Saturday, March 31, 2018

Season 11 superfinal, games 1-10

This season's superfinal is between Stockfish and Houdini.

After 10 games Stockfish leads 2-0 with 8 draws. 

In game 1 Stockfish had a pawn advantage from the opening, evals increasing over 1. After a few exchanges the evals dropped to 0 and Houdini regained the pawn. The engines traded down to a RB vs RN position and then a 6-man draw. In game 2 the evals stayed under 0.6. Stockfish gave a second pawn for a king side attack, then sacrificed a bishop to get a repetition draw in a line the engines saw 10 moves in advance.

After a few early exchanges in game 3 Stockfish's eval started to increase. Evals were over 1 when Stockfish gave a pawn and opened the king side, after exchanging queens there was no immediate threat to Houdini's king in a RRB vs RRB position. Stockfish got the pawn back but only got a drawn bishop ending. Game 4 reached a drawn rook ending on move 26, evals close to 0 from the start.

Game 5 started with an unusual Q vs BB and pawns imbalance. Evals were over 1, with such a biased opening the question was: can black hold the position?



Houdini tried to break through the center. Stockfish gave another pawn and its king was totally exposed facing a pair of black passers. On the other hand the passers formed a screen preventing black rook attacks on the white king. Houdini also had to keep an eye on the white passer on the 7th rank.



After the advanced passers were eliminated Stockfish gave the queen back and traded down to a RR vs RB position. Evals indicated a win for white.



The black king was trapped in the corner and the white king was very active in the center. Stockfish started to capture pawns and the game was adjudicated quickly. First win in the superfinal, but could Houdini win playing white?
In game 6 Houdini took the time to castle on the queen side and to find a safe spot for its king. The engines were essentially shuffling from move 20, Houdini had an eval advantage close to 1 but evals did not move for a long time. More than 40 moves later Houdini opened up the position and evals turned slightly in Stockfish's favor. Houdini found a spectacular draw, sacificing all its pieces for a stalemate.

After a few early exchanges in game 7 the engines created long pawn lines across the board, with only a few possible break points. Stockfish had a knight which could be better than Houdini's bishop in such positions.



The engines shuffled for 25 moves and then Stockfish crashed through the king side, sacrificing the knight and clearing the king side pawns. Stockfish's eval jumped over 2, Houdini's eval stayed below 1, which engine was right?



Stockfish used the black exposed king, backrank pins and threats on the queen to win a pawn on the queen side. Houdini's eval slowly increased as it saw the danger. At the right moment Stockfish exchanged queens.



Stockfish gained another pawn and its 4 passers, two connected pairs, were strong enough for a win. Two wins in a row for Stockfish playing white, this time the opening was not so biased like game 5.
In game 8 Houdini pushed pawns on the king side and it kept its king uncastled, away from the queen side where Stockfish pushed its pawns. Stockfish had a bishop pair and was a pawn up, its eval around 1, Houdini countered with better rooks and threats on the black king. Stockfish gave a rook for a knight, and after exchanging queens evals came down in a RR vs RB position. Houdini had to watch out for a black passer advancing, resulting in a draw.

Game 9 started with Houdini a pawn up and with an eval advantage. Both kings remained in the center, Houdini gave the pawn back and after queen exchange the evals dropped to 0. The game reached a drawn double rook ending. In game 10 both kings castled, Stockfish kept the pawn advantage and the pawn lines became static after move 22. Stockfish avoided a 50-move draw but the game drawn with a 3-fold repetition.

Thursday, March 29, 2018

Season 11 premier division, final results

Final crosstable after RR12


The premier league is over, Stockfish and Houdini will fight for the title in the superfinal, Fizbo and Ginkgo will be relegated to division 1 next season.

Komodo finished third, 3 points behind Houdini. Considering the 84 games each engine played this is not a large difference, but its losses to Houdini and Stockfish (five losses, one win in total) were the deciding factor.

The closest race was for 5th place, Andscacs after another weak RR score of 2/7 lost its place to Chiron (3.5/7).

The superfinal will consist of 100 games, 50 openings chosen by Jeroen Noomen. The time control will be 120+15, game duration will be close to 4 hours. Stockfish will be updated, Houdini will not. There's always some risk involved in updating the engine, hopefully there will not be serious bugs found during the games. Based on the premier division results (12 games, 11 draws, one win for Stockfish) the engines' strengths are very similar, and it is hard to say what the outcome will be.

A few interesting games

Fire - Ginkgo: After many early exchanges only QB vs QB were left on move 27. Fire had a better pawn structure, and after exchanging queens it had a pair of connected passers, enough for a win.

Fizbo - Fire: After move 14 the engines did not exchange pieces for a long while. Fizbo had a lot more space behind its pawns, and its eval was over 2, yet Fire seemed solid enough as the engines shuffled. They got to move 150 under time pressure when Fire took a rook for a knight. In the QRB vs QRR position Fire had only one pawn beyond the 6th rank. Fizbo's eval jumped as it forced a queen exchange on move 184. Fire realized it was losing, though it took a while for Fizbo to find the way to win. An unexpected result, Fire does not play well under time pressure.

Ginkgo - Komodo: The game was balanced until move 35 with most of the action on the queen side. Ginkgo grabbed a pawn, leaving its king and queen exposed to attacks. This allowed Komodo to gain the initiative, leading to a deadly attack on the white king and a win.

Stockfish - Ginkgo: Stockfish had an eval advantage after the start which increased gradually. Ginkgo chose to trade RB for Q and evals jumped. Ginkgo's rooks and bishops harrassed the white queen, but eventually Stockfish managed to trade pieces and reach a winning Q vs RB ending with 2 pawns up.

Komodo - Fizbo: Komodo took advantage of Fizbo's exposed king with not enough black pieces on the king side to develop a strong king side attack. Komodo drove the black king to the center and started to win material. The game was adjudicated in a QBN vs QN position, Komodo a pawn up.

Fire - Chiron: Fire was a pawn up with a small eval advantage after exchanging queens. Chiron allowed Fire to create a passer on the queen side while it created a passer on the king side. Chiron traded a rook for a bishop and in return got a second passer on the 2nd rank, but no further. Fire's rooks were too strong, after trading the advanced passers Fire's remaining pawns started to march, eventually forcing Chiron to lose a bishop and the game.

