a big eval advantage. Vajolet2 was forced to lose material, reaching a BN vs B ending, 3-4 pawns on the king side. The game was adjudicated before there was any material change, but with black's pawns being targets for white's bishop. A huge result for the qualification of both engines, but nothing is determined yet.

Gaviota held the pawn line against Andscacs for almost 60 moves, but the position deteriorated quickly after Andscacs exchanged a rook for a knight and created a hole in the line.

Stockfish had a strong king side attack that forced Critter to lose a knight, leading to a winning RN vs R endgame. Stockfish winning the stage is almost certain.

Jonny had an eval advantage over Critter for 40 moves, with a significant advantage in space. When the position opened up a little it was suddenly Critter who was up on eval, but with only RB vs RN left a draw was more probable. The game ended with mutual queenings and a perpetual check.

Andscacs was hoping to be the first engine to beat Stockfish this Season. It held an eval advantage right from the opening for over 30 moves. Despite opposite castling the game remained relatively calm with no serious threats on either side. Eventually the evals dropped back to zero and a draw was reached. Jonny may still tie for first place, and Andscacs is still not qualified for sure, last game against Critter may be the decider.

Vajolet2 beat the Baron without too much trouble. It is still 9th but is not out of the race yet.

Chiron had a significant eval advantage over Fire, but as the game progressed with less and less material on the board it turned out that Chiron was two pawns up in an opposite bishop ending. The position became static, black not moving and white unable to move a pawn without losing it. And then for no reason Chiron crashed!! So sad... The last round is going to be tense.

Naum - Nirvana was one of those long and boring draws, this time in a BB vs BN ending. Naum waited 70 moves to lower its eval long enough for adjudication.

Jellyfish - Naum seemed to be heading for a draw. Surprisingly, Naum was not able to get a significant advantage and the pieces were being taken off the board until on move 50 only N vs N was left, Naum with an extra pawn. But Jellyfish blundered in the ending allowing Naum to win. This secures Naum's qualification.

Fire beat Arasan with ease, securing its qualification as well.

Chiron also qualified after beating the Baron in a rook ending playing black.

Now with only two more qualification spots, it was down to Vajolet2, Critter and Andscacs. Gaviota playing against Vajolet2 exchanged a rook for a knight and pawns early in the game, but when the position opened Vajolet2's rooks were a big advantage. Gaviota was forced to lose more material and lost the game. Vajolet2 was now tied with Andscacs and 0.5 points ahead of Critter with better SB.

Stockfish demolished Firefly as expected, and won the stage.

Critter - Andscacs was the last game of the stage. Critter needed a win to qualify ahead of Vajolet2, but Andscacs had other plans. Both evals reached zero on move 25, and from there it was a long boring game waiting for the pawn moves to stop long enough for adjudication. Vajolet2 and Andscacs are the qualifiers, Critter 9th on SB.

Featured game: Nirvana - Jonny
Stage 1b, round 15
Link to game on TCEC

Jonny was co-leader with Nirvana 0.5 points behind before their last match of the stage. Stockfish was going to win for sure so Jonny needed a win to stay at the top. The game reached a QRR vs QRR position at move 21 with evals close to zero. The most likely outcome was a draw, Jonny finishing second. After 40 moves the position was:

White could alternate pressure on the a6 and e6 pawns, the black queen confined to the king side too far to help with the a6 defense. Nirvana had a small advantage, and perhaps Jonny's defense was not perfect, allowing Nirvana to win the a pawn.

The b pawn now became a passer. Nirvana managed to advance the pawn slowly, keeping an eye on the black queen to avoid checks. Eventually Jonny had to block with the rook to avoid queening, and Nirvana could set up a winning position.

In the end it was Nirvana who came in second, Jonny only third. Both had a great first stage, we'll see in the next stage how much skill and luck were involved.

Friday, June 10, 2016

Season 9, stage 1b, rounds 10-12

Table after 12 rounds:

The draw rate is  29/96 (30/2%) which is almost the same as stage 1a.
Nine engines fighting for the eight qualifying places for stage 2. Combining the points and the remaining games I would say that Fire, Vajolet2, Andscacs and Critter are are the most likely candidates for finishing in 9th place. The Fire-Vajolet2 game in round 13 and the Critter-Andscacs game in round 15 are therefore critical. Any decisive result here can have a large effect on the final ranking.
At the top Stockfish has weaker opponents than the others, I think it will finish first. However Jonny and Nirvana still have a shot, both haven't lost a game yet and are playing well. Relevant games are Naum-Jonny (round 13), Naum-Nirvana (round 14) and Nirvana-Jonny (round 15). Chiron will probably also move up in the ranks, but I don't think it has a shot at number 1.

Some of the game highlights:

Jonny - Gaviota reached a RBN vs RBB endgame by move 20. The position looked drawn but Jonny soon outplayed Gaviota, despite the bishop pair. One bishop became almost useless and the other exchanged for a knight, and from there Jonny won comfortably.

Critter beat Fire quite easily, a bit surprising for me. Critter was ahead on eval right from the start, so it may be a biased opening or Fire misplaying the opening. Fire had a bishop pair for Critter's knight pair, but with most pawns on the board the knights were much better. Fire is still fighting for qualification and it needs better results.

Chiron beat Naum without too much difficulty, again somewhat surprising (although Naum has an extra point due to the Fire crash). After applying pressure on the queen side Chiron was able to get two connected passers in the center with only RRB vs RRB remaining. From there the win was just a question of technique (way over my head but still straightforward).

