Tuesday, December 26, 2017

Season 10, rapid tournament statistics

Draw rate, wins



Final draw rate was 38%. Looking at the subtable of only the top 10 engines the draw rate was 53.3%.

Game termination

The three most common game termination causes were:
57.8% - TCEC win rule
15.9% - TCEC draw rule
12.0% - TB position

There were only 11 crashes.

Moves per game



Median= 57
Average= 62.9

The distribution has a tail of long games. There were 47 games (8.5%) with more than 100 moves, the longest was 318 moves long, Fire - Booot.

Time per game (hours)
  

Median= 0:45
Average= 0:46

The time distribution is almost symmetric, when playing many moves the engines played faster.

Openings

There were 2-move book openings in this stage chosen by Cato. The first letter of the ECO codes was distributed as follows:



Open games (C) and indian openings (E) were under-represented, very similar to the rapid stage in season 9. Is this a Cato preference or do the engines prefer to transpose to other openings? 

If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
Sicillian - 86 times
English - 63 times
QGD - 48 times

The engines had a lot of freedom to choose the opening variant. Only 38% of the game pairs repeated the same ECO code twice, and only 20.7% repeated the same opening variant. In 18.5% of the game pairs the ECO first letter was not repeated twice.

Reverse pairs, wins



Reverse pairs, same moves 



The reverse games diverged very quickly, 43% diverged in the first move and 78% diverged at or before the third ply. Only 12 games repeated more than 6 plys.

Monday, December 11, 2017

Season 10 LTC is over

Season 10 is over, at least the games with long time control. There is a rapid tournament playing now, and it will be followed by a blitz tournament. I may watch when I have time but I won't follow these games closely or post reports. There are just too many games and I need a break.

Season 10 was the first season under chessdom that felt well managed. The games ran smoothly, there were very few server breakdowns and those were due to the larger number of viewers. A replacement for Martin as director was found. Anton was able to persuade more websites to follow the games, and he also posted interesting interviews with engine creators. Martin visited the chat from time to time and I think he can be proud of this season, definitely after the problems of the last one.

This season the old Chatwing chat was replaced by a twitch channel and chat. The transition went smoothly and the new chat became popular. One of the consequences is that there is a lot less trolling, especially by him who shall not be named. The twitch channel also allows a replay of a game including the chat in real time, acting as a chat archive which was impossible before.

In stage one most of the attention was focused on the qualification race. In the final rounds there were 5 contenders for four spots, in the last round it was determined that Gull was the one left out. Stage 2 was a very close race between the top three engines, in the end it was Houdini and Komodo who went on to the superfinal. Stockfish was 0.5 point down with worse results against the lower ranked engines, not qualifying despite not losing a game and beating Komodo once. The superfinal initially looked like a one-sided match, Houdini with 5 wins in the first 24 games. From that point the match became much more balanced, Houdini won the match but the remaining games had a score of 10-9. With this win Houdini completes a comeback which started mid season 9: version 4 was dropping out of the top 3, version 5 reached the season 9 superfinal, and version 6 wins in season 10.

Looking back at the season there were three main issues discussed in the chat that come to mind. In stage 2 it was the 'c7f7' event. Andscacs, a well respected engine, did not perform well in stage 2. Its worse moment was when it threw away a game by making an illegal move (c7 goes to f7). In the beginning of the superfinal it was the suspected problem with Komodo. It was noticed that Komodo's node per second rate was lower than expected, and the Komodo team found out that the cause was a problem in the compilation (this was later given the name 'compiler-gate'). The TCEC decision was to not allow a mid-stage fix, submitting new versions can be risky. Closer to the end of the superfinal Google dropped the AlphaZero bomb. With its huge resources and a learning framework Google was able to give a program the rules of chess, and with only self play it managed to reach a very high level of play, beating Stockfish 8 in a 100 game match 28-0. There was a lot of discussion, mainly defensive arguments (match was not fair, can't really tell which engine is "better"). There is no question that this is a new approach to computer chess, and it would be exciting to see how it will affect the future of the field.

Season 11 will start soon, perhaps right after the blitz tournament. The format is going to change to 4 leagues of 8 engines each, ordered by engine strength. The league system will include relegations and promotions of two engines from league to league. In this system there will be less games between engines with extreme strength difference. The lower leagues will be played with shorter time controls, and the higher leagues will have more rounds. The top two engines in the top league will play a superfinal match as before.