Andscacs - Komodo: Material was balanced when Komodo pushed its pawns on the queen side. Andscacs tried to counter on the king side, sacrificing a bishop for pawns it exposed the white king. Komodo defended carefully, giving back the material and trading down to a double rook ending. A queen side passer on the 2nd rank was enough for the win.

Houdini - Fizbo: The center was locked and Houdini a pawn up after opening the black king side. Evals were over 2 as Houdini moved its king to the queen side, apparently preparing for a king side attack which never started. Instead Houdini slowly strangled Fizbo, capturing a second pawn on the queen side. When the game was adjudicated Fizbo could not move without losing material. 

Stockfish - Andscacs: Stockfish started with an eval advantage, its pieces had more space while Andscacs' pieces could hardly move. After exchanging queens Stockfish opened the king side and pushed a passer forward, Andscacs lost a knight to stop it. The game reached a RRB vs RR position with white passers advancing in the center, a win for Stockfish. After this loss Chiron and Andscacs have the same number of points with two games remaining.

Komodo - Fire: Komodo traded a rook and pawns for two knights early in the game. Fire was 3 pawns up after a few exchanges, the engines started to stabilize the pawns and shuffle from move 27. Evals started to jump suddenly on move 84, Fire probably blundered and Komodo saw a path to win. With 4 moves left on the 50-move counter Komodo forced a knight for rook trade, and then opened up the board and traded down to a won RN vs R position.

Andscacs - Houdini: Houdini's eval started to jump on move 26. It sacrificed a knight for pawns to open the king side and then concentrated forces to attack the white king. Andscacs gave a rook to clear the black queen side pawns, in the hope that its queen side passers will slow down the attack. Houdini traded down to a N vs R ending, the white pawns were easy targets, game over. After this Andscacs drops down to 6th place, 0.5 points behind Chiron.

Chiron - Fizbo: Chiron had an eval advantage from the start, it kept its king in the center and pushed pawns on the king side. Fizbo sacrificed a knight for pawns and after a series of exchanges both kings were exposed, Chiron's eval over 1.5. Fizbo's attack on the white king was not effective, Chiron created a passer on the queen side and exchanged pieces when possible. The game was adjudicated in a QRN vs QR position with a white pawn on the 7th rank. With the win Chiron will finish 5th, Andscacs can not close the gap in the remaining game.

Featured game: Stockfish - Komodo
Premier division, round 84
Link to game on TCEC

Stockfish started with a significant eval advantage from the opening. For the first 25 moves Komodo held the position, keeping the evals close to 1. The king side looked like a fortress around the black king, on the queen side Komodo had a rook stuck in the corner, only good for protecting the a pawn.



Stockfish's eval started to increase a little, then the engines had a very long PV agreement which they started to play out. It appeared the game was heading towards a drawn RN vs RN ending.



Suddenly the evals jumped over 2. Both engines realized that by moving the knight to e3 on move 41 Stockfish could protect its king, keep the queens and force a rook exchange, leading to a QN vs QN ending.



Stockfish had checks available and it could avoid a Komodo counter attack on its king. Within a few moves Stockfish was a pawn up, the game was adjudicated before reaching a winning knight ending. This game was an excellent way to close the division.

Wednesday, March 28, 2018

Season 11 premier division statistics

Draw rate, wins

 

Final draw rate was 61.0%


Game termination

The three most common game termination causes were:
37.8% - TCEC win rule
36.3% - TCEC draw rule
14.9% - 3-fold repetition

There were no engine crashes in the division, there was one game where the runner software crashed (game 268, Ginkgo - Houdini). The game reached a drawn 6-man position and was manually set as a draw.

Moves per game

 

Median= 60
Average= 69.1

Time per game (hours)
 

Median= 3:00
Average=2:57

Openings

The first double RR was bookless, and a summary of the bookless openings can be found in a previous post. Here I summarize the results of the remaining games, with book sequences by Cato. The first letter of the ECO codes was distributed as follows:
 

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

The opening variant was mostly determined by the book sequence. 96.4% of the game pairs repeated the same ECO code twice, and 87.1% repeated the same opening variant. In all of the game pairs the ECO first letter was the same in both games.

Reverse pairs, wins

Again I'm restricting to the games with book sequences.



There were only 2 game pairs with a 1-1 biased result.

Reverse pairs, same moves 

Here I'm also restricting to the games with book sequences.



Pairs of reverse games diverged very quickly, 51.8% of the pairs diverged immediately after book, 78.6% repeated at most 2 plys.

 
 

Tuesday, March 27, 2018

Season 11 premier division, after RR11

Crosstable after RR11


The rankings haven't changed in RR11. With 7 games to go for each engine it would be highly unlikely that the top and bottom two engines will change in the last RR. The only undecided race is for 5th place. Andscacs with a very disappointing 1.5/7 score in the RR allowed Chiron to close the gap to only 0.5 points. 


A few interesting games

Stockfish - Chiron: Chiron sacrificed a pawn to break Stockfish's center, then traded down to a QBN vs QBB position. The black king was completely exposed, Chiron thought the bishop pair gave some compensation but Stockfish's eval kept increasing. Stockfish traded down to an opposite color bishop ending, usually a draw but evals jumped over 2. Stockfish had a connected pair of passers that were unstoppable, Chiron realized this as well and the game was adjudicated before they actually moved.

Fizbo - Andscacs: On move 20 Andscacs started a series of sacrifices and exchanges that resulted in a very unusual BBN vs R position on move 32, with Andscacs 4 pawns up. Fizbo's eval started to jump, Andscacs realized it was losing a few moves later. After exchanging pieces Fizbo had only a bishop and a single pawn left, but the pawn was protectable by the bishop while Andscacs' king couldn't protect its remaining pawns.

Chiron - Andscacs: Chiron had an eval advantge from the start, evals approached 1 when it created an advanced central passer. Andscacs advanced its passer on the queen side, Chiron captured it on the 2nd rank. Andscacs then gave a rook for a bishop, and after a series of exchanges only RN vs NN were left. Chiron then captured a pawn and reduced to a winning knight ending. Two losses in a row for Andscacs, the gap to Chiron in 6th place is down to 1.5 points.

