Tuesday, September 27, 2016

Season 9, rapid tournament, rounds 17-24

Table after 24 rounds


Houdini is still going strong, 7 points in the last 8 games and 1.5 points clear in the lead. Komodo with 7.5 points climbed to second place, recall the slow start it had. Stockfish on the other hand dropped to third with 6 points (4 draws), tied with Ginkgo with a result of 7.5 points. Jonny and Fire, 5.5 points each, are a little behind. These six engines stand out leading the table so far. Houdini, Stockfish and Komodo remain unbeaten, though this may change in the next rounds when they meet each other.

The draw rate is very low, but many draws in the table are between engines in the top half of the table. For example 26 of 47 games (55%) within the top 12 engines are draws.


Games that caught my eye

Round 17
Chiron - Houdini: Chiron couldn't stop Houdini even with the advantage of white pieces. There were many early exchanges, and Houdini managed to get space advantage and attack with a pawn storm supported by the king. Houdini then converted the advantage to a winning N vs B ending with a pawn up.

Critter - Stockfish: Stockfish thought it was better in a RRB vs RRB position, then pawns were exchanged as well as the bishops and both engines saw that with rooks only the game was a draw. Stockfish now 1.5 points behind Houdini, tied with Fire. Jonny is in second place.

Round 18:
Fruit - Komodo: Komodo equalized quickly with many early exchanges. With RB vs RB remaining Fruit was able to hold a draw.

Stockfish - Jonny: After move 17 only RRB vs RRN were left and the shuffling began. There was almost no change until the drawn end almost 100 moves later.

Houdini - Critter: Another solid game for Houdini. It gained a space advantage in the opening, Gull's king and pieces were cramped in the corner. Houdini exchanged everything and got a winning rook ending, extending its lead to 1.5 points to Jonny and Fire, 2 points to Stockfish.

Round 19:
Gull - Fire: This was a long game. Fire thought it was better while there were still many pieces on the board. Then after most pieces were exchanged it was Gull who was optimistic. In the end though, Fire sacrificed a knight for the dangerous white pawns and Gull could only draw the N vs pawns endgame.

Jonny - Houdini: Jonny had space advantage in a closed position, but Houdini played better to reach a double rook ending a pawn up. The game became static and Houdini could not improve its position, leading to a draw.

Texel - Stockfish: Another long game, the evals stayed close to 0 most of the time. The game reached a drawn 7-man RB vs RB position that continued more than 40 moves before adjudication. Stockfish missed an opportunity to get closer to Houdini.

Round 20:
Nirvana - Jonny: Jonny was better out of the opening. It kept the pressure on for more than 80 moves while Nirvana remained calm in defense. The game ended in a draw.

Fire - Andscacs: This game was a draw from start to end, with evals close to 0. At least it didn't last very long.

Round 21:
Chiron - Fire: Fire's pawns formed a V shaped line blocking the whole board. When Chiron broke through it was a pawn up with only opposite color bishops left. The engines realized it was a draw 70 moves later. Fire continuing to drop down in the ranks, now in 6th place.

Round 22:
Fire - Critter: Fire had a big eval advantage after the opening and then chose a BN vs R imbalance. Critter's king was isolated and Fire launched an attack, eventually winning material and the game.

Round 23:
Jonny - Fire: The game was balanced for 40 moves until Jonny's queen took a poisoned pawn, losing time and allowing Fire to start a king side attack. Jonny tried to exchanged pieces and gave a rook for a knight but it's position soon became hopeless. Fire overtakes Jonny in the rankings.

Protector - Houdini: The game turned into a rook ending by move 30 and Protector held a draw without too much trouble. Houdini loses a half-point, can the other engines close the gap?

Hannibal - Stockfish: Stockfish tried hard to win, it kept pieces on the board and got an eval advantage. After 45 moves queens were exchanged and the game reached a BN vs BB position. Stockfish won a pawn but could do no more, the engines started a long shuffle that resulted in a draw almost 90 moves later. Stockfish remains 2 points behind Houdini, level with Fire. Komodo and Ginkgo used this opportunity to climb to 2nd/3rd 1.5 points behind Houdini.

Round 24:
Arasan - Ginkgo: After 20 moves it was a RBB vs RBN position. Ginkgo hoped Arasan would blunder, and it seemed to be the case when the position opened a litle and Ginkgo was up a pawn with evals above 1. Arasan had to give a bishop for a passer, reducing to a R vs RN ending. Ginkgo lost all its pawns however, resulting in a tablebase draw.

