Season 9, stage 3, RR 1

Stage 3 has started, 8 of the best chess engines fight for a place in the superfinal of the season. We have a new version of Houdini, will it change the balance of powers, can it compete with Stockfish and Komodo? In this stage each engine is given 150 minutes + 15 seconds increment per move. Many games are more than 5 hours long, everything is slower than the previous stages.

Important Note: In the next two weeks I will be busy with other summer activities. Therefore I may be late in posting and my posts may be less detailed.

Table after 1 RR

The draw rate is 19/28 (68%). Cato did his best to choose openings that give advantage to white but do not force a win, and in most cases black manages to get a draw. Only one black win so far, and that was quite a surprise.
Komodo and Stockfish are at their natural 1-2 places, Houdini's new version seems better than before but I don't think it will be able to challenge the leaders. Rybka will probably be the easiest engine to beat this stage, recall it qualified last as well. It remains to be seen if Jonny can perform as well as it did in stage 2.
The games of the next RR will complete the reverse color game pairs, and then the rankings will be more meaningful.

Some of the game highlights

Houdini - Komodo: What a start to the stage. In the opening game the new Houdini used the opening advantage to defeat the champion Komodo. As this was going on, while Houdini fans were declaring its return to greatness, the Komodo developers noticed that the settings for Komodo were not what they instructed. The organizers decided to replay the match using the right settings for Komodo, and you can imagine the craziness that went on in the chat around this decision.
In the replayed game Houdini had an eval advantage and an imbalance of RPP for BN. There were only a few pawns but many pieces on the board, Komodo had to defend but it wasn't in any apparent danger. The RB vs BBN endgame was long but always a draw. Did the settings change anything? Was the first win just luck? Who knows.

More problems at the start: 6 hours down time, no live coverage of games 2-3.

Andscacs had an eval advantage from the start against Rybka. Rybka played defensively from the start, not moving a pawn for more than 45 moves and just waiting to see what Andscacs will do. Evals increased a little after the pawn move, and then the game became static and it seemed the 50-move rule would be applied. Eventually, with both engines in blitz mode with very little time, Andscacs did find a way to break Rybka's defenses on the king side, creating a significant mating threat and winning the game. Did Rybka blunder in time trouble?

Stockfish took advantage of the favorable opening it had against Jonny. Without stopping to castle it applied pressure on Jonny's pieces, forcing them to retreat until they almost could not move. An opening on the king side was enough for Stockfish to attack Jonny's king, evals jumping fast. The game ended before the action started, but the PVs showed significant material advantage for Stockfish.

In its second game Komodo was back to normal with no technical problems. It started with an eval advantage against Fire and kept it alive for 30 moves. It repeated a move twice to confuse Fire, and us, but then found the path to victory. A strong king side attack led to a superior ending with an advanced passer and a bishop pair.

Houdini - Andscacs: Very early in the game Andscacs created a defensive line and played waiting moves, challenging Houdini to try to attack. When Houdini finally attacked it gave a rook for a bishop but exposed the black king, with evals climbing to around 1. Andscacs defended well, exchanging down to a Q vs RR position with very few pawns, and the game ended in a draw.

Stockfish was able to get a pawn advantage against Houdini, but Houdini exchanged everything and got a draw.

Komodo had a big eval advantage right from the start against Jonny. Jonny gave a rook for a bishop with the hope that a bishop pair will be enough as defense. It also tried a king side attack but it was weak. Komodo was about to win material in a QR vs QB ending when the game was stopped. Will Komodo hold the reverse game?

Houdini got into a double rook ending with Gull, and although material was equal Houdini's had the initiative. That was enough to win a pawn and reach a winning pawn ending.

Andscacs - Stockfish reached a RR vs RBN ending. Andscacs had what looked like dangerous connected passers but Stockfish kept calm with eval at zero. There was some excitement near the end when Andscacs' eval jumped to nearly 4, but this was due to lack of tablebases, the final RP vs B position was a draw.

Stockfish - Komodo: another draw, third one this season.

Jonny had a pawn advantage and an eval advantage right from the start against Rybka. However, with Jonny's performance so far in the stage and its tendency to exaggerate it was doubtful it would win the game. Then the endgame was RB vs RB with opposite color bishops, and the pawn advantage was a doubled pawn, sure signs of a draw. Yet the eval advantage persisted. Somehow, Jonny found a way to infiltrate with a king and rook, to create a passer and to get a winning position. A big surprise for me.

Komodo had a small eval advantage from the start against Rybka, and Rybka was able to hold for about 30 move before it started to play inaccurately. In the RN vs RB endgame Rybka had an advanced passer but Komodo controlled the center and had a stronger rook. It was a slow process but Komodo succeeded in taking out the passer, and then the central pawns started marching and it was all over. Two defeats in a row for Rybka.

Stockfish eval advantage against Gull became a position where material was equal but Gull had weak pawns and an exposed king. The game reached a queen ending with a pawn advantage for Stockfish. The win was clear for Stockfish, especially after it took another pawn. Gull, with no tablebase support, took longer to convince.