This season I had a few live blogging sessions, where I could react to the game as it happens and not while looking at it in the archive. This was a lot of fun though time consuming, I will try to have more of these sessions in the future. Other blog news:
  • With the growing popularity of TCEC I had many more hits than the previous two seasons. There were over 3000 hits in November, almost 3 times the previous record for a month.
  • All-time number of hits above 15000.
  • The blog was mentioned and quoted in a post on chessdom
  • The blog continues to be read by people from all over the world, I'm very happy about that.
That's all folks! I will be back for next season, till then bye, and
GO TCEC!

Friday, December 8, 2017

Season 10 superfinal, games 91-100

Houdini won the season 10 superfinal !!
The final result is 15-9 and 76 draws.

Game 91 started with an eval advantage for Komodo. Between moves 15-31 most of the pieces were exchanged and the game reached a double rook ending. Komodo was a pawn up, but after Houdini forced a rook exchange the extra pawn was only enough for a draw. It took another 40 moves to adjudicate the game. In game 92 there were less exchanges after the opening and Houdini kept its eval of about 0.4 longer. There were no open files until move 40, and then all rooks were exchanged and Houdini closed the game with a check repetition.

Houdini started game 93 with negative evals and a pawn down, though after a few moves the evals became positive. The kings castled in opposite directions, Komodo pushed pawns on the king side and Houdini placed its king behind a white pawn, giving a second pawn. For a while Houdini had the initiative and an attack on the white king. Komodo defended well and its major pieces took over the central files. In a series of exchanges Houdini then gave a bishop for the last white pawns. Komodo had to defend its exposed king against checks from the black queen, and eventually queens were exchanged. The engines agreed to a draw in a RBN vs RN position. In game 94 there were no pawn storm on the king side. Houdini gave back the pawn and opened the king side while Komodo countered on the queen side. Both kings faced the major pieces of their opponent, Houdini attacked first and drove the black king to the queen side through checks. Komodo exchanged queens to get out of trouble, and the game reached a drawn rook ending.

Game 95 was a king's gambit and Houdini had an eval advantage. The engines castled in opposite directions, Komodo did not have enougn pawns on the king side to threaten the black king and the focus was on the queen side. Houdini kept an eval of about 0.6 for a long time, a file opened on the queen side and Komodo's king walked forward out of the corner and found safety. When queens were exchanged the evals dropped to 0 and the game reached a drawn rook ending. In game 96 Houdini opened the black king side and exposed the king. Komodo had an advanced passer which had to be blocked. After a long series of exchanges only BN vs BN remained and Houdini captured the passer. After a few pawn moves the draw rule stopped the game. Two king's gambits and two draws, somewhat surprising.

Game 97 started with an eval advantage for Komodo. The center was blocked, most of Houdini's pieces were on the king side with little room to move, while Komodo had more space on the queen side. Houdini managed to push a pawn to g3, further blocking the king side, its eval turning negative. Komodo still thought it had the edge.



Komodo chose to give a knight for two pawns and open the center. There was a long series of exchanges, most of the minor pieces were exchanged and Komodo had three passers in the center. Houdini had an extra knight and evals were in its favor though still less than 1.



Komodo's strong point in the position were its passers. Houdini blocked two with its queen and knight, and Komodo pushed the third to the 7th rank where it too was blocked. Komodo couldn't get any further. 



Houdini had several easy targets, the pawns on d7, a5 and g2, as well as a back rank mate threat. The a5 pawn was first to go, and after exchanging a pair of rooks Komodo agreed its position was hopeless.

In game 98 a few exchanges released the block in the center. Both kings were slightly exposed but after most pieces were exchanged and only RN vs RN remained both kings were safe. Evals were close to 0 and a draw was just a matter of time.

In game 99 a series of exchanges left only RRB vs RRB on move 19. Evals stayed low as the engines captured pawns and exchanged a pair of rooks. Houdini gave the bishop for the remaining white pawns leading to a drawn position. In game 100 more pieces stayed on the board. The engines castled in opposite directions, Houdini advanced pawns on the king side and Komodo concentrated forces on the queen side. On move 28 only QRB vs QRB remained, Houdini gave two pawns to advance a passer but evals stayed at zero. Komodo was able to stop and capture the passer but gave the material back. After exchanging queens the game reached a draw.