Stockfish - Rybka: The game seemed to be a shuffling draw when Rybka blundered and let Stockfish open the queen side and create a passer. This cost Rybka a piece and eventually the game.

Fire - Texel: Fire opened the king side with its king uncastled, allowing Texel to launch an attack on the white king. Texel sacrificed a rook for a bishop, but later regained the material. After the attack was over Texel had a clear win with 3 extra pawns. Excellent result for Texel, Fire dropping to 6th place.
 

Wednesday, September 21, 2016

Season 9, rapid tournament, rounds 9-16

Table after 16 rounds


Houdini in the lead ! With 7.5 points in the last 8 games Houdini is 1 point clear of the field. Stockfish had 3 draws, including one that was a huge surprise against Delphil. It is now second together with Jonny that had 8 straight wins. Fire dropped to 4th after two losses and a draw playing against the top contenders. Komodo is still 2 points behind the lead after a good result of 7 points. Only Houdini, Stockfish and Komodo remain unbeaten in the rapids.

Texel with a huge 7.5 result, and Raptor with 6.5 are the only 3000- engines in the top half, Laser is still over-performing with a 50% score. Naum is still under-performing, the only 3000+ engine in the bottom half with less than 50%.


Games that caught my eye 

Round 9:
Nirvana - Stockfish: Stockfish used a moment when all Nirvana's pieces were on the queen side to launch a king side attack. Stockfish used the exposed white king to get a material advantage until it was up a rook. Nirvana's first loss of the stage.

Gull - DisasterArea: Gull only managed to get a draw, another good results for DisasterArea.

Protector - Komodo: After 70 moves with both engines shuffling in a static position Komodo found a breakthrough. What looked like a draw became a winning N vs R endgame.

Round 10:
Komodo - Hannibal: An easy win for Komodo. The opening eval advantage turned into a winning RN vs RB ending.

Fizbo - Raptor: A very long game. Fizbo was up a rook against a knight and pawns, and thought it had an advantage. After a long shuffle the position opened up and it was Fizbo's passers against Raptor's rook, Raptor thought it was better. After another long shuffle with 5 pawns for a rook Raptor was able to exchange queens and get a winning endgame.

Houdini - Nirvana: Houdini thought it had a better RN vs RB ending with a pawn advantage, but Nirvana turned it into a drawn rook ending. Houdini down to 3rd place.

Round 11:
Nirvana - Laser: Nirvana thought it was ahead against Laser in a N vs B ending. evals climbed to over 1.5 before coming back down when not enough pawns were left.

Delphil - Stockfish: Only a draw !! After many early exchanges the position was a BB vs BN ending at move 27. Delphil was able to hold, quite a surprise given this was only its second draw of the stage having lost the other 9 games.

Fire - Naum: Naum seemed to be holding a RB vs RB ending, but Fire found a way to advance its two passers to get a winning position.

Rybka - Komodo: Komodo equalized easily but could not get more than a draw. It is getting far behind the table lead.

Round 12:
Laser - Komodo: A 3-fold repetition after 31 moves, great result for Laser, not so great for Komodo. It doesn't seem like it can close the gap to the top.

Stockfish - Fire: Fire played the opening badly and Stockfish took control of the game from move 10. The black king stayed in the center and was under constant attack. With Fire losing material the game ended on move 36. Fire's first loss, and Stockfish is in the lead.

Round 13:
Fire  - Houdini: Houdini defeated Fire convincingly playing black. First it had two advanced passers in front of an exposed king, and then Houdini converted this advantage to a winning RN vs RR endgame. Fire with two losses in a row, Houdini and Stockfish leading together.

Komodo - Ginkgo: Komodo showed its endgame strength beating Ginkgo in a RN vs RN endgame, aided by Ginko's no-tablebase handicap. Ginkgo's first loss in the rapids.

Round 14:
Bobcat - Critter: Bobcat surprised Critter in a B vs R ending. Critter blundered and allowed the white king to get behind the unguarded black pawns, creating enough passers for a win.

Nirvana - Fire: Nirvana had an eval advantage out of the opening, Fire was able to equalize and then get more space in a closed position. Playing with no time on the clock Fire opened the position and gained eval, but with pawns running out the game reached a drawn ending.