Naum - Arasan was a long game where Naum held a significant eval advantage but was not able to penetrate Arasan's defense. The game almost ended by the 50 move rule twice, but somehow Naum found the path to victory (by random shuffling and luck?).

Delphil and Firefly drew the battle for last place, making sure that both get points in this stage. Both will probably not get any more points... And we still don't know which should be last.

The Baron managed to get a draw against Critter. Pieces were exchanged quickly and the pawn structure was symmetric enough that neither engine could develop any significant threats. For a short while the Baron had a pawn advantage and hopes that it can make something out of it, but there was not enough material left on the board.

Jonny - Stockfish was a battle for first place. The game was interesting, including a queening as well as a knight sacrifice towards the end. However both evals fell to zero after 26 moves so the draw was inevitable.

In Critter - Gaviota there was a long struggle in the center, the evals stayed relatively low for over 50 moves. Gaviota had an isolated advanced pawn which it tried to defend, but in the end Critter found a way to take the pawn, gain control of the center and win the game.

Arasan was a pawn down against Nirvana and the evals were in Nirvana's favor. However in the N vs B endgame Nirvana couldn't convert its advantage and the game ended in a draw.

Fruit tried a bishop for three pawns exchange against Naum, but this backfired in the end. Naum played the QR vs QRN ending better taking the white pawns one by one until the game was adjudicated.

Featured game: Gaviota - Stockfish
Stage 1b, round11
Link to game on TCEC

Gaviota was supposed to be one of the weaker opponents of Stockfish, so I expected a quick demonstration of power and next game please. Things didn't turn out quite as smoothly. Gaviota held its own for a long while, all evals close to zero and material being reduced. After 35 moves there were only RN vs RB left, with no significant weakness for either side.

This was surely a draw, but you can never let your guard down against Stockfish. After a few innocent moves (errors by Gaviota?) Stockfish saw an opportunity to force a rook exchange.

Suddenly the black bishop has targets and white's pieces are too weak to defend all of them. A few moves later it was all over.

Thursday, June 9, 2016

Season 9, stage 1b, rounds 4-9 overview

There have been so many games on TCEC while I was away, and I have no way of going over all of them while the tournament continues. I'll just have a quick look at the results and write about things that look interesting.

Gaviota managed to draw Naum in the longest game of stage 1b (so far). Gaviota had an eval advantage but this only reflected a space advantage in a blocked position. The engines avoided the 50 moves draw several times, dragging the game for a long time.

Andscacs beat Chiron in another long blockade game. Andscacs found a way to get a decisive advantage, though parts of the game seemed like random shuffles. A win against one of the stronger engines of the stage shows that Andscacs is contender for qualification.

Arasan was able to hold Andscacs to a draw, and then it crashed. What a missed opportunity.

Stockfish beat Naum with black pieces. All the pieces were concentrated on the king side, and Stockfish's pawns limited Naum's movement until it was ready for the killing blows. This was the first strong engine Stockfish beat this season.

Jonny - Fire ended in a draw. Jonny is doing well so far in this stage, better than I expected.

Stockfish - Nirvana was a 21 move miniature that ended in a 3-fold repetition. Nirvana is also doing very well so far, and Stockfish is not dominating the field.

Andscacs couldn't get more than a draw against Fruit, a little disappointing for the ambitious newcomer.

After 6 rounds the top of the table was:
Nirvana 5
Jonny, Stockfish, Andscacs 4.5
Naum, Chiron, Fire, Critter 4
Vajolet2 3.5

Chiron lost to Jonny after it went for an early pawn sacrifice which did not work. Another important win for Jonny in the qualification race.

Naum beat Andscacs in a long and technical RB vs RB endgame, using an advanced passed pawn on the queen side to win pawns on the king side. An important result for both engines in the qualification race.

Nirvana beat Critter in a rook ending. Critter playing with tablebases seemed to blunder a drawn ending, and Nirvana playing without tablebases saw the win 10 moves before Critter did. Critter will have to work hard to qualify, while Nirvana's performance is excellent.

Gaviota was a pawn down against Fire, but managed to hold a pawn line, filling the gap with rooks. When Fire broke through to avoid the 50 move draw Gaviota was able to draw with perpetual checks. Respectable result for Gaviota, Fire still in danger considering it also crashed earlier.

Andscacs had an eval advantage over Nirvana but it never seemed that Nirvana was in any danger and the game ended in a draw. Another solid result for both engines.

Vajolet2 could only get a draw against Fruit. Vajolet's qualification chances are declining.

Arasan - Jonny reached a RN vs RB endgame on move 16, Jonny sacrificed pawns to get a dangerous passer but Arasan got the draw through perpetual check. Arasan won't qualify but it can still be a tough opponent.

Naum - Vajolet2 was a long and complicated game. In the bishop ending Vajolet2 kept seeing a Naum win while Naum was shuffling without an idea of how to proceed. Ironically it was Vajolet2 that moved a pawn, resetting the 50-move counter and letting Naum see the path to victory.

I would have thought that the Fire - Stockfish game would be the fight for first place of stage 1b, but that is not the case. In the game Stockfish never had an advantage but it also never gave Fire anything to work with. Fire thought it had something near the end but it was just an illusion, the game ended in a draw.

After 9 rounds the top of the table was:
Nirvana 7.5
Jonny, Stockfish, Naum 7
Andscacs, Chiron, Fire, Critter 6
Vajolet2 5