Fire used the opening bias to get an eval advantage against Stockfish, and it slowly increased to about 1. To get out of trouble Stockfish sacrificed a knight and got an attack on the white king. This led to a series of exchanges and a QN vs Q position, and the white king was too exposed for anything but perpetual checks. 

Featured game: Andscacs - Fire
Stage 3, round 6
Link to game on TCEC

Andscacs had the opening advantage, but the evals did not go over 1. After several exchanges in the center, including the queens, Andscacs was able to reach the 6th rank with a passed pawn supported by a bishop.

This looked encouraging for white. But Fire blocked any more support for this pawn and it couldn't advance further. The evals came back down to zero, so probably a draw, right? Then Andscacs made an unexpected move.

Fire liked a4 quite a bit, causing Fire's eval to jump to (negative) 0.33. I don't know what was behind Andscacs' move, it allowed Fire to have a supported passer and to free its bishop. The evals continued to rise for black and a few moves later Andscacs chose to give a rook for a knight.

Most probably Andscacs valued the bishop pair and the two passed pawns, and was willing to lose some material for them. This turned out to be a serious mistake. After trading bishops suddenly the white pawns were very weak and easy targets for the black rook.

It wasn't long before the king side pawns were all gone, and the black passer was going to cost at least the white bishop for a nice black win.This was a very surprising result, the only black win, Fire's only win of the round and Andscacs' only loss.

Season 9, stage 2 stats

Draw rate, wins

Final draw rate was 55%. There was a preference for white in the decisive games, 3 times more than black wins.

Game termination

The three most common game termination causes were:
40.8% - TCEC win rule
30% - TCEC draw rule
14.2% - TB position

There were 5 crashes, not a lot but too high for my liking for a second stage. The last one by Jonny shows that we may get more in stage 3.


Moves per game

Median= 64.5
Average= 70.61

The games were a bit longer than in stage 1, about 5 moves more.


Time per game (hours)

Median= 3:53
Average= 3:48

Despite having a few more moves the game duration was comparable to that of stage 1.


Openings

There were 8-move book openings in this stage chosen by Cato, so the engines had almost no say. The first letter of the ECO codes was distributed as follows:

A small bias for Sicillian openings. There was a similar graph in stage 3 of season 8.


If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
Sicillian - 58 times
English - 24 times
French - 20 times

The 8 predetermined moves usually made sure that the reverse pair games had the same opening definition. There were 7 cases where the full opening name was different, of these 3 had different ECO codes.

Reverse pairs, wins

Only 7/120 openings are suspect of a bias - each engine wins once, while 60/120 openings have a win+draw combination - a proof of engine strength, winning from one side and holding the other side. Cato is my hero.

Reverse pairs, same moves 

Most reverse pairs diverged within the first two moves. One game had the same move sequence for 17 plys (Protector - Chiron, 1/2-1/2).

Season 9, stage 2, rounds 28-30, final results

Final results of stage 2 after 30 rounds:

The final draw rate is 132/240 (55%).
The qualifiers for stage 3 are: Komodo, Stockfish, Jonny, Gull, Fire, Andscacs, Houdini and Rybka. Congratulations to all!
Komodo won the stage with Stockfish a close second. Will these two be in the superfinal yet again? Stage 3 will have a new version of Houdini that is an unknown element, perhaps it will surprise everyone.
A win in the last round was enough for Rybka to qualify for stage 3. Ginkgo could not close the gap and came half a point behind.

Some of the game highlights

Andscacs - Rybka started more or less equal, but at some point the evals for white started to increase. The white knights took control of the center and were making life hard for the black rooks. By the time Rybka managed to get rid of one knight its rooks were stuck on the queen side, hardly in the game at all. Out of desparation black gave a piece to create a passer, but Andscacs was in full control for a nice win. Can Rybka's rivals close the gap?

Hannibal had an eval advantage over Jonny, but considering Jonny's performance in the stage I expected the evals to go back down eventually. There was a long series of moves with evals not moving too much and the position more or less the same. Hannibal had an advanced passer that wasn't moving, and Jonny gave a rook for a knight in the hope that its knights will be stronger in a position with many pawns. Hannibal then opened up the queen side through a series of exchanges, and in the R vs N ending had a clear advantage. A very unexpected loss for Jonny, first in the stage. Could Hannibal be making a comeback? Not likely, it's still a point down on Rybka with two games remaining.

Nirvana could only get a draw with black against Chiron, a missed opportunity to get closer to Rybka. A point down may be too much for Nirvana.

Protector played the opening better than Vajolet2, and as a result had a strong bishop pair in a BB vs BN ending. Together with a dangerous passer this was enough for Protector to get to a winning position.

Komodo - Fire got into a closed position with all pawns on the board. Fire had less room and was mostly waiting while Komodo planned how to break the pawn line. By move 40 a few pawns were exchanged and the evals were at 1.5. several moves later Komodo was ready to execute its plan. After a series of exchanges only RN vs RB remained but Komodo was two pawns up, an easy win - at least easy for Komodo. You need to be over 3000 to fully appreciate a game like this... First loss for Fire that can't be attributed to the opening.

Stockfish quickly took over control playing black against Naum. Using an almost unstoppable passer Stockfish could arrange the pieces so that after exchanging everything the position was a winning 6-man rook ending. Stockfish has the same score as Komodo, but Komodo takes the SB tiebreak.