Round 15:
The Baron - Nirvana: The Baron found a 3-fold repetition to stop Nirvana's pawn from queening. A rare score against a top contender for the Baron.

Gull - Houdini: Houdini in black had little trouble in this game, patiently preparing in a closed position and then opening the position to a RR vs RR ending with connected passers.

Andscacs - Stockfish: Stockfish couldn't find a way to get an advantage in the game, Andscacs finished with a rook sacrifice and 3-fold repetition. Stockfish is 0.5 points behind Houdini.

Round 16:
Stockfish - Chiron: This was a long game in a closed position. Stockfish thought it had an advantage with the opening of the board in a RN vs RB ending, but with time running out it only got a drawn rook endgame.

Houdini - Andscacs: Houdini's eval started to climb on move 28. It combined a passed pawn with an attack on the king, enough to guarantee a win. This puts Houdini 1 point ahead of Stockfish.

Nirvana - Gull: Nirvana chose a RPP for BB imbalance and then was outplayed by Gull. The bishop pair was too strong and Nirvana gave a rook for a bishop to stop it. The extra piece won the game for Gull.

Thursday, September 15, 2016

Season 9, rapid tournament, rounds 1-8

Rapid tournament, not my cup of tea. The games fly by and there is no way to keep up with all the action. There will be one double RR, hardly enough to get a clear picture of which engines are the best at this time control. Cato is providing 2-move book openings, some of which are very biased, so luck will be an important factor. All 32 engines are participating, and many games are really one-sided because of the big difference in strength. The only good thing is that even boring games with endless shuffling are over after an hour and a half.

I may not like this tournament but people are watching and commenting in the chat, not a lot of complaints. TCEC is for chess entertainment, so have fun !

Table after 8 rounds


Fire leads with 8 wins, Stockfish and Houdini right behind with 7 wins and one draw in the game they faced each other. Komodo is 2 points behind the leader, still unbeaten. Ginkgo and Nirvana are the only additional engines with no loss so far. Of the 3000+ engines Naum and Critter are a little behind at this stage with a less than average score. DisasterArea, Laser and The Baron are performing better than expected so far for 2800- engines. It should be remembered that the table means little at this stage since each engine has only met 8 opponents of 31. Comparing engine performance is unreliable when each engine faced a different set of opponents. The table will get clearer as the tournament progresses. There's also the white/black bias and opening bias, so the best results will only be available when the reverse games have been played, but who wants to wait that long to say something about engine performance...

Games that caught my eye

Round 1:
Bobcat - Laser: Laser surprised with black. After 40 moves in a closed position Laser sacrificed a knight to open the queen side and get a QN vs RR imbalance. The rook pair was very strong and Bobcat had to exchange the queen for one of the rooks. Laser played the BN vs RB ending well, getting the win.

Texel - Raptor: In a static RRB vs RRN position and with no time on the clock Texel was tempted to take two pawns. Raptor quickly took over, using the exposed white king to gain material that made up for the lost pawns.

Round 2:
Laser - Raptor: Laser outplayed Raptor in what appeared to be a drawn rook ending. It appears Laser's rank should be higher in rapid games after two wins against better ranked engines.

Protector - Jonny: Jonny converted a pair of connected passers to a winning position. Two losses in a row for Protector, it is underperforming in the rapid games.

Round 3:
Andscacs - Ginkgo: With both engines in time trouble Andscacs was late to realize it was losing a piece at the end of a series of exchanges. The game looked balanced in the first 40 moves.

Critter - Rybka: Critter played a closed position two pawns down, but couldn't stop Rybka when the position opened after exchanging pieces. I expected Critter to be a tougher opponent.

Texel - Protector: Texel beat Protector in a RB vs RB ending, Protector's bishop was trapped in the corner. Three losses in a row for Protector, statistical noise?

Round 4:
Rybka - Jonny: Jonny outplayed Rybka with black, slowly gaining space and eval and then simplifying to a better endgame. Rybka is getting too old for this time control.

Fizbo - Gull: Around move 30 Fizbo's eval started to increase. It trapped the black king and queen in the corner and then won the game with a central passer. An unexpected loss for Gull against a lower ranked oppponent.

Round 5:
Critter - Ginkgo: The game was very close to a 50-move draw, but when the position opened it was Ginkgo with two pawns advantage. When there were almost no pieces left on the board the pawns got Ginkgo the win. Critter continues to struggle.