Season 10, superfinal stats

Draw rate, wins



Final draw rate was 76%, lower than what Cato predicted though close to the draw rate of the last superfinal. There were 4 black wins, two for each engine.

Game termination

The three most common game termination causes were:
45% - TCEC draw rule
24% - TCEC win rule
19% - TB position

Moves per game




Median=62
Average=70.7

There were 10 100+ move games, two of these were decisive.
 
Time per game (hours)



Median= 4:04
Average=4:05

The short increment made sure the game duration would not far exceed 4 hours.

Openings

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


A small bias for Sicillian openings and open games, probably close to human game opening distribution.

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

Sicillian - 14 times
King's Indian - 10 times
French - 9 times

Some of the book lines were short enough to allow different ECO codes in reverse games, this happened in 3/50 paired games including the first pair which had no book at all.

Reverse pairs, wins


There were 3 biased openings with two white wins, a side effect of the generally low draw rate.


Reverse pairs, same moves  



Reverse games diverged very quickly, 38% diverged immediately after book, 70% diverged at most after two plys. One opening had 22 plys repeated after book, games 43-44 Caro-Kann, Advance opening.
 

Thursday, December 7, 2017

Season 10 superfinal games 81-90

After 90 games Houdini is in the lead, 14-9 and 65 draws. Komodo reduced Houdini's lead slightly to 5 games. With 10 games left it is safe to say Houdini will win (it is very unlikely that an engine will win both sides of an opening in the match as the openings are all slightly biased). As someone said in the chat, the match would have been much more exciting if the game order was reversed, the score in games 25-90 is 9-9.

There has been a lot of heated discussion surrounding the recent Google chess AlphaZero paper. Google published a manuscript where they describe their program learning chess from the basic rules, and improving through self play and a general learning framework. They claim to have reached a level of play that can defeat Stockfish 8 (not the most recent version) after a few hours of training. In my opinion this achievement is incredible, regardless of the criticism expressed in the chat (not a fair competition because the hardware Google used for each program was very different, Stockfish was not configured optimally, a few hours on Google hardware is equivalent to years for Stockfish testers). I for one would love to see a similar contestant in TCEC.

In game 81 there were no pawn exchanges after the opening. Houdini pushed pawns on the king side and evals became slightly negative. Komodo's queen went on a risky journey, capturing a pawn and creating a passer on the queen side but giving a rook for a bishop. Houdini had to control the white passer, this allowed Komodo to exchange down to a drawn queen ending. There were many exchanges in game 82, evals got as high as 0.5 but then dropped to 0. The game reached an opposite color bishop ending, and after 35 moves the engines agreed to a draw.

Game 83 started with a small eval advantage for Komodo. All knights were exchanged early in the game. Komodo was a pawn up when it chose to give its queen for a rook and bishop.



Evals stayed low while Komodo pushed its e pawn to create a strong bishop and pawn pair on the 6-7 ranks. Komodo limited the possible squares available for Houdini's queen until it was almost trapped in the corner. Komodo's eval rose above 1, Houdini was still calm.



The board was full of traps, Komodo's pieces coordinated perfectly. Komodo's eval was close to 3 before Houdini realized it was in deep trouble. The black queen was out of the corner and even captured a pawn. In order to let the queen get back and help the defense Houdini weakened the king side pawns.



Now Houdini's eval jumped, Komodo closed in on the black king and Houdini allowed the white passer to advance. Houdini took the passer and exchanged a pair of rooks but was about to lose a piece when the game was adjudicated.

In game 84 there were many exchanges. After queens were exchanged evals dropped to 0 and stayed there for the rest of the game,  reaching a drawn rook ending.

Komodo had a small eval advantage in game 85 even though its king stayed uncastled in the center. All knights were off the board and with many pawns the remaining pieces had limited mobility. After queens were exchanged only RB vs RB remained and evals dropped to 0. Houdini gave two pawns to be able to place its rook on the 2nd rank, the game was adjudicated and the PV showed a series of exchanges and a drawn rook ending. In game 86 the center was open and both engines castled. Komodo gave a rook for a knight and got a pawn majority on the queen side. Evals dropped as the engines exchanged pieces. In a QR vs QB position and an open position the queens posed many threats, finally Komodo forced a draw by perpetual check.