Ginkgo managed to get a draw against Houdini, now only half a point behind Rybka.

Ginkgo gave a bishop against Raptor, demolishing the pawns defending the black king. The resulting attack on the king finished Raptor off very quickly. Ginkgo is temporarily in 8th place leading by half a point, Rybka plays last in this round.

Houdini had an eval advantage against Naum, and the evals gradually increased. After the opening stage was over Houdini had a strong center with a supported passer, while Naum's counter on the queen side was weak. After pawns were exchanged on the king side Naum had to deal with mating threats as well and its position crumbled.

In their second game this season Stockfish and Komodo drew again. Stockfish tried to keep a small eval advantage but there was nothing it could do against Komodo's defense. Komodo is still leading on SB tiebreak.

Nirvana - Hannibal ended in a repetition draw. This was a bad result for both engines. Hannibal is now out of the race being a point down and with a crash handicap. Nirvana is also a point down, almost no chance of qualifying.

Jonny was ahead on eval the whole game against Andscacs, but it didn't seem that Andscacs was ever in any danger. The game ended in a draw before the 50-move rule kicked in.

Rybka - Gull was another draw, a quick one this time. Rybka is leading Ginkgo on a tiebreak before the last round. Who shall qualify?

Raptor was happy to take a rook for a bishop against Rybka, but Rybka used this to turn the game around. In a RR vs RB position the black pawns and bishop controlled the center and the white rooks were playing defensively. Rybka's pair of advanced connected passers were enough for a win. With this result Rybka is the qualifier. It has a point advantage and a big SB lead in case of a tiebreak with Ginkgo. The games remaining are just to complete the stage.

Gull - Jonny reached a queen ending with a pawn advantage for white. Jonny found a perpetual check draw but then crashed, a disappointing end to a glorious stage by Jonny.

Hannibal came out of a closed position with better pawns and piece coordination against Vajolet2. A supported passer gave Hannibal the win.

Chiron was outplayed by Fire. When almost all pieces were exchanged Fire was a pawn up with two advanced passers in a B vs N position, an easy win.

Protector could not hold back Stockfish for long. The white king remained uncastled and Stockfish had a lot of firepower in the center and on the king side while its king was hiding behind a white pawn on the other side. Stockfish broke the pawn defense with a knight sacrifice and the game was over before the whole position collapsed. Stockfish has a good chance of winning the stage, Komodo needs a win in the next game (see below).

Naum - Ginkgo was a long draw, Ginkgo finishing 9th half a point behind Rybka.


Featured game: Komodo - Houdini
Stage 2, round 30
Link to game on TCEC

Komodo needed a win to beat Stockfish for first place in the stage (not really important but still...).
The game started with evals close to 0 and no clear advantage for either engine. Material was equal but the position was not static with chances for both engines. On move 32 Komodo seemed to offer Houdini a rook, a poisoned offer?

Houdini did not accept the offer, though it could for several moves. I can only guess that the passer on the e file could become dangerous and that the bishop pair would be strong enough to compensate. Komodo's eval started to climb and Houdini was feeling the pressure, finally it took the rook.

Note the rook on h6, what is it doing there? Already it cannot get out safely though it can move a little. The evals started to increase rapidly as Komodo threatened to open up the position. With a strong bishop pair staring at the black king and black playing with a rook handicap Houdini was in real trouble.

The h5 rook is really miserable, at least the king is keeping it company. When pawns and pieces started to be exchanged Komodo's advantage was very clear. The game ended in a BB vs RB position.

The black king side is virtually frozen, the rook can be taken any time and the black bishop cannot defend against all the threats on its own. Komodo is the well deserved winner in this game and of the stage.

Season 9, stage2, rounds 25-27

Table after 27 rounds:

The draw rate is 121/216 (56%). 
Stockfish is now leading the table after closing the gap to Komodo with three straight wins.
In the race for 8th place Rybka beat Hannibal, a rare win both for Rybka and in games between 8th place contestants. This result puts it in front a full point ahead of Ginkgo. With three games remaining this could be enough. Remaining games are (underlined opponents playing black):
Rybka: Andscacs, Gull, Raptor
Ginkgo: Houdini, Raptor, Naum
Nirvana: Chiron, Hannibal, Andscacs
Hannibal: Jonny, Nirvana, Vajolet2

Some of the game highlights

Stockfish played the opening much better than Raptor. It was ahead a rook for a bishop in a RR vs RB position with evals at 4. Raptor had one last counter-measure, an advanced passer threatening to queen. Stockfish had no real problem getting rid of the black pawns before taking the passer and the bishop for a rook. The position was a 6-man won rook ending, but then there was a repeat of the Fire - Vajolet2 game: Stockfish played non-optimal moves and Raptor refused to give up. Sadly, this went on for 50 more moves before finally Raptor saw it was lost.

Fire had a space advantage against Houdini, the center and king side pawns were locked but Fire had a stronger queen side. The evals steadily increased as Fire created a 7th rank passer which was eventually turned into a bishop advantage for a win. Houdini had a big problem with this opening, losing from both sides.