Komodo - Andscacs: Komodo was down a pawn early in the game. Andscacs locked the center and had a weak king side. Komodo concenterated forces and advanced pawns on the king side, when it opened a file the attack on the king was deadly and the game was adjudicated quickly. Another Andscacs loss, Chiron is only 1 point behind.

Stockfish - Fizbo: Stockfish had two strong knights in the center, which it traded in a pretty combination to get two pawns on the king side. Evals continued to increase, Fizbo got a pawn back but had to protect its exposed king. It gave its queen for two white rooks, leading to a QB vs RRB position with Stockfish up two pawns. A king side passer meant more material for Stockfish and another win. This is Stockfish's 10th win over Fizbo in 11 games, remarkable.

Houdini - Ginkgo: Houdini had an eval advantage of over 1 in a RB vs RB position but Ginkgo held the draw. 

Fire - Komodo: In a closed position  Fire was tempted to capture a pawn while opening files, one on each side. Komodo applied pressure with a queen side rook on the 2nd rank and attacked the king on the king side. Fire's defense collapsed and the game was adjudicated as it lost its queen for a rook.

Houdini - Andscacs: Houdini had an eval advantage that stayed under 1 for more than 40 moves. Material was balanced, Andscacs' king was exposed in the corner with an open file that Houdini's pieces could use to make threats. On move 58 Houdini created a central passer and evals were over 1.5. After exchanging queens only RBN vs RBN were left, Houdini captured a pawn and created a second passer. Andscacs started to lose material and the game was adjudicated when it was a piece down. 

Ginkgo - Fizbo: Fizbo's eval was high early in the game, over 1.5, Ginkgo's eval was close to 0. Fizbo had more space but Ginkgo's position seemed solid. Fizbo gave a rook for a knight and opened the king side, and then evals started to jump. Fizbo traded down to a RN vs BN position, it was 2 pawns up with two king side advanced passers. Fizbo brought the king forward and captured another pawn to create a queen side passer. When the pawns started to march the game was adjudicated.

Fire - Houdini: Evals were close to 0 as the engines exchanged pieces gradually. Fire was a pawn up with most pawns on the board and the engines started to shuffle from move 40. 45 moves later Fire opened the queen side with a knight for pawns sacrifice. Evals jumped to 1.5 as Fire's passers reached the 7th rank. Houdini gave a rook to get rid of the passers, leading to a QR vs QN position. Houdini's eval was signalling that it was lost, but Fire couldn't find the right moves and Houdini managed to reach a draw.


Featured game: Stockfish - Fire
Premier division, round 76
Link to game on TCEC

After many early exchanges the game reached a RRB vs RRB position on move 27. A few moves later Fire captured a pawn and equalized the material. Each engine had a passer and the bishops were of opposite color, it was not clear why Stockfish's eval was over 1.



Evals kept increasing slowly. After exchanging rooks and a pair of pawns, Stockfish brought the king forward. All its pieces active and centralized, Fire's pieces were mostly blocking and defending.


Stockfish captured the black passer and pushed its own forward. Fire used its pieces to block the passer, allowing Stockfish to capture a second pawn, creating another passer on the queen side.

The white passers were too strong, Fire soon lost its bishop for one of them and the game was adjudicated



Friday, March 23, 2018

Taking a break for a few days

I will be away from the blog and TCEC for a few days. The R11 report will probably be delayed, and this may also affect the R12 report. I will try to post the statistics report for the premier division on time.

I hope there are no scandals or big upsets while I'm away.

Wednesday, March 21, 2018

Season 11 premier division, after RR10

Crosstable after RR10



There were many draws in RR10, only 8/28 games were decisive. As a result not a lot changed in the table. Each engine has 14 games left to play, the standings haven't changed since Andscacs overtook Chiron in RR7 and will almost certainly will not change until the end.


A few interesting games

Chiron - Stockfish: Evals were close to 0 after the opening with an advantage for Chiron. The king side was open and Stockfish hid its king behind a white pawn. Chiron planned to advance rooks on the queen side, and it tried to exchange queens when Stockfish's eval jumped to almost 3. With a bishop and knight sacrifice Stockfish suddenly had a pawn on the 2nd rank with all the remaining white pieces unable to help. Chiron managed to block the passer for a while, allowing Stockfish to capture two pawns and get its rooks on the 2nd rank for the win.

Stockfish - Ginkgo: Stockfish had an eval advantage that stayed under 1 for the first 30 moves. Ginkgo's queen moved forward on the king side, evals started to increase when Stockfish trapped it partially. Stockfish traded down to a QR vs QR position, the material was balanced but the black pawns were weak and the king exposed. The game reached a queen ending with Stockfish 2 pawns up, an easy win.

Komodo - Fizbo: Fizbo pushed pawns on the queen side and Komodo pushed pawns on the king side. Komodo's king was exposed, yet it had an eval of 1 after it won a rook for a knight. Evals jumped over 3 when the king side opened and the black king was under attack. Fizbo tried to exchange pieces but Komodo was too strong, the game was adjudicated before it gained more material.

Andscacs - Komodo: Komodo was a pawn up early in the game, Andscacs had a strong center and a bishop pair as compensation. Evals became negative and increasing as the game progressed, Komodo punched a hole in the center and exchanged a bishop for a knight, the black pawn majority on the queen side became more important. Komodo created two passers, Andscacs gave a second pawn to get activity for its pieces but it was not enough. Komodo exchanged queens and pushed pawns on the queen side for the win.

Fizbo - Stockfish: Fizbo gave a pawn and opened a file on the king side to get an advanced passer. Stockfish captured a second pawn as Fizbo's queen attacked the back rank and pinned a knight. When Stockfish captured a third pawn Fizbo was desperate, and after Stockfish exchanged the pawns for a rook the game was over.

Ginkgo - Houdini: The game reached a drawn 6-man position, since Ginkgo doens't have tablebase support the game continued past move 120. Then the software running the game crashed, stopping the tournament for about an hour until it was restored (I saw all this later on the twitch archive). The game was adjudicated manually as a draw, there is a rule exactly for this case. The game itself does not appear in the TCEC archive.

Houdini - Fizbo: Houdini's king stayed in the center and hid behind an advanced passer. Fizbo's king was exposed in the king side, and the queen side rook and bishop were not developed. After exchaging queens Houdini created king side passers and captured two pawns. When the passers started marching towards the black king the game was over.