Komodo started with an eval advantage in game 87, and evals gradualy increased. The center and king side were mostly blocked, both kings were apparently safe, Komodo had more space and a bishop pair that could explain the eval.



The engines shuffled for a while, evals continuing to increase. Komodo then pushed the c pawn, disrupting Houdini's pawn structure and threatening the weak pawn on e6. As a result Komodo was able to create a hole in the center and a passer, with evals over 2.5. 


The e passer was well supported and threatened a black rook. Houdini tried to use a pin so Komodo moved the queen out of the way. After a series of exchanges most pieces including queens were removed, Houdini gave a piece to get rid of the white passer.



The rest of the game was straightforward, Komodo exchanged rooks and the piece advantage was enough for a win.

In game 88 the center and king side were again mostly blocked. This time black castled on the king side and the queen side pawns for both engines had more freedom to move. The engines exchanged most pieces through the open queen side until only RN vs RN were left. 



Komodo had a queen side passer and material was equal, yet Houdini had an eval advantage of 0.7 and gradually increasing. Komodo had no counterplay and had to defend its weak pawns, in particular the backward e pawn. Houdini's knight replaced the rook in blocking the a passer, and this allowed the rook to get behind the black pawn line.



Komodo's king and knight could not move for fear of losing the e pawn. Houdini's king started the long walk around to the queen side. Once it was covering the a pawn the knight joined the attack and it was game over for Komodo. The black pawn line crumbled and a few moves later the game reached a won rook ending.

Komodo started a pawn up with a queen side passer in game 89 (Benko gambit). A long series of exchanges left only BB vs BN with equal pawns on move 26. Both engines' evals were close to 0, the bishop pair was not sufficient to push the white passer forward, game ended in a quick draw. Houdini kept the pieces on the board in game 90, with evals increasing to about 0.8. Komodo used its major pieces to defend the queen side, and with the rest of the board static the engines shuffled for a while with evals close to 0. The engines pushed a few pawns forward, opening files and exchanging pieces, as a result the queen side threat was removed and the game reached a tablebase draw.

Tuesday, December 5, 2017

Season 10 superfinal, games 71-80

After 80 games Houdini is in the lead, 13-7 and 60 draws. Houdini remains 6 games in front with only 20 games left. There were two more wins in black, one for each engine, and both in openings that started with an advantage for white.

In game 71 Houdini gave a knight for pawns, cleared the queen side from white pawns and supported an advanced passer. Komodo blocked the passer and exchanged pieces whenever it could. After exchanging queens only BN vs B were left and evals at 0, the extra piece enough to handle the passers and a draw. In game 72 there were almost no exchanges, Houdini's pawns advanced across the board gaining a lot of space. Houdini's eval was over 1 in a closed position and the engines started to shuffle. Komodo gave a rook for a bishop and created a wall, evals dropped to 0. After a few exchanges in the center only QRB vs QBN were left, Komodo had a fortress on the queen side for its king. Houdini's king was exposed and had to be defended by pieces. When all the king side pawns were gone there was a dynamic stand off that was played for another 60 moves until the engines agreed to a draw.

In game 73 the engines formed a long diagonal pawn line that blocked the center and the king side and almost divided the board into two separate halves. There was a small opening on the queen side for pieces to pass through. Houdini's bishop was completely blocked out, the only target for Komodo was the black b pawn.



Suddenly Komodo saw a plan and the evals jumped above 1. Houdini did not have enough defenders available and Komodo could take b5 easily. However, it took its time mainly to block any infiltration attempt and to keep an eye on the king side, a knight sacrifice could lead to a dangerous king side attack. Once the pawn was taken the black QR could move more freely and Komodo was very careful and took its time. Eventually it arranged all its pieces just right and pushed the b pawn. 



With the pawn wall breached Komodo created a connected pair of passers and exchanged queens. The game was soon adjudicated, the PV showed that Houdini was losing material to stop the passers. A deep plan and patience won the game for Komodo.