Vajolet2's eval advantage against Ginkgo was quickly reduced to nothing, and then Ginkgo developed a strong king side attack. Starting with a knight sacrifice Ginkgo took out the white pawns defending the king. There followed a series of exchanges which resulted in a RBN vs RBN position with equal material. However, Ginkgo found a way to win (Vajolet2 blunder?) by exchanging down to create an unstoppable passer.

Jonny thought it had something against Komodo, even what appeared to be a solid pawn advantage. Komodo remained cool, defending and exchanging pieces until reaching a drawn rook ending. Jonny still doesn't have what it takes to beat Komodo.

Rybka went a pawn up early in the game against Protector. Many exchanges later and after a gradual eval increase the position was RN vs RB and Rybka won a second pawn. As the pawns marched the evals continued to increase, Protector had to lose a piece to avoid queening, losing the game.

There were many early exchanges in Gull - Chiron, and by move 30 the position was a N vs B endgame. Gull had more space and Chiron was playing defensively but the game was heading for a draw ... and then Chiron crashed yet again.

In Hannibal - Gull a complicated set of attacks and pins in the center was resolved into a bishop endgame with a pawn advantage for Gull. Hannibal thought it was losing immediately, perhaps giving up too soon, but Gull did find the win eventually.

Komodo couldn't find a weakness in Nirvana's position for almost 50 moves. When Nirvana made the crucial inaccuracy Komodo's evals started to increase rapidly. With clever queen maneuvers it won a pawn and created a passer. Nirvana couldn't defend its king and the queening threats, the game was soon over before we saw the actual collapse.

Naum - Vajolet2 was on the sure path to a draw. Naum had an eval advantage in a QBB vs QBN position but both engines were shuffling for many moves and the 50-move rule was getting close. After Vajolet2 moved a pawn Naum's eval started to increase. Naum found a way to get a pawn advantage and then exchange all pieces to get a won tablebase position. First win for Naum in this stage, all engines have at least one win this stage.

Houdini had a strong pawn attack on Raptor's king, resulting in a RB vs RB endgame with a pawn advantage for Houdini. The black king hid behind a white pawn but was also trapped as a result. When Houdini's king marched forward it made the difference, and the game reached a rook ending with two pawns advantage, and easy win for Houdini.

There were many early exchanges in Stockfish - Ginkgo and the game reached a RNN vs RNN position. For 35 moves the engines shuffled their pieces with almost no pawn moves, the game was clearly a draw. Ginkgo moved a pawn and Stockfish's eval jumped a little. Another inaccuracy and the evals started to rise fast. Through careful maneuvering Stockfish was able to exchange some pawns and get a crucial passer. Ginkgo was forced to lose a knight for the passer, leading to a RN vs R ending. For the final blow Stockfish gave back the material to reach a won king and pawns ending, really beautiful in the PV but not played out.

Fire only got a draw against Naum, Vajolet2 held a draw against Komodo. Disappointments for the top engines, Komodo slipped to second place behind Stockfish.

Nirvana won a rook for a bishop in the start against Protector. However after queens were exchanged it turned out that the black bishop pair was stronger than the extra white rook. Protector's eval started to climb as it managed to slowly take white pawns. In the open position Nirvana's king started to be in danger and Nirvana gave back a rook for a bishop, but then it was three pawns down. It took a while but Protector's win was inevitable. A blow to Nirvana's qualification chances.

Jonny had to settle for a draw against Chiron, another top engine disappointment.

Rybka - Hannibal had many imbalances and an intense struggle in the center, but after most pieces were exchanged the game reached a rook ending. The material was equal but Rybka had an eval advantage with its pawns more advanced. However, it was not clear whether the advantage was enough for a win. Indeed, after several pawn exchanges the position was a 7-man draw. Hannibal did not have tablebase support, and as a result it blundered and let Rybka win the endgame. Rybka is now first in the race for 8th.

Gull was up a bishop for pawns against Andscacs. The white king was exposed but it managed to escape perpetual check. Still the game appeared to be heading for a draw. Twice the TCEC draw rule was approached and missed at the last moment, Gull insisted it was slightly in the lead. Then suddenly it really was ahead, it got a pawn back and created a passer. After queens were exchanged in a RN vs R position the last white pawn proved unstoppable and cost Andscacs the rook, giving Gull the win.


Featured game: Houdini - Stockfish
Stage 2, round 26
Link to game on TCEC

The game started with an eval advantage for Houdini, but Stockfish equalized rather quickly. The first sign of trouble for Houdini came when it threatened to win a rook.

Stockfish thought the knight should be exchanged for the e6 bishop and was happy to take it with the d6 bishop even for a rook. Houdini then chose to leave the rook and took back the bishop. A bit strange, and an eval jump for Stockfish. Now Stockfish had strong central pawns, and the a5 bishop was in danger. A few moves later it found safety in the corner.

The bishop was completely trapped! After a long series of shuffles the game reached a QRB vs QRN position, with the bishop still trapped. Stockfish had time to arrange the attack and at the right moment the pawn advanced.

Houdini's position was in ruins, the bishop was free to move but Houdini had to face queening threats and defend its hanging pawns and its king. Stockfish won a pawn and after queen exchange won another.