Jonny - Komodo: Komodo did not have a good start in this tournament, but in this game playing black against a strong opponent Komodo got a win. It is still unbeaten and not to be ignored, even though it didn't reach the superfinal and it is 2 points behind the table lead. The game also featured the rare "pawn square"



Round 6:
Laser - Protector: Laser was two pawns up but ignored Protector's connected passers. Protector won a rook in exchange for these pawns and won easily. The statistical noise for these engines is starting to average out.

Naum - Stockfish: Naum was a pawn up with eval advantage, but couldn't hold against Stockfish. Stockfish attacked white's exposed king, got the pawn back and captured more pawns until it had a decisive lead.

Round 7:
Houdini - Naum: Houdini used a strong king side attack to gain two pawns and get an winning endgame. Naum is not strong enough for the top engines.

Round 8:
Stockfish - Houdini: Both engines had 7 straight wins before this game, and these are also the two superfinalists.  However instead of an exciting decisive game we got a boring draw. All the tension was released in a massive piece exchange that left a drawn bishop ending. Fire is the real winner, now in the lead with 8 wins. 

Friday, September 9, 2016

Season 9, stage 3 stats

Draw rate, wins


Final draw rate was 65.6%. There was a preference for white in the decisive games, 4 times more than black wins.

Game termination

The three most common game termination causes were:
39.3% - TCEC draw rule
30.8% - TCEC win rule
19.2% - TB position

There was one crash by Jonny, a bit disappointing for such a late stage. This was mentioned as a possibility in the stage 2 statistics report. At least it didn't affect the qualification outcome.

Moves per game


Median= 62.75
Average= 71.4

The distribution is similar to the one in stage 2.

Time per game (hours)

Median= 4:57
Average= 4:54

Games were longer than in stage 2 by about an hour, due to the longer time control - 30 minutes more per engine. I'm not sure what causes the double peak.

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:

Open games (C) were under-represented, and there were many flank openings (A), mainly Dutch and English, as well as Sicillians (B).

If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
Sicillian - 51 times
English - 22 times
Dutch - 20 times

The 8 predetermined moves usually made sure that the reverse pair games had the same opening definition. There were 6 cases where the full opening name was different, of these 5 had different ECO codes.

Reverse pairs, wins


Only 4/112 openings are suspect of a bias - each engine wins once, while 53/112 openings have a win+draw combination - a proof of engine strength, winning from one side and holding the other side. Cato's opening choices manage to keep the draw rate down with opening bias while letting strong engines find defenses the weaker engines struggle with.

Reverse pairs, same moves 

The reverse games diverged quickly but less so than in stage 2. About a half of the game pairs had the same first two plys. One game had the same move sequence for 25 plys (Houdini - Komodo, 1/2-1/2).

Season 9, stage 3, RR 8

Final stage 3 table


Stockfish and Houdini qualify for the superfinal !!

Stockfish is still undefeated this season, as expected it won stage 3. Is this going to be Stockfish's season? So far it is in a class of its own.

Komodo could not close the gap in the last RR, it needed something extraordinary and one win against a bottom-4 engine just wasn't enough. The turning point was Houdini's win against Fire, after 7 drawn games between them. Looking at their performance Houdini and Komodo were very close, Houdini's single win over Komodo was the deciding result - compensating for the extra loss to Stockfish.

Houdini made a great jump in strength with this new version. Can it improve again for the superfinal? Will the new Stockfish release be even better? After the rapid tournament the superfinal begins, and we will see the final battle of season 9.


Some of the game highlights

Komodo faced Houdini in the first game of the RR, hoping to get a win and show Houdini fans that the opening bias caused Komodo to lose in the reverse game (Cato's fault?). The evals were in favor of white like before, but in this game the pieces were exchanged much faster. After 31 moves the game reached a rook ending, Komodo was a pawn up but it wasn't enough. With 3 pawns vs 2 on the queen side, the game was a theoretical draw. Does this mean Houdini is better? Which engine will reach the superfinal? Still 0.5 points between them and 6 games each to go.

Komodo and Houdini drew their games both playing black (against Fire and Andscacs, resp.), no change between them, now only 5 games to go. Andscacs managed to get a rook ending with two pawns against Houdini, but the pawns were f+h, a known 6-man draw.