Komodo closed down the queen side in game 74, then its queen wandered through the white territory until it was exchanged. The remaining pieces could only shuffle, and a few exchanges later the game was adjudicated in a RNN vs RNN position.

Game 75 started with an eval advantage for Komodo, its queen out taking pawns on the queen side. Houdini took a rook for a bishop to compensate. After closing the center Houdini could only defend, but Komodo could not push its queen side pawns and the engines started to shuffle. After 3 rounds of resetting the 50-move counter at the last minute the engines finally allowed a draw, the longest game of the match so far. 

Game 76 started in a much more relaxed manner. Material was balanced and there were almost no exchanges, Houdini kept an eval advantage and had more space. The queen side looked stable and Houdini's pawns on the king side were advanced, its king slightly exposed.



Evals started to increase as Houdini's pawns advanced on the king side. Komodo blocked them without opening files, but it gave a rook for a knight in the process. In a QRR vs QRB position Houdini's king could not be attacked as long as white controlled the back ranks, while Komodo's king was vulnerable in the corner if white could penetrate.



Komodo tried to avoid exchanging queens, Houdini managed to force it for two pawns. With a pair of rooks also gone the remaining R vs B position was a win for Houdini, too many pawns were hanging and the black king could not get out of its corner.

There were very few exchanges after the start of game 77. The evals started in favor of Komodo but turned negative. Komodo pushed a passer through the center, Houdini advanced pawns on the king side.



Through an exchange Houdini created a passer on the queen side, and it also went a pawn up after capturing Komodo's unsupported passer. Evals continued to climb, approaching 1.5 for Houdini.



After a series of exchanges Houdini's a passer was unstoppable. Komodo chose not to stop it, instead it continued to exchange and capture pieces. The result was an unusual QQ vs QB position with Komodo 2 pawns up.



Komodo captured another pawn but its position was hopeless against two queens. This was Houdini's first win in black in the match.

In game 78 the center was closed and evals close to 0. The engines opened a file on the king side and through it exchanged pieces until only QB vs QN remained. Komodo gave the knight to stop a passer and forced a perpetual check draw.

Komodo had a small eval advantage in game 79. The center was blocked and action focused around the open file on the king side. The engines exchanged most pieces and reached a knight ending, Houdini blocked any chance Komodo had to gain a pawn. Komodo continued to try for another 35 moves before adjudication.

Game 80 started with a small eval advantage for Houdini. The center was blocked after the start but then Komodo unblocked it, compromising its king safety and losing one of its bishops in the process. The evals started to go negative, still close to 0.

Komodo now had a pawn majority on the queen side. Houdini was reluctant to open files on the queen side and tried not to exchange pawns there. As a result Komodo was able to advance its pawns until two became connected passers. Evals jumped over 1.



Houdini had to deal with the threat of both the passers and the strong central bishop. There was a series of many captures and exchanges, including a queen sacrifice and a pawn queening by Komodo. After (all) queens were exchanged only RB vs RR remained, evals approaching 2.



The endgame was not easy, Komodo trapped the white king and Houdini had to protect against mate threats. Eventually the game was adjudicated a win for Komodo, in the PV it forced a rook exchange and the remaining bishop could not defend the white pawn on the king side. A second win in black for Komodo.

Sunday, December 3, 2017

Season 10 superfinal, games 61-70

After 70 games Houdini is in the lead, 11-5 and 54 draws. Houdini is 6 wins ahead of Komodo with 30 games left. It was 5-0 after 24 games and Komodo's record since then is 6-5, not too bad for an engine with a known bug slowing it down.

Komodo had a small eval advantage in game 61 but no real advantage on the board. After closing the queen side Houdini attacked on the king side, advancing pawns supported by many pieces. Komodo cleared the queen side and countered, leading to a series of exchanges that left a drawn knight ending.

In game 62 Komodo exchanged many minor pieces after the start, Houdini's pawn structure had holes on both sides and its king was exposed. After exchanging queens the kings were not in danger and Houdini's more advanced pawns became an advantage. Houdini's eval was starting to increase.



Houdini exchanged the bishop for the knight and evals were jumping. The white king was very active, it marched forward and attacked the queen side pawns. Komodo tried to counter with its pawns on the king side, but it wasn't enough.