Another beautiful win for Stockfish, the second win against Houdini in this stage.

Season 9, stage 2, rounds 22-24

Table after 24 rounds:

The draw rate is 111/192 (58%). There have been 72 openings played from both sides so far, and in only 4 both engines won - potentially a biased opening. One of these was due to a crash. Another feather in Cato's cap.
Komodo increased its lead to a full point. It had two quality wins against Gull and Andscacs. In the race for 8th place the only thing that changed is that Rybka has made a small step forward. Hannibal's crash is also important since it will now lose all tiebreaks automatically. Now only half a point separates the 4 engines in the race, we may not know the qualifier until the last round.

Some of the game highlights

Raptor - Chiron was a draw after 217 moves, a close second to the record for this stage, 219 moves.

Protector and Hannibal started the game ok, but in the 15th move Hannibal had a very long think. After more than 80 minutes, when all fans were sure that it was locked in an infinite loop, Hannibal finally moved, and immediately crashed. A serious setback for Hannibal's qualification chances.

Stockfish had an eval advantage from the start against Nirvana, and it grew steadily. Nirvana did not castle and had a weak bishop in a QRN vs QRB position, with only a threat of an attack on the white king. Stockfish hid its king behind a black pawn and then it had better placed pieces and a dangerous passer. Nirvana gave two pawns to force a queen exchange but the game was already lost.

Fire - Vajolet2 started with many early exchanges and evals close to zero. At some point Fire's eval started to grow to about 1, and the game reached a queen ending which can be very tricky. Fire's advantage was tablebase support, which Vajolet2 lacked. In one move Fire's eval jumped to over 9, it won a pawn and threatened to queen another. As a result Fire got a two pawn advantage and the game reached a won 6-man position. Funnily, Fire kept playing non-optimal moves while Vajolet2's eval kept resetting the win rule, and the game continued for another 20 moves until a pawn was taken.

Fire couldn't beat Raptor with white, Stockfish couldn't beat Vajolet2 with black, a disappointment to both. Nirvana tried hard to beat Houdini, but only got a rook ending with f+h pawns, a known 6-man tablebase draw.

Jonny had a very effective king side attack against Ginkgo. Ginkgo tried to exchange pieces to get out of trouble but ended up a bishop down in a RB vs R ending. This was an easy win for Jonny, continuing its good performance this stage.

Gull - Komodo was another masterpiece by Komodo (see featured game). Gull opened a file in front of its king allowing Komodo to apply pressure. When Gull blocked with pieces it also supported these with its king, creating pin opportunities for Komodo which ultimately led to Gull losing a bishop. The game continued for more than 20 moves with evals slowly increasing and Komodo targeting the white pawns until adjudication. Impressive that Komodo beat Gull from both sides of this opening.

Andscacs' pieces were much more active than Protector's, and this was reflected in the eval. When the position opened up a little Andscacs also had a serious threat against the black king. Most pieces were exchanged leaving a BN vs R ending, but black couldn't defend its pawns. It was two pawns down when the game was stopped.

Tired of crashes Hannibal and Chiron played a 14 move miniature with a repetition draw.

Andscacs had more space in a closed position playing black against Chiron. There were long stretches of shuffling moves in the game. The first breakthrough came when Andscacs trapped a white rook, and then brought a knight from the other side of the board for an exchange. The second breakthrough was 40 moves later when the black queen found a way behind the white pawn line. Andscacs was declared the winner a few moves later, a double win against Chiron.

Naum held its position against Jonny for over 40 moves. Naum had a pawn advantage but Jonny's pieces were much more active while Naum's pieces could hardly move. After a few exchanges Jonny had advanced passers in a RN vs RN position, and after the rooks were exchanged the game was over. A first double win for Jonny.

Ginkgo - Nirvana was important in the race for 8th place. Ginkgo had an eval advantage of over 1.5 at some point, and the game reached a QN vs QB ending two pawns up for white. However Nirvana was able to draw with perpetual check.

In Houdini - Vajolet2 the black queen wandered far from its lines and was a bit caught up in the center. Houdini saw a blunder and immediately its eval jumped over 3. The impressive thing was how Houdini planned almost all the remaining moves in advance. For over 25 moves it stopped to think twice, and other than that played blitz with under 5 seconds per move. After exchanging almost all pieces the game reached a won 6-man position. A first double win for Houdini.


Featured game: Komodo - Andscacs
Stage 2, round 22
Link to game on TCEC

Andscacs' performance this stage has been quite strong, and its qualification for stage 3 is almost certain. Could it withstand Komodo's power with the black pieces?

The game started with a small eval advantage for Komodo, Andscacs defended well for about 25 moves before starting to play small inaccuracies that allowed Komodo to improve its position and increase the evals. Komodo had a dangerous passer in the center, and Andscacs had to block with pieces and was feeling the stress. The black king tried to help with the defense of a knight, but was that a good idea?

By moving to g7 the king self-pinned the knight, and when the rook on d6 was eliminated the king too couldn't move so as not to lose the knight. Andscacs was getting under more and more pressure, after rooks were exchanged the passer became too big a threat.