Stockfish - Andscacs: The engines were essentially shuffling from move 26. Stockfish had an eval over 1, it couldn't improve for 50 moves.The break came on the queen side, Stockfish created a passer that was well supported and Andscacs couldn't block it. The pawn got to the 6th rank, there Andscacs captured it and traded rooks. This opened files and allowed the white queen to move behind the pawn line. The black king was exposed in a QRN vs QRB position, the black pawns were vulnerable. Stockfish quickly exchanged down to a winning rook ending 3 pawns up.

Komodo - Chiron: Komodo had an eval advantage after the opening, Chiron's king was a little exposed on the king side with most pieces on the queen side. Evals kept increasing as the engines exchaged pieces, until only RN vs RB remained. The black bishop was trapped in the corner and Komodo's rook could attack both the black king and the weak pawns. It took a while but when Komodo started capturing pawns the game was over.


Featured game: Houdini - Fire
Premier division, round 70
Link to game on TCEC

The game reached a closed position on move 28, with most pieces and pawns still on the board. Houdini had an eval advantage and more space, the queen side and center were blocked. The king side pawns could move, but the engines chose to freeze the pawns and shuffle.



After exchanging bishops and shuffling for 30 moves Houdini captured a pawn on the queen side. A few exchanges later there were open files on the queen side and only QRN vs QRN were left.



The engines exchanged queens, and the evals were constant for a while. Fire took over the 2nd rank with its rook and exposed the white king, while Houdini cleared the queen side and created a connected pair of passers.



Houdini's eval started to increase quickly as it pushed the b pawn forward. After forcing a rook exchange the knight ending with a pawn up was winning for Houdini. Fire could stop the queen side pawns only at the cost of its king side.

Saturday, March 17, 2018

Season 11 premier division, after RR9

Crosstable after RR9



RR9 was one of the more interesting RRs of the division. Nine of the first eleven games were decisive, though only 5 of the remaining 17 games. Houdini lost its first game of the division, Ginkgo won its first.
Stockfish had its best RR score with 6/7, and with a win against Houdini it extended its lead to 4 points. Despite the loss Houdini is still 4.5 points ahead of Komodo, Houdini beat Komodo for the third time. 
Fizbo with a weak 1.5/7, including a loss to Ginkgo, it is now 4 points behind Chiron in 6th place.

A few interesting games
 
Stockfish - Chiron: Stockfish had an eval close to 1 after the start, it had a lot more space and it pushed pawns to open the king side. Evals jumped after Stockfish blocked the retreat of a forward black bishop, locked the queen side and started attacking on the king side. Chiron's position collapsed, it was down a rook for a knight when the game was adjudicated, with more material loss coming.

Houdini - Komodo: Houdini had an eval advantage that increased steadily. It advanced pawns on the king side, its king was exposed but safe with no black piece around. On the king side the pawns pushed back Komodo's bishops. Komodo exchanged one bishop but the other was trapped. Komodo tried to counter on the queen side, it sacrificed a pawn and created a passer. Houdini stopped the passer on the second rank and was two pawns up, it had no problem to convert this to a win - though the game was adjudicated before the win could be seen. This was Houdini's third win against Komodo in the division.

Ginkgo - Fire: The game reached a RBB vs RBN position on move 28 with evals close to 0. Fire had a strong pair of central pawns and Ginkgo traded a rook for a knight to get rid of them, leading to a B vs R endgame. Fire's evals increased constantly as the pawns were eliminated. The game reached a winning 6-man position, that was then played out for 60 moves since Ginkgo had no tablebase support.

Fire - Fizbo: Fizbo pushed pawns on the king side giving Fire the opportunity to exchange them and expose the black king. Fire gave a rook to get rid of a strong Fizbo knight and immediately regained the material to reach a QR vs QR position with evals over 1.5. The black pawns were all isolated and Fire was quickly two pawns up and winning.

Komodo - Ginkgo: Evals remained low while the position was closed. After opening a file on the queen side evals climbed over 1, and stayed there in a QRN vs QRB position as the engines shuffled for 15 moves. Ginkgo then opened another file on the queen side and captured the knight, this allowed Komodo to launch a deadly attack on the black king and force a R for Q trade. A queen side passer cost Ginkgo its bishop and the game was over.

Houdini - Chiron: Houdini's eval was over 2 already on move 16. Chiron's pawns were all isolated and Houdini had a strong king side threat, but evals did not move for 10 moves. Then Houdini shifted to the queen side, got a rook on the 7th rank and captured 2 pawns. The white passers on the queen side were enough for a win.

Ginkgo - Stockfish: Stockfish had an eval of about 1, the engines shuffled for a while behind the pawn lines. When Ginkgo walked its king to the queen side Stockfish opened files there and moved its pieces to prepare an attack. Ginko's king had no pawns to protect it, Stockfish had full control of the light squares with its LSB. Soon Ginkgo ran out of ideas and the game was adjudicated, PVs showing total collapse.

Fizbo - Komodo: Fizbo was ahead by a rook for a knight, Komodo with two extra pawns to balance the material. The game reached a RRB vs RBN position with evals slightly in Komodo's favor. In the endgame Komodo played much better, its king walked to the queen side and after exchanging bishops the queen side passers were unstoppable.

Stockfish - Fizbo: The engines castled in opposite directions and Stockfish had an eval advantage. Fizbo tried a pawn storm on the king side, with one pawn reaching the 2nd rank. Stockfish hid its king behind the black pawn and countered on the queen side with a much stronger attack. Fizbo had to give a rook for a bishop, leading to a winning QRB vs QBN position.

Houdini - Ginkgo: Houdini started with an eval advantage, Ginkgo's king remained uncastled in the center. Ginkgo had advanced pawns on the queen side, after move 24 the evals turned negative since the pawns were trapping a bishop and a rook in the corner. Evals increased to around 1, Houdini was busy blocking the pawns and it seemed that Ginkgo could find a way to break the defence. However, after move 46 the evals dropped to 0, Ginkgo lost the initiative and the engines exchanged down to a drawn RRB vs RRB position. Still no win for Ginkgo, a win in black against Houdini would have been an incredible upset.

Chiron - Komodo: Evals were close to 0 with a small advantage for Komodo. After exchanging queens the evals stayed constant for a long time in a RBN vs RBB position. The engines shuffled except for a few pawn moves and exchanges, reaching a N vs B ending. Komodo's eval suddenly jumped as it found the way to win. It brought the king forward and Chiron couldn't defend both sides of the board.