Houdini's rook found its way behind the black pawns, and the game was declared a win for white. Komodo's PV suggested that Houdini would be two pawns up soon in a winning endgame.

Komodo's queen came out early in game 63, and the queens were exchanged before too long. Evals were low though Komodo was two pawns up with an advanced passer in the center. Houdini created passers on the queen side and exchanged down to a RN vs RB position and forced a draw. In game 64 Komodo's queen came out early again, this time as black. Houdini's eval stayed above 0.5, it was a pawn up and it had more space on the queen side. In a long series of exchanges the evals jumped above 1 and only RRN vs RBB remained. Komodo found a defense, first exchanging a pair of rooks and removing the white queen side pawn, then exchanging a bishop for a knight. The R vs B ending was a 7-man draw that was played for another 25 moves.

Komodo gave a knight for pawns at the start of game 65. Evals were at 0.8 after queens were exchanged, Komodo took over the center and with a series of exchanges removed all the black queen side pawns. 



The white rooks took over the 7th rank and the queen side pawns began to march. Houdini tried to counter with its pawns on the king side, and by exchanging a pair of rooks. The evals continued to increase over 2.



Eventually Houdini was forced to give the knight to stop the a pawn. In the RB vs RB ending Komodo was two pawns up. The game was stopped before the win was clear, the PV showed that after exchanging bishops the rook ending was won for white.

In game 66 there were very few exchanges at the start. Houdini had a space advantage and evals that got up to 0.8, the position remained mostly closed and the engines started to shuffle. After more than 60 moves and exchanging all rooks evals dropped to 0, Houdini marched a passer forward and Komodo responded with a perpetual check draw.

There were many exchanges at the start of game 67. The engines castled in opposite directions and Komodo advanced pawns on the king side facing the black king. Evals increased to about 1 as the king side started to open up with all the white major pieces attacking. Houdini exchanged a pair of rooks and took over the center, in the QRN vs QRB position Komodo's attack was over. The engines shuflled for a while and closed the game with exchanges and a drawn rook ending.
In game 68 again the engines castled in opposite directions. Both engines advanced pawns towards their opponents' king, Houdini also lined up its major pieces on the g file, looking very dangerous for black.



Instead of opening files the pawns on both sides locked to create long diagonal lines. Houdini's evals remained constant for a long time, while Komodo's slowly increased. The engines shuffled and it was not clear how Houdini could improve its position. Then Komodo chose to exchange a rook for the white knight.



Komodo perhaps thought the knight can be strong in these closed positions.The exchange allowed Houdini to move its pieces to the queen side and open it up with a pawn push. Evals started to jump as most of the queen side pawns were removed and the rooks took over the open files.



Houdini's king was exposed but Komodo was too busy defending with the white rooks on the 7th rank. The threat of mate forced Komodo to lose material and the game was soon over.

In game 69 Houdini sacrificed a knight for pawns and got the initiative with very active pieces. Komodo had an eval advantage of up to 0.6, it exchanged pieces to stop Houdini's attack. After queens were gone only RN vs R remained, Houdini's pawns were strong enough to compensate for the knight, and the engines agreed to a draw. In game 70 the queens were exchanged early. Only RN vs RN remained by move 27 and evals were close to 0. The game ended in a drawn R vs R position.

Saturday, December 2, 2017

Season 10 superfinal, games 51-60

After 60 games Houdini is in the lead, 9-4 and 47 draws. There were 4 decisive games in the last 10. The first two by Komodo, giving the appearance that it was closing the gap. The next two by Houdini, and with 40 games left it will probably win the match.

Evals were low in game 51, most of the queen side pawns were exchanged and then the four knights filled it, two self-protecting pairs that could not be moved. Komodo appeared to attack the black king through the long diagonal, yet its eval remained 0. A series of 7 moves that included 11 captures cleaned the board leaving a drawn knight ending.
In game 52 Houdini advanced its king side pawns, closing the king side with very little room to move for black. Komodo won a pawn on the queen side while Houdini concentrated forces on the h file, it looked as if one wrong move by Komodo and its king would fall.