With the black knight pinned and the black queen in the corner too far to assist, Andscacs had no choice but to lose a piece to get rid of the passer. The evals were over 2 and rising, but with only QBN vs QN did Komodo have enough material for a win?

The pieces danced on the board with slowly increasing evals but it was difficult to see what was the improvement. After 12 moves a first breakthrough for Komodo, it was able to force a knight exchange.

And the evals jumped to 250 ! A sure win for Komodo, though how it was going to do it was still a mystery. The experts in the chat indicated that all that is needed is a queen exchange to finish the game, the bishop and king were enough to get rid of black's pawns. But the PVs did not explain how Komodo was going to do this. The pieces continued to dance for more than 30 moves, Andscacs stayed cool and refused to exchange, and the 50-move rule was getting closer and closer. And then, finally, Komodo found the right series of moves.

Andscacs could not avoid the queen exchange and the game was finally over. Beautiful game, great display of Komodo's strength.

Season 9, stage2, rounds 19-21

Table after 21 rounds:

The draw rate is 97/168 (58%). The last three rounds have been more drawish. 
Komodo and Stockfish are still 0.5 points apart, Jonny is still unbeaten but Fire has lost against Nirvana, its first loss of the stage. In the race for 8th place Nirvana has improved its chances with this win and is now leading. With 9 rounds to go all four engines ranked 8-11 have a fighting chance. There are now 8 games in their internal crosstable, only one is decisive.

Some of the game highlights

Nirvana started with an eval advantage against Raptor, but the evals dropped back to zero later for both engines. Both kings were exposed and there was a possibility of a perpetual check. Nirvana gave a rook for a bishop but the evals did not move. Raptor kept trying to draw but Nirvana was playing to win. After queens were exchanged the two advanced passers were enough for a win in a RBN vs RRB position.

Jonny had a pair of knights against Vajolet2's pair of bishops, in a closed position where knights were much better. After locking the queen side Jonny sacrificed a knight to open the king side and expose the black king. Vajolet2 had to give the material back to defend but this left a RR vs RR position with Jonny two pawns up for the win.

In Andscacs - Houdini white had a space advantage and the black pieces had little room to move. Andscacs used its advantage to clear the black queen side pawns and create dangerous passers. After queens were exchanged Houdini was two pawns down, Andscacs won a third pawn before the game was stopped. This was the first opening where both engines won with white.

Chiron survived more than 75 moves playing black against Komodo, but then its position collapsed. I don't know if Chiron blundered or not, but Komodo managed to exchange pieces and get to a winning bishop ending with better pawns.

Houdini - Gull was by far the longest game of the stage so far with 219 moves and 5:44 hours of play. At first it seemed that Houdini was in some danger from Gull's attack on its king. However, the position became static and both engines were shuffling and the game became boring. The game was extended by pawn moves and exchanges until finally it was over. Thankfully there were no 50-move game changing blunders.

Stockfish created a pair of advanced passers in the center against Rybka, and this forced Rybka to give a bishop for 3 pawns to get rid of them. Although arithmetically material was equal, in practice this led to a N vs pawns ending for white and the extra piece was enough for winning.

Nirvana - Fire reached a rook ending quickly. Nirvana was a pawn up which usually is not enough, and it also had no tablebase support, which Fire had. Therefore it was quite a surprise when gradually Nirvana's eval increased, and after several pawn exchanges the position was a won 7-man position for white, and both the engines saw the win. Another opening with two wins by white.

Jonny - Stockfish was a draw without any incident, but there was some talk in the chat that Stockfish may have missed a win in the endgame. I wonder...

Ginkgo was defending with black against Gull, and it chose to give a queen for RB. At first it seemed it had a solid fortress, but when pawns started to be exchanged (Ginkgo blunder?) Gull's queen became more dominant. When Gull took two black pawns the game was over. Double white win opening again. 

Andscacs and Naum exchanged pieces quickly reaching a RN vs RN position. The white pieces were much more active, leading to a rook endgame with a pawn advantage for Andscacs. After taking another pawn the game was practically won for white. Andscacs with wins from both sides of this opening.

Featured game: Fire - Jonny
Stage 2, round 20
Link to game on TCEC

Fire - Jonny was a clash of the 3rd and 4th ranked engines of the stage, both unbeaten so far. Fire had an eval advantage but pieces were exchanged very quickly and a queen ending was reached.

Queen endings can be very hard to predict. Both engines started to advance passers, and with the support of a queen the passers were hard to stop. Then came the highlight of the match:

Four queens on the board !! So much firepower, so many ways to mate. However, since both engines had two queens whatever Fire tried in its attack Jonny could find a defense. In the end one pair of queens was exchanged, the evals went down as did the excitement, nothing but a draw.

Season 9, stage 2, rounds 16-18

Table after 18 rounds:

The draw rate is 81/144 (56%).
Remember that the first 12 games of the stage were draws? 8/12 of the rematches were decisive ! Another confirmation of Cato's excellent opening choices, as if it is needed. Cato is the best.
Komodo and Stockfish are leading, only 0.5 points separating them. Jonny and Fire are still unbeaten in the stage. The battle for 8th place is fought between Ginkgo, Rybka, Hannibal and Nirvana. Only one of seven games within this group has been decisive so far.