Gull - Jonny: Jonny tried a knight sacrifice for two pawns early in the game, but did not have enough compensation. Gull was able to exchange pieces to a RRN vs RR position, and then go after the black pawns. The extra knight helped Gull to create a passer and win the game. A rare win for Gull in this stage, third win against Jonny though.

Stockfish had an easy win against Rybka. Rybka took a pawn and left its king almost defenseless, allowing Stockfish to develop a deadly attack. Rybka started to lose material and was a rook down with mate in sight when the game ended.

Komodo could not crack Andscacs' defense, only a draw against one of the bottom-4 engines. Qualification chances don't look good for Komodo.

Houdini and its fans had hopes of a win against Stockfish, the opening was quite biased and Stockfish won the reverse game. In the game Stockfish's defense proved to be very effective. Unlike the reverse game Stockfish playing black kept most of its pieces. The evals dropped to 0 by move 30 and stayed there until the end. Four games to go, still 0.5 points to Komodo, Houdini is getting closer.

Jonny - Fire was headed to a draw with evals at 0 when Jonny crashed. A bit disappointing that engines are crashing at this late stage of the season. Lucky that it happened in a game that has no impact on the qualification race. A crash in a game against Houdini or Komodo would have created a huge debate (fair or not, replay or not).

Komodo defended well playing black against Jonny, evals came down to 0 rather quickly and the game ended in a draw. Houdini played black against Gull, and for a while it seemed that Houdini had a chance with evals turning negative. Gull had a poor endgame performance this stage and there was a real chance it could blunder. Twice the evals got near 0.5, but in the end Gull held the draw. Three games to go, things are looking good for Houdini, Komodo is getting desperate.

Stockfish displayed its strength against Andscacs with a strong king side attack from move 20. Evals constantly increased as Andscacs tried to move pieces from the queen side. With the black king exposed Stockfish started to gain material, and the game was stopped before it was a full rook up. With a lead of 4 points Stockfish is sure to win the stage.

Komodo played against Stockfish with its back to the wall, but couldn't get any advantage in the game. Stockfish's eval was 0 after 15 moves and Komodo's eval stayed below 0.4 the whole game. Komodo is behind Houdini in SB score, it needs at least 1 point more than Houdini but it isn't winning at all so far this RR.

Houdini - Fire: see below. Houdini beat Fire and is now 1 point ahead of Komodo with two games remaining.

Rybka had a rook for knight advantage against Komodo, but Komodo was comfortable with eval at 0. When Rybka grabbed a pawn its rook was too far from the action and Komodo could attack the exposed white king, the evals started to turn negative. Rybka gave the material back but by then Komodo's connected passers were too strong in a bishop ending. Komodo's win in black gives its fans a final ray of hope, but they still need a Houdini loss.

Jonny had an eval advantage with a queen side attack against Houdini, but after a series of exchanges the evals came down and the game reached a drawn QN vs QB position. Houdini's 0.5 point lead should be enough for qualification with one game to go.

Komodo had nothing in the game against Gull, the game ended in a draw. This was Komodo's last game of the stage, it is tied with Houdini but loses the tiebreak. Houdini qualifies!! The remaning gmaes won't change this fact.

Stockfish had an eval advantage against Fire, right after Fire's first move. Fire's king stayed in the center and had to escape Stockfish's attacking pieces. Stockfish won a piece and reached a winning RN vs R position.

Houdini-Rybka was the last game of the stage. Rybka let Houdini develop a very strong king side attack so fast that the game ended before move 40 with Rybka sacrificing a queen and with both PVs showing mate soon. Now it is official, Houdini will meet Stockfish in the superfinal.

Featured game: Houdini - Fire
Stage 3, round 54
Link to game on TCEC

There were 7 draws between Houdini and Fire so far in the stage, the only engine pair with no decisive games. Houdini started with an eval advantage which grew to about 1 after 20 moves. Fire's eval was below 0.5 and it wasn't clear which engine was right, but there was a clear trend of gradual increase. At move 23 the two engine PVs agreed on the next 15 plys, unusual when the evals are diverging.

As the sequence of moves played out the evals started to climb again. Many pieces were exchanged and the position was simplified to RRB vs RRB. Houdini was a pawn up and had a doubled passer while Fire had a pawn advantage on the queen side.