Stockfish - Fire: Stockfish had an eval advantage, climbing over 1 after it captured a pawn on the queen side. Stockfish gave a pawn to create a queen side passer, and Fire lost a piece to get rid of it. Fire had three pawns as compensation, but Stockfish used a king side attack and a second queen side passer to get a decisive advantage. Fire had to give a rook for a knight, and Stockfish converted the QRR vs QR position easily.

Houdini - Andscacs: For 35 moves the evals were close to 0, after exchanging queens Houdini's evals started to climb in a RBN vs RBB position. Andscacs' king was trapped in the corner and under threat of mate, with only a pinned bishop on the back rank defending. Houdini allowed all its pawns to be captured as it brought the knight forward, Andscacs escaped the mate but lost its rook. The black pawns were not enough in a RN vs B ending, a win for Houdini.

Ginkgo - Fizbo: Ginkgo was a pawn up early in the game, the engines started to shuffle with evals at around 1. After several exchanges only QRB vs QRB were left, the evals climbed to a new level and the shuffling resumed. After 30 moves Ginkgo was ready with its pieces and it attacked the queen side, creating a passer. After trading the queen for a rook Ginkgo queened the passer, and with a rook up it was winning. Finally a win for Ginkgo in the division.



Featured game: Stockfish - Houdini
Premier division, round 58
Link to game on TCEC

Stockfish had an eval advantage from the opening. Houdini chose to give a knight for pawns early, exposing the white king for counterplay.



Stockfish was not afraid of leaving its king on its own, even moving it forward. Houdini played defensively, trying to avoid exchange pieces. Stockfish had time to get its knight to a strong square on the king side where it could reach both the central pawns and the black king. The evals climbed higher, though Houdini was still not convinced.



Houdini gave a rook to get rid of the white knight, capturing another pawn. Stockfish was up a rook for 4 pawns, and many of the black pawns were easy targets. Houdini's position was desperate, and its only hope was to get the queen forward and try to check the exposed white king.



Stockfish had to be careful not to allow a perpetual check draw. Eventually Houdini ran out of checks and it gave up. This was the first loss for Houdini in the division, it was the only engine that was undefeated up to this game. Stockfish lead grew to 3.5 points over Houdini, it is hard to imagine that Houdini can close this gap in the remaining games.

Tuesday, March 13, 2018

Season 11 premier division, after RR8

Crosstable after RR8



Stockfish with 5.5/7 in RR8 extended its lead to 2.5 points over Houdini. Komodo is losing sight of the superfinal, it had another low score of 3.5/7 including two losses, the gap to Houdini is now 4.5 points.
Fizbo scored only 2/7 in RR8, the gap to Chiron in 6th is now 3.5 points. Andscacs is now firmly in 5th place, 1.5 points ahead of Chiron. Andscacs' win over Komodo was quite a surprising result.

A few interesting games

Chiron - Stockfish: The game was quiet for the first 30 moves in a balanced position. Stockfish's eval started to increase when it applied pressure on Chiron's weak pawns. Evals were over 1 when Stockfish used a gap in the pawn wall to penetrate with its queen and make threats on the king side. After some hesitation (two 2-fold repetitions in a row) and possibly Chiron inaccuracies Stockfish traded down to a QR vs QR position. Chiron had to give its queen for a rook to save its king, a win in black for Stockfish.

Andscacs - Fizbo: Material was balanced but Andscacs had a connected pair of passers on the queen side. When the passers pushed forward Fizbo was forced to give a knight to stop them, with evals over 1. The position was complicated with no pawns on the queen side and many pieces on the board. Fizbo traded Q for RN and tried to build a fortress for its king in a QRB vs RRB position, Andscacs broke through with a bishop sacrifice and its pawns marching towards the king won the game.

Chiron - Houdini: The engines castled in opposite directions, with the queen side open Houdini placed its pieces facing the white king. Very patiently Houdini increased the pressure with evals constantly increasing. After almost 30 moves Houdini was ready, finishing off with a queen sacrifice and a knight fork, leading to a won RR vs RN ending.

Komodo - Fizbo: Fizbo held a fortress for 50 moves, then Komodo found a way to break it. After a series of exchanges only QN vs QB were left, Komodo two pawns up with evals over 2. Komodo had no problem converting to a won queen ending.

Fire - Andscacs: Andscacs kept its king in the center, Fire was a pawn up with a potential king side passer. Andscacs tried to attack the white king but didn't have enough strength. Fire attacked the exposed black king from both sides and advanced its passer on the king side, causing Andscacs' position to collapse. The game was adjudicated before Andscacs lost more material. After 7 draws Fire finally wins a game against Andscacs.

Andscacs - Komodo: For the first 30 moves the game was balanced with all pieces on the board, Komodo controlled the only open file while Andscacs had a small eval advantatge. Then the engines exchaged a few pieces, Andscacs went a pawn up and Komodo's pawn structure was ruined on the queen side. Evals were over 1, but they remained constant for over 30 moves as the engines exchanged down to a RN vs RB position. There were no pawns on the queen side and the position looked drawish, especially after the game reached a rook ending. Then Komodo's eval jumped to 4 suddenly, and two moves later Andscacs saw that its advanced central pawns were strong enough to force a win. This was quite a shock, very unusual for Komodo to lose in such a way in an endgame, while Andscacs is not considered an endgame specialist and it does not even use tablebases.

Fizbo - Stockfish: Stockfish had an eval of about 1 when the engines started to shuffle on move 20. Nothing happened except a few pawn moves and exchanges, with the draw counter on 3 on move 100 Stockfish gave a rook for a bishop to continue playing. The position opened up a little and the shuffling continued. With the draw counter at 15 and both engines on increments Fizbo blundered the game away. Starting with a queen sacrifice and knight fork Stockfish reached a R vs N endgame with too many advanced pawns that Fizbo couldn't stop.

Houdini - Fizbo: Houdini activated its rooks early and Fizbo's king was stuck in the center, evals were over one by move 16 and kept increasing. Fizbo had to defend its king and block a queen side passer, while keeping its pieces safe. Eventually Fizbo's position collapsed, it lost a rook for a knight and its second rook was trapped. The game was adjudicated in a won QRR vs QRN position.