Some of the game highlights

Raptor became very excited against Naum, a rare occasion for Raptor in this stage. It had a strong queen side attack on the black king and Naum's pieces were on the other side of the board. First Raptor won a rook for a bishop, then Naum's counterplay resulted in further exchanges leaving Raptor a full knight ahead. When all other pieces were removed the knight guaranteed the win for white.

Ginkgo's king came under attack very quickly against Komodo. Evals for black increased rapidly as Komodo gave a rook for a bishop, exposing the white king to its QBB. Ginkgo exchanged pieces to stop the attack, reaching a RBN vs RBB endgame two pawns down. Komodo continued patiently to improve its position, taking pawns and exchanging pieces slowly. The game was stopped in a bishop ending with Komodo four pawns up.

Houdini - Protector had stable pawn lines and stable evals that were close to zero for about 50 moves. Then a few pawns were exchanged and white's evals increased a little, but still it seemed Protector was holding for another 20 moves. The first breakthrough came when Houdini won a pawn, and then exchanged a rook for a knight and pawns. The evals were above 2 in a RNN vs RRN position, and stayed there for a while. Houdini finally sealed the win when the pawns started to march and Protecor started to lose material. Slow win, 114 moves.

Stockfish was up a pawn against Chiron at the start of the game. The evals started jumping when Stockfish was able to get a rook and a queen to the black back rank, the win was just a matter of time. The game was stopped with Stockfish a knight and three pawns up.

Vajolet2 - Andscacs started with an advantage for white but Andscacs equalized and then took the initiative. For a long time the evals remained stable as pieces were being exchanged. In a R vs BN position with two king side pawns for each engine I was sure the game was a dead draw. However, instead of the 50-move rule suddenly the evals started to jump again. Andscacs found a way to get the king to help and take a white pawn. The game was won with a queening threat for black. Not using tablebases - well done Andscacs.

Jonny had an eval advantage against Raptor from the start. For a while it seemed Raptor could hold the pawn line but Jonny found the right moves to open up the king side. The white attack through the g and h files was deadly, game was stopped before Raptor lost material.

A king side attack gave Gull a pawn advantage over Vajolet2, and then through gradual maneuvering was able to attack the weak pawn structure of black and gain two more pawns. Vajolet2 tried to counter with an attack on the white king but Gull had no trouble defending. It was just a matter of time until the game was over.

Hannibal had an eval advantage against Stockfish, but Stockfish equalized and gave two rooks for a queen. After exchanging pieces and pawns the position opened up and Stockfish gained the initiative in a RRN vs QB endgame. It won a pawn and created a passer, while the black king found relative safety in the center. Hannibal could not stop the passer until it had to commit a rook for blocking, the game was over a few moves later. Stockfish won both sides of this opening, a first for this stage.

It was a mini surprise that Komodo could only draw against Naum. Komodo had a clear eval advantage out of the opening but Naum equalized quickly and most of the game was waiting for a draw with evals at zero.

Raptor chose to exchange RN for a queen early against Komodo, and both engines thought they had an advantage. This went on for more than 25 moves, Komodo's eval kept increasing until Raptor realized it had a problem. While it was keeping the rooks out of its back ranks Komodo grabbed a pawn, advanced its pawns and the evals climbed to winning values. The game stopped without showing us the path Komodo would take to victory, but the PVs displayed more pawn grabbing and exchanging down. Komodo with a first double win as well.

Ginkgo had an eval advantage against Chiron, but it was difficult to get past the pawn line. By exchanging rooks Ginkgo created a passer that forced Chiron to block it with a piece. When Ginkgo found a way to open up the position it just exchanged everything, grabbing pawns when possible. The end position was a rook ending with white two pawns up, three passers.

Houdini had an eval advantage over Hannibal, but it seemed that Hannibal's defense was strong. Both engines started shuffling pieces and we were waiting for the 50-move rule to end the game. Suddenly Houdini's eval jumped to over 3 and Hannibal agrees - what? If you think it's a blunder why make the move?? The whole position collapsed with pawns and pieces exchanged, and the end result was a winning rook ending with two pawns up for Houdini. 

Stockfish was up a rook for a bishop against Andscacs. For a while it seemed the black bishop pair was enough compensation for a defence, evals stayed more or less constant. In the RN vs BB ending Stockfish was able to find a win, perhaps it was the lack of tablebases for Andscacs that helped.

Featured game: Jonny - Rybka
Stage 2, round 16
Link to game on TCEC

Jonny was unbeaten so far in the stage (not counting Chiron's crash in a winning position...)
and generally playing better than what was expected. In the first round game Jonny played black. The game was uneventful and ended in a draw. Could Rybka, as a past champion, stop Jonny in white? 

After a series of exchanges in the center Jonny was a knight down for two pawns, one pawn was an advanced passer. The evals were still very close to zero.

Rybka was a little careless in the defence. It had several opportunities to exchange queens (Jonny expected this in its PV) but it chose to grab a pawn instead. In the meantime Jonny slowly advanced the e pawn until it connected with the d pawn.