The eval trend was unmistakable at this stage. Houdini wasn't winning yet but it didn't seem that Fire could hold. After more exchanges Houdini's advantage was clear. Fire's passer was weaker and slower, Houdini's rook made sure the black king cannot help, and the white passers looked dangerous. 


It took a few more moves for Houdini to get the win. This was the only decisive game for this engine pair after 7 draws, and with this win Houdini almost surely qualified to the superfinal. A 1 point lead over Komodo with two games to go against bottom-4 engines, as well as a tiebreak score advantage, should be impossible to beat for Komodo.

Friday, September 2, 2016

Season 9, stage 3, RR 7

Rapid tournament announcement - Anton (TCEC director from chessdom) announced that after stage 3, instead of the superfinal, there will be a rapid chess tournament. All 32 engines that participated in the season will play a double RR, with 25 min + 10 sec time control. The superfinal will start after the rapid tournament, somewhere in October.

I must say I was a little disappointed. Who asked for this change? Anton claimed popular demand, not sure who these people are. I know some people complain about the long TC, boring games etc. I'm sure others will now complain about the short TC. 992 games, 20 games a day is too much. That's more games than the regular season. And why now and not after the superfinal? Waiting for better hardware? Someone suggested that this will allow Houdini to get a better version - if it qualifies to the superfinal.

In any case, I will not be following this event closely. It's impossible for me to cover that many games. I may post a statistical report after the event is over.

Now, back to normal TCEC. 

Table after 7 RR


Another solid RR for Stockfish with 3 wins, one of the wins against Houdini. Stockfish is still undefeated not only in the stage but for the whole season, impressive.  Stockfish will be in the superfinal, with a 3.5 point lead over Komodo in third place and 7 games to go.
Houdini started with a win against Komodo, boosting its qualification chances. This was later offset by the loss to Stockfish, but still Houdini is in second place with a lead of 0.5 points. Komodo clawed its way back, but it still needs a good last RR. A lot depends on the Komodo - Houdini reverse game, the first game of RR 8.

Some of the game highlights

Houdini - Komodo: What a start to the RR, after 6 draws and only two direct matches left between the engines. The evals were high out of the opening, good for Houdini, trouble for Komodo. There was not a lot of change in evals for 50 moves, Houdini kept the pressure, Komodo defending but looking good. Time was running out for both, especially for Houdini. And then without warning, with less than two minutes remaining, Houdini's eval took off. The strange thing was that Komodo saw its problem at once (why did it allow the position or evaluated it so low before?). With no time to think Houdini was able to reach a queen endgame with a pawn advanage, create a passer and find a win. Amazing. With this win Houdini is 1.5 points ahead of Komodo, its qualification chances much higher. Can Komodo beat Houdini in the rematch?

Stockfish had control of a long diagonal and used it to drive Jonny's king out of the corner to the center of the board. Then through a combination of mating threats, pins and a pawn on the 7th rank Stockfish gradually improved its position in a QRB vs QRB position. Jonny was a pawn up but was too busy defending to advance the queen side pawns. The game was over just as Stockfish was about to win a piece after queening a pawn. A good start for Stockfish in the RR, keeping its 2.5 points lead.

Jonny started with an eval advantage against Gull, but Jonny tends to exaggerate and can't be trusted. At some point Jonny trapped a black rook on the queen side and started a pawn storm on the king side. After a series of exchanges Jonny's eval was at 3 in a QBN vs QRN position two pawns up, with a paralyzed black rook. After queen exchange the white pawns took over, Gull lost a piece and the game was over.

Houdini didn't get a lot out of the opening against Andscacs, but Andscacs made two mistakes: it allowed Houdini to create a passer and it sent its queen to chase two poisoned pawns on the queen side. By the time the queen came back the white passer was almost unstoppable, and evals were close to 4 as it reached the 7th rank. Andscacs gave a rook for the pawn and Houdini won the game easily. Houdini now 2 points ahead of Komodo, fourth win in a row for Houdini.

Stockfish - Houdini: Stockfish won !! (see featured game below). Houdini 1.5 points ahead of Komodo.

Komodo's evals stayed at around 0.5 for 30 moves against Jonny. Just as Komodo's bishop pair started to take hold of the center Jonny chose to give a rook to take out one bishop. Komodo welcomed this exchange, and after more exchanges reached a R vs N ending which was easy to win. First win for Komodo this RR, it need more to have hope of closing the gap to Houdini.