Stockfish - Andscacs: The engines castled in opposite directions and Stockfish attacked on the king side, evals increasing rapidly. Andscacs traded a rook for a bishop and two pawns, opening files on the king side. Stockfish's rooks and queen were very active, when Andscacs tried to block a file with a knight Stockfish traded a rook for it and shattered the black defence. Stockfish was soon up a piece and the game was adjudicated.

Komodo - Fire: Komodo took advantage of the uncastled black king and concentrated all its forces on the king side. Fire traded down to a QRR vs QRR position, Komodo used its pawns to open the king side. Fire had no counter and its king tried to run away to the queen side. Komodo's threat on the back ranks was enough to gain material and win the game. This is Komodo's first win against Fire in the division.

Fizbo - Ginkgo: Fizbo started with a rook for bishop trade and forced the black king to move uncastled. Then came a knight sacrifice and an all out attack on the black king. Ginkgo defended by exchanging everything, leading to a BN vs BN position with Fizbo a pawn up by move 30. In the endgame Fizbo's king was very active, Ginkgo didn't (or couldn't) stop it from penetrating the king side and isolating the black pawns there. Fizbo captured a another pawn and exchanged bishops, leading to a won knight ending.



Featured game: Stockfish - Komodo
Premier division, round 56
Link to game on TCEC

Stockfish had an eval advantage after the opening, Komodo's eval was closer to 0. Stockfish had a bishop pair which could be powerful, but the position was relatively closed with not so much room for the pieces to move.



After move 25 Stockfish's eval jumped over 1 as it pushed pawns on the king side. Komodo had a strong center and its eval remained low, it accepted Stockfish's pawn offer, losing a tempo and creating more room on the queen side. Meanwhile Stockfish opened a file on the king side and its eval was already over 2.



Komodo started to realize it was in trouble as more files were opened. The rooks and bishops had a lot more squares to move and the knights were too slow. Komodo gave the pawn back and exchanged queens but at this stage the evals were too high to have any hope of a draw.



The black king and pieces became targets for Stockfish's rooks and bishops, Komodo had no defence. The game was adjudicated as Stockfish was about to win material. Stockfish finally with a win against Komodo, now their head-to-head score is even after Komodo won in RR1. A second loss for Komodo in RR8.

Saturday, March 10, 2018

Season 11 premier division, after RR7

Crosstable after RR7



Stockfish with 5/7 and Houdini with 5.5/7 continue to lead the table. Komodo was only 1 point behind when the previous RR ended, but with a very weak score of 3/7, with no wins and a loss to Houdini, the gap is now 3.5 points. 

Andscacs continues to move away from relegation, after this RR it is 5th, 0.5 points ahead of Chiron in 6th. Fizbo with a score of 3.5/7 gained on both but needs 3 points to close the gap.

A few interesting games

Fizbo - Andscacs: Fizbo had an eval over 1 while Andscacs' eval was 0. The engines shuffled from move 25, after they exchanged bishops the evals increased but the engines continued to shuffle. Andscacs captured a pawn, Fizbo waited for the draw counter to reach 3 before forcing another exchange (move 109 !!). The reasonable outcome was a draw but after 25 more moves Fizbo's eval rose suddenly above 6, a mating threat forced Andscacs to lose material and the game. Did Andscacs blunder under time pressure? Did Fizbo stumble upon a win by mistake?

Houdini - Chiron: Only bishops and rooks were left by move 16, Houdini had an eval advantage due to Chiron's isolated queen pawn. After reducing to a RB vs RB ending Houdini found ways to improve its position, evals slowly increasing. Houdini pushed passers on both sides of the board, once the black pawns on the queen side were gone Chiron couldn't stop the white pawns.

Ginkgo - Stockfish: Ginkgo held the position for 45 moves before Stockfish's eval started to increase. Stockfish exchanged down to a RB vs RN position, captured a pawn and reached a winning endgame.

Fizbo - Komodo: Fizbo was a pawn up in a RBN vs RBB position. After exchanging a pair of bishops Fizbo had a pawn majority on the king side, its eval was over 2. Komodo's eval peaked at 1.4 but still it found a defence to get a draw. With better endgame skill Fizbo could have won perhaps.

Chiron - Fire: Chiron pushed pawns on the king side making its king exposed. Fire took a long time to prepare its attack, with evals gradually increasing. When Fire was ready it sacrificed a bishop to remove the final pawn defender, its eval was over 3. Fire quickly regained the material and the game was adjudicated just as it was gaining more, leading to a winning QB vs QR position.

Stockfish - Fizbo: Stockfish captured the b pawn with its queen, and later had to retreat to safety. It created a queen side passer while giving Fizbo a rook for a knight. For a long time the engines mostly shuffled with evals constant, only after move 60 evals started to increase. Stockfish exchanged pieces keeping its queen side passer until the game reached a won queen ending.

Houdini - Ginkgo: Houdini used the open f file to develop a deadly king side attack while many Ginkgo pieces were stuck on the queen side. Ginkgo gave a rook for a knight but evals were already over 6 and the game was soon over.

Stockfish - Fire: Stockfish's eval was over two already at move 15, Fire's pieces were undeveloped on the queen side leaving the king side vulnerable. With queens gone Stockfish doubled rooks on the open g file, Fire was forced to lose material to get out of trouble. The game was adjudicated before Stockfish won a piece.

Houdini - Andscacs: The game reached a RRB vs RRB position with Houdini a pawn up and evals about 1. Andscacs lost a second pawn but traded down to an opposite color bishop ending, however it was not a draw in this case. The game reached a won 6-man position, again Andscacs losing an endgame partly because it has no tablebase support.

Andscacs - Ginkgo: The game reached a QR vs QR position, Andscacs a pawn up with an advanced central passer. Evals remained constant for over 30 moves and it seemed that Ginkgo was holding. However, Andscacs found a way to win by moving its king to the queen side and then exchanging down to a won king and pawns ending.

Featured game: Houdini - Komodo
Premier division, round 43
Link to game on TCEC

Houdini had a small eval advantage after the opening,which slowly increased to about 1. Houdini found a great outpost for its knight on d6, Komodo had many pieces on the king side, perhaps anticipating an attack there. Instead Houdini opened the queen side, with evals approaching 2.