BOOM. The evals jumped to almost 3. The threat of the pawn pair was too much for Rybka, and it immediately gave a bishop to take them out. By that time Rybka realized it was in trouble, and Jonny's evals were over 11.

Jonny had too many threats on the exposed black king. To get out of the mating net Rybka had to give significant material, and the game was stopped. Impressive, Jonny is playing well in this stage.

Season 9, stage 2 rounds 13-15

Table after 15 rounds:

The draw rate is 70/120 (58%).
Interesting to note that the top 5 engines have so far drawn every game they played against each other. Will this trend continue for the rest of this stage as well as the next?
Houdini has lost to all the top 3 engines of this stage. Even if it qualifies for the next stage it needs an update to be a real contender. 


Some of the game highlights

Gull exchanged a rook for a bishop against Raptor, but with three extra pawns Gull had the advantage. After almost all the pieces were exchanged the B vs R four pawns up was an easy win for white. Gull gets one back after the loss on the previous game.

Jonny had an eval advantage over Hannibal from the start of the game, and was a pawn up later. After exchanging pawns on the king side Jonny had two passers, and Hannibal had to control them and defend its remaining pawns. With many pieces on the board the going was slow, but Jonny concentrated its forces, took another two pawns, enough for the win.

Fire - Komodo was an interesting battle between two top performers of this stage (even though Fire is not considered a part of the top-3 engines). The game though was not very eventful. Fire had an eval advantage in a closed position, which grew temporarily when the position opened up a bit, but then fell back down. Komodo was never in real danger.

Stockfish had an eval advantage against Naum in a closed position, and it increased when Naum opened up the queen side exchanging a few pieces. Stockfish had a better pawn structure, and when the only pieces left were N vs B the white pawns were out of the bishop's reach and one was an advanced passer, enough for a win.

Raptor - Ginkgo reached a RB vs R endgame fairly quickly, where Ginkgo had four extra pawns for the bishop. Both engines thought they had an edge, I trust Ginkgo's eval more than Raptor's. The outcome was a queening for both engines and perpetual check, so the truth was somewhere in the middle.

In the much anticipated Komodo - Stockfish game nothing really happened. Stockfish's eval reached zero very quickly and Komodo followed, and the rest of the game was just exchanging pieces and waiting for adjudication. This reminded me of the season 8 superfinal where many games were short draws. When Stockfish and Komodo see a draw coming they lock their evals at zero and let TCEC call the draw.

Chiron was glad to give a rook for a bishop against Vajolet2, ruining the black pawn structure and with more active pieces. Many exchanges later it was the advancing passer that ensured the win. This was Chiron's first win of the stage.

Hannibal - Nirvana was very strange. Hannibal had a significant eval advantage but Nirvana held a static line in a semi-closed position with many pieces. After almost 50 shuffling moves, with 2:30 minutes on the clock, Hannibal moved a pawn and suddenly the evals jumped. It looked as if Nirvana blundered a piece, and suddenly Hannibal was a rook up and winning. What was that?

Gull executed a king side attack against Rybka, with most of the black pieces stuck on the wrong side and unable to help. The game was declared a win for Gull when it was a rook down, with the black king exposed and just before Rybka started to lose material to save it.

Rybka gave a rook for a bishop against Raptor, and in return had a passer and a strong bishop pair to back it up. Raptor gave the material back to get rid of the passer, but this allowed Rybka to create another passer on the other side of the board. Again Raptor had to give material, leading to a RB vs BN ending with one pawn each. White had the advantage and Rybka found the win, though it was not obvious for me.

In retrospect Vajolet2 - Hannibal was very funny. Hannibal's eval constantly increased in a RB vs RB endgame. It queened a pawn in exchange for the other pieces, leading to a RB vs Q position. This was perhaps a bad decision since Vajolet2 seemed to be holding its position for a long time, with evals fairly constant. Hannibal's eval jumped after it exchanged RB for Q again, and both engines queened. The 6-man position was a tablebase draw, yet Hannibal was sure it was winning (eval above 11).

Fire was down a rook for a bishop against Chiron, but its bishop pair was very strong in the open position and the black king was exposed. As a result Fire was able to win material and get to a won QB vs Q ending.

Stockfish's eval started to increase after 30 moves against Protector. It took over the center winning a pawn, and hid its king behind a black pawn. When Stockfish won another pawn it was the beginning of the end for Protector. The game was adjudicated in a RRN vs RBB position, both engines saw that eventually the extra pawns were unstoppable.



Featured game: Houdini - Komodo
Stage 2, round 15
Link to game on TCEC

Could Houdini playing white be a match for Komodo? I expected at least a draw. The game started with a small eval advantage for white, but Komodo soon equalized and got a small advantage of it own, even though it was a pawn down.

Komodo advanced the d pawn as a way of opening up the position. Houdini tried to block the pawn but couldn't in the end.

As a result Houdini gave a rook for a bishop and was a little behind on material. Komodo's eval advantage grew but the way forward was not clear.

The king side pawns were taken off, leaving Komodo with only the doubled c pawns. Why is it so happy in its position? Then I saw it, both the white knight and rook are trapped !!

Adjudication came quickly though the game was far from over. The PVs show that Houdini was about to lose material in a few moves.