Houdini started with an eval advantage against Gull, and Gull held out only 20 moves before the evals started to rise. Most of Gull's pieces were ineffective on the queen side, while the black king was in danger on the other side. Through a clever series of exchanges Houdini created a RPPP vs BN imbalance, opening up the king side and exposing the black king. This led to a RB vs RN position with Houdini 3 pawns up, it took a while but the win was clear. Gull added a little joke, evaluating the final tablebase win at 0.61. The gap to Komodo is still 1.5 points, Houdini matching every Komodo win.

Stockfish, playing black against Andscacs, was able to take the initiative. It was not enough for a win though, but Stockfish can afford to draw as black. It has a 4 points cushion to Komodo in third, and only 10 games remaining.

Stockfish - Komodo was a draw. Komodo needs wins but it probably won't win black against Stockfish. It has some hope for the reverse match.

Gull surprised Andscacs (and most everyone else) with a win. It started with a modest advantage and gradually increased it. Gull gave a rook for a knight, and later this turned into a BN vs R imbalance. After queens were exchanged the white bishop pair took over the center, and a passed pawn cost Andscacs a piece. Andscacs still had 3 extra pawns in a RBB vs RB position, but after all pawns were gone Gull did not blunder the ending for a change.

Fire - Houdini was another draw, seventh draw for these engines in this stage. Houdini had no problems in black against a tough opponent. No change in the lead over Komodo.

Komodo - Rybka: after 20 moves Komodo sacrificed a pawn to get a deadly king side attack. As a result Komodo won a piece and reached a winning RN vs R position. Komodo putting some pressure on Houdini.

Houdini seemed to be on a sure path to victory against Jonny, keeping an eval advantage of 0.8 for 30 moves. A first signal that Houdini was confused was the dance the white knights performed on the queen side, moving forward and coming back with no purpose. After a series of exchanged of pieces and pawns the evals started to come down. In the end the game reached a drawn rook ending. The gap to Komodo is now 1 point with one game per engine left in this RR, both with black. Is Komodo coming back?

Stockfish took over the game against Gull with evals rising after 20 moves. There was nothing spectacular, just a better position that led to a rook ending with Stockfish a pawn up and pawns on both sides of the board. Stockfish improved to 3 pawns advantage and then it was just a formality to get the win.

Gull thought for over 51 minutes on its second move after book against Komodo. That was quite unusual, and also did not change the eval, what was Gull thinking so much about? Then on move 18 the experts on the chat claimed that Gull missed a move (18. Ng5), an attack that would have given it a big advantage if played correctly. Komodo also missed this, was it real or not? Anyhow, as the game continued to a RN vs RB ending Komodo took the initiative. Gull's endgame abilities were no match for Komodo, after rooks were exchanged Komodo grabbed two pawns and got the win. Komodo finishes the RR with a win in black, can Houdini do the same?

Rybka was a pawn up against Houdini, and it didn't let go of the advantage. Houdini easily got a draw but no more than that. The game continued for a long time since Rybka misevaluated a drawn 6-man position, thinking it had a clear advantage. The gap between Houdini and Komodo is back down to 0.5 points, same as the gap after the previous RR.   After all the excitement we've had the race is still the same, only one RR to go.


Featured game: Stockfish - Houdini
Stage 3, round 45
Link to game on TCEC


Stockfish had an eval advantage of about 0.8 out of the opening, quite a lot. By move 18 It was a pawn up, but there were so many exchanges that only RRB vs RRB remained. The position looked very drawish, and with opposite color bishops as well.


For a very long time the engines developed their pawns and tried to find the best squares for their pieces. The evals did not change a lot, a little bit higher than before.


Stockfish attacked on the king side and was able to capture a second pawn and get a passer. The price was exchanging a pair of rooks. A RB vs RB position with opposite color bishops could be a draw after another rook exchange, even with two pawns advantage. Could Stockfish pull this off?


The target was the pawn on b7, but it was defended by a bishop and could only be attacked by the rook. Houdini used its rook to threaten the king side pawns so that the white king would stay and protect them. Stockfish then found a way (Houdini blunder?) to get the king involved, and the evals shot up.


With the king going to b6 Stockfish could give a rook for a bishop as well as two pawns to take out the black pawns, that would be a winning BPP vs R position. Houdini had nothing to do against this, and the game was over in a few moves. Stockfish and Komodo fans were delighted, although Houdini still has better chances in the race for the superfinal.