After exchanging a few pieces Houdini captured a pawn and created a queen side passer. In a QRN vs QRB position Komodo's bishop was on the wrong side of the board and not useful.



Houdini traded down to a queen ending, and the connected pair of passers on the queen side was enough for a win. This is the second win for Houdini against Komodo, extending the gap between them to two points. The win looked too easy, perhaps the opening bias is too great here. Can Komodo beat Houdini in the reverse?

Wednesday, March 7, 2018

Season 11 premier division, after RR6

Crosstable after RR6



With 6 RRs finished the division is at the halfway point. It is safe to say that Ginkgo will finish last, and that two of the leading 3 engines will qualify for the superfinal. It is likely that Fizbo will be the second engine to be relegated, and that Stockfish will be in the superfinal.

Stockfish continues to lead the table. The race for second place between Houdini and Komodo is close with 1 point separating the engines. In RR6 both engines scored 4.5/7. At the bottom Fizbo had another disappointing RR with 2.5/7, while Andscacs scored 4/7 increasing the gap to 4 points. Andscacs has caught up with Chiron, winning their head to head game in the RR.

A few interesting games 

Fire - Ginkgo: Fire had an eval advantage from the start. The evals slowly increased in a closed position while Fire prepared a king side attack. When the king side opened Ginkgo lost a bishop that was trapped there, Fire just exchanged everything to get a winning endgame.

Andscacs - Fizbo: In the game between 6th and 7th place Andscacs had an eval of over 1 after the opening. Fizbo gave a knight for pawns and countered with a connected pair of passers. Andscacs allowed a pawn promotion but gained material to get a QBN vs QR position, leading to a long shuffle and a draw.

Andscacs - Chiron: Andscacs pushed its pawns on the king side, Chiron felt the pressure and evals climbed quickly. Andscacs opened the king side with a knight for pawns sacrifice, Chiron could not defend its exposed king against the white pawns without losing significant material. With this win Andscacs moves further away from the relegation zone, catching up with Chiron in 5th position.

Fizbo - Fire: Fire took the initiative after a few moves, by move 31 it was a pawn up with passers on both the queen and the king side. Evals remained constant for 20 moves, then Fire found a winning line. After a series of exchanges only N vs B were left, Fire was two pawns up and winning. Not sure whether Fizbo blundered or Fire was winning all along.

Ginkgo - Komodo: The game was balanced until move 30, then Komodo sacrificed a pawn and opened the king side. The white king was exposed and Ginkgo gave a rook for a bishop to remove the immediate threat. Komodo was in control in the QRB vs QRR position, it took 20 more moves of slow improvement to complete the win. Komodo is now only 0.5 points behind Houdini.

Stockfish - Ginkgo: Ginkgo was up a pawn, while Stockfish had a strong queen side attack with a rook on the back rank. Stockfish's eval was over 2 in a QRB vs QRB position, it used its control of the back ranks to get the pawn back. Ginkgo had to give another pawn to free its bishop, Stockfish then traded down to a winning queen ending.

Fizbo - Stockfish: Stockfish's evals were over 1 in a closed position. For a while the engines shuffled and the evals drifted down, but then Stockfish found a winning line. It opened the center and after trading a rook for a bishop and exchanging queens it created a group of 3 central passers (one doubled). Fizbo's counter attack on the black king was not strong enough to prevent queening and a decisive material advantage.

Chiron - Ginkgo: Ginkgo's pieces had very little room to move, Chiron patiently organized its pieces and didn't open the position until it was ready. The evals increased all the time and were over 2.5 when Chiron opened the king side. Chiron continued to improve slowly, collecting pawns when it could, it was two pawns up with evals over 5. The game was soon adjudicated.

Houdini - Fizbo: Fizbo held the position for 40 moves, but when its queen moved forward alone on the queen side Houdini attacked through the center. Using the exposed black king Houdini was soon two pawns up and then it traded down to a winning rook ending.

Stockfish - Andscacs: Stockfish had a significant eval advantage from the opening. It slowly arranged its pieces for a king side attack while the evals increased. When it was ready Stockfish sacrificed a knight to open the king side, eval already over 4. Andscacs gave back a bishop and tried to hide its king behind a white pawn, but Stockfish had too many pieces aimed at the black king and Andscacs couldn't avoid losing material.

Komodo - Chiron: Chiron's king remained uncastled in the center, making it hard to develop the king side rook. Evals started to increase when Komodo dominated open files with rooks on both sides of the board. After exchanging queens Komodo was a pawn up. It then pinned a black knight on the back rank, eventually Chiron couldn't free the bishop or defend it, enough for a win for Komodo.

Fizbo - Ginkgo: Fizbo sacrificed a knight to open the king side. For a while Ginkgo's eval remained low, its king seemed to be relatively safe. Fizbo attacked through the center and pushed a passer to the 7th rank, Ginkgo gave the piece back and exchanged pieces, leading to a RB vs RN position. Fizbo was a pawn up and evals were climbing, it took another 20 moves for Fizbo to convert the win.

Ginkgo - Andscacs: The engines shuffled pieces with evals close to 0 until Ginkgo captured a pawn opening a file on the queen side. Andscacs' rooks used the open files to take over the queen side. This allowed Andscacs to capture two pawns, trade all pieces and get a winning king and pawns ending.

Stockfish - Komodo: Stockfish had an eval advantage after the opening, it chose to sacrifice a knight for 3 pawns very early in the game. Komodo's eval stayed close to 0 as the engines exchanged pieces. Stockfish's eval peaked at 1 but dropped quickly after the game reached a R vs RN position. Stockfish was 4 pawns up but couldn't get more than a draw.


Featured game: Houdini - Fire
Premier division, round 42
Link to game on TCEC

Houdini's eval jumped very early in the game, it saw an opportunity to be a pawn up. A few moves later Fire saw it too with a long PV agreement. At the end of the series of moves not many pieces were left, Houdini had a pawn majority on the queen side while Fire had a central passer.



Houdini chose not to exchange queens, after a few pawn moves the engines started to shuffle. Evals remained constant and Fire seemed to be holding. After 20 moves Houdini found a way to improve, it exchanged a pair of rooks and cleared the queen side for its pawns.