The engines shuffled for a while, Komodo defended the possible entry points and Houdini didn't find a safe way to break the defense. Evals came down gradually and even became slightly negative. Houdini was getting low on time. Komodo opened the queen side and exchanged pieces, Houdini managed to open the h file but did not have enough pieces left to use it. Komodo's extra pawn was evident in the evals, about 1 for black.

Evals gradually increased as Komodo slowly improved its position on the queen side.The bishop on g7 was enough to stop any counter by Houdini on the king side. It took Komodo 30 moves to force a queen exchange.

Both engines were in blitz mode. The endgame was far from trivial, Komodo pushed the d pawn forward until Houdini was forced to lose a bishop to stop it.



The extra piece was enough to guarantee a win for Komodo, the game was adjudicated without showing exactly how. A wild game, Houdini's plan ran into a wall and Komodo turned the game around, winning in black for the first time in the match. Houdini's time trouble probably helped as well.

Komodo had a small eval advantage in game 53. There were several pawn exchanges creating space and open files, Komodo was a pawn up and had a bishop pair. Evals increased to about 0.8, then Houdini exchanged pieces until reaching a rook ending a pawn down, securing a draw. In game 54 both engines did not castle, Houdini's king side pawns moved forward and Komodo infiltrated with two pieces. Following a rare king fork Houdini won a rook for a knight, leaving QRR vs QRB in a closed position. Evals were at 0.8 but Houdini could not break Komodo's defense. The position opened up a little and Komodo's advanced passer compensated for Houdini's material advantage. The black king was safe in the corner, Komodo harassed the exposed white king until a draw was agreed.

Game 55 opened with a pawn advantage for black, Komodo soon equalized through a series of exchanges giving a knight for pawns. Both kings were exposed, after queens were exchanged Komodo took over the 7th rank and forced Houdini to give the material back. The game reached a rook ending, gradually capturing pawns until reaching a draw. In game 56 the evals stayed negative for a long time. Komodo gave back the pawn and exchanged queens early, but had a pawn advantage on the king side and a castled king. When the engines closed the queen side and center they started to shuffle, both not wanting to make contact. Komodo moved a pawn to avoid the 50-move mark, then Houdini opened up the center. Both engines were low on time and the evals started to jump, pawns advancing on both sides and exchanging pieces. When the dust cleared only NN vs RN remained. Komodo's material advantage was not sufficient, a draw was agreed a few moves later.

Komodo's evals increased rapidly after the start of game 57, already at 1 on move 14. Houdini was a pawn up while Komodo had an advanced passer in the center and open diagonals looking directly at the black king. Houdini's queen side was not developed and unable to assist.



Komodo sacrificed a knight to open the king side, with evals jumping above 2. Then, starting from move 22 the engines played out a long PV they both agreed on. Many moves and exchanges later the material was balanced but Komodo was in a winning position, its passer almost unstoppable.



Houdini had to lose a piece for the passer, and the RN vs R ending was won for white. Komodo is closing the gap, now only 7-4.

Game 58 developed into a closed position and the engines started to shuffle early. After a few weird looking rook and king moves Houdini started to push pawns on the king side and the evals started to jump. When Komodo took a pawn in the center with its knight Houdini pinned it immediately and evals were over 2.


Komodo tried to exchange all the pieces it could, losing at first a knight for three pawns. The black king was still in danger and Komodo had to lose more material to escape.As a result it ended up a rook down for three pawns.



The black queen side pawns could not be protected and the game was over within a few moves. The opening was biased and Houdini strikes back immediately.

There were many exchanges after the start of game 59, Houdini was a pawn up but its king remained uncastled in the center. Komodo had the initiative but could not get any advantage. Evals stayed low and after exchanging queens only rooks remained, leading to a drawn tablebase position.
In game 60 both engines did not castle. Houdini had more space and managed to activate both rooks, Komodo's king side rook could not get out .



Evals stayed constant for a while as the engines moved their pieces behind the pawn lines. Then Houdini doubled rooks on the c file and closed the queen side, and evals jumped over 2. Komodo had to prevent the rooks from getting to the back rows, and to prevent black's knights from getting to f6. After a while it couldn't handle all the threats. 



Komodo's position crumbled and it immediately started to lose material. After the queens were gone Houdini put a rook on the 7th rank, the black queen side pawns could not be defended, another Houdini win.