Houdini started to push its pawns forward and Fire had many threats to deal with: the king was in danger, its pieces were attacked, the white queen side passers became an immediate threat. The evals increased rapidly as Fire used up all its defences.



Fire had to give a rook for a knight to capture one of the queen side passers, the other passer reached the 7th rank. Fire launched a desperate QB attack on the white king, only to drive the king near its passer. Houdini was about to win more material when the game was adjudicated. 

Saturday, March 3, 2018

Season 11 premier division, after RR5

Crosstable after RR5


Stockfish and Houdini are in the lead, but Komodo is the winner of RR5 with 5.5/7 points. The gap between Houdini and Komodo is down to one point, making the race for the superfinal interesting.
At the bottom Fizbo had 2 losses in the RR, increasing the gap to Andscacs by a point.

A few interesting games 

Komodo - Ginkgo: Ginkgo chose to trade a R and pawns for BN early in the game, Komodo is an expert in these situations and its eval started to increase. Ginkgo then added to the imbalance by trading RB for Q, leading to a QBN vs RRN position. Komodo attacked the black king and the black pieces defended passively, the 3 extra black pawns were no compensation. The game was adjudicated just before Komodo gained material.

Stockfish - Houdini: Stockfish had an eval advantage after the opening. For a while the engines shuffled behind their pawn lines. Houdini gave a pawn to open the queen side and threatened the white king, Stockfish's eval started to increase when it seemed the white king was safe. The eval peaked over 1.5 but then dropped after Houdini gave two more pawns and found a perpetual check draw. Houdini still without loss in the division.

Houdini - Chiron: Houdini sacrificed a bishop for pawns, opening the king side to create a deadly attack on the black king. The game was over quickly.

Fizbo - Komodo: The game was balanced for the first 40 moves. Fizbo sacrificed a bishop and exposed the black king, but it soon realized it miscalculated and that Komodo could find a defence. When the attack was over Komodo was still a piece up in a winning Q vs QN ending. This was Komodo's first win against Fizbo in the division.

Komodo - Andscacs: Komodo was ahead on eval despite being a pawn down and Andscacs having a bishop pair. In a series of exchanges Andscacs traded down to an imbalanced QN vs RRB position, with evals over 1.5. It took a while but Komodo had no real problem to convert the win. This is Komodo's 3rd win in the RR, still in the race for 2nd place.

Stockfish - Fizbo: Stockfish had an eval advantage in a closed position. Evals started to increase as the position opened up, Stockfish had a pawn majority on the queen side which later became a passer. Together with a back rank attack on the black king this was enough for a win.

Houdini - Ginkgo: The game reached a RB vs RB position on move 28. Houdini's eval was over 1, it had a pair of connected passers on the queen side while Ginkgo had a central passer. When Houdini started to push its passers Ginkgo was forced to give its bishop to stop them, resulting in a won B vs pawns ending.

Andscacs - Ginkgo: Andscacs took advantage of Ginkgo's uncastled king, sacrificed a knight and attacked through the center. Evals jumped over 2 on move 15 and over 4 by move 22. Andscacs got the material back while Ginkgo struggled to get its king to safety and its pieces developed. The game ended with Andscacs in a QR vs QB position with Andscacs two pawns up. After this win Andscacs' gap to Fizbo increased to 2.5 points.


Featured game: Chiron - Komodo
Premier division, round 34
Link to game on TCEC

The game was balanced for over 40 moves, with evals very close to 0. The engines gradually exchanged pieces until reaching a RB vs RB ending on move 44.


Komodo had a passer that Chiron had to watch, all other the pawns were static and invisible to the bishops. Komodo's eval started to increase, it kept the white king on the king side while moving its king to the queen side. There the black king was safe and applied pressure on the weak a2 pawn.


Komodo then used the bishop as a blocking piece, it isolated the a pawn so the rook couldn't protect it and captured it.



At this stage Chiron admitted defeat. The PV showed a possible rook for bishop sacrifice to remove the queen side pawns and create new passers. Komodo's fourth win of the RR, second win in black.

Friday, March 2, 2018

Season 11 live blogging, Friday 2nd March

Finally I have some time to look at a game live, lately I have been barely able to keep up with the game reports.

This is the round 34 Ginkgo - Fizbo game. Ginkgo has no real chance of avoiding relegation, even though there are still many games to play in the division. This is however a chance to get a win, none for Ginkgo so far. Fizbo is still hanging on, hoping to catch up with Andscacs in 6th place. In RR5 Andscacs gained 0.5 points on Fizbo, Fizbo needs wins but playing black it will be a challenge.

move 11: Ginkgo was given an advantage out of the opening, so far Fizbo is not worried, it will probably castle its king soon.

move 13: The black knight on b4 looks annoying, black also has some control on the long diagonal facing the white king and Ginkgo blocks that first, and then a3 to drive the knight away.

move 15: It seems Ginkgo wants to exchange LSBs, the black bishop on b7 is a danger for the white king. However, pushing e5 creates weak squares in the center. Fizbo looks comfortable though Ginkgo has a space advantage.

move 16: Evals are going slowly down towards 0, if we see exchanges in the PVs the game may become a fast draw. Fizbo wants to get rid of the e5 pawn to free its knight. It appears it wants to go d6, in the previous PV it thought about f6.

move 19: As expected Fizbo moved d6 to remove the e5 pawn. The c and d files are open with black major pieces on them. Fizbo's PV shows all bishops exchanged, the game is heading towards a boring draw. We need some kind of blunder to make it more interesting.

move 21: Ginkgo exchanged the LSBs first, Fizbo's eval is negative for the first time, now the black queen is on the long diagonal.

move 24: Fizbo is looking better now that the center is cleared and with its bishop and queen aiming at the white king. Ginkgo is exchanging rooks, Fizbo likes it and its eval is a little more negative.

move 26: the white d pawn is isolated and the f pawn is also weak. The engines are fighting for control over the e5 square, a potential for more exchanges.

I had to go for an hour, came back to move 60, looking at what I missed...
Fizbo's evals continued to be negative, even getting close to 1. Ginkgo exchanged everything and by move 50 only queens remained. On move 63 Fizbo realized it had nothing and the game will either be adjudicated or end by repetition.

move 72: queen exchange, the engines are prolonging the inevitable draw.

The draw is not good for either engine, Ginkgo still without a win, Fizbo is 2 points behind Andscacs.