Season 14 division 2, final results

Final crosstable:



Leela and KomodoMCTS continue to move up, promoting to division 1. Gull and Texel are relegated to division 3.


With 3 rounds to go Leela was leading and sure of first place. The race for 2nd place was between KomodoMCTS and Xiphos, both with 14.5 points. Texel was last and could not avoid relegation. The race for 7th place was between Booot 12.5, Nirvana 11.5, ChessBrain 10.5 and Gull 10.5. ChessBrain had one crash which gave it a tiebreak disadvantage.

Nirvana - Booot: Evals stayed low as the engines exchanged pieces until adjudication.

Leela - Xiphos: The game reached a same colored bishop ending on move 32. Leela's eval jumped to 1.5 but Xiphos defended well, the game ended in a tablebase draw on move 143.

KomodoMCTS - ChessBrain: ChessBrain had a small eval advantage in the middle game, the game reduced to a rook ending which KomodoMCTS drew easily.

Texel - Gull: Gull had a real chance to win this game. It created a queen side passer and Texel gave a knight to remove it. The game reached a B vs BB position with negative evals over 2, but Gull couldn't find a way to convert. The game ended in a tablebase draw after 65 more moves.

After 4 draws there is no change in the table. Booot is almost surely safe from relegation.

Texel - Nirvana: Evals were low, the engines reduced to a rook ending and a draw.

Gull - KomodoMCTS: Evals were close to 0, Gull gave a rook for a knight and created a connected pair of passers on the queen side in a BN vs RB position. Then evals turned negative, KomodoMCTS' eval increased rapidly as it created a king side passer. Gull lost a knight for the black passer and KomodoMCTS won the endgame.

ChessBrain - Leela: The game quickly reduced to a drawn rook ending.

Xiphos - Booot: The engines exchanged all pieces and reached a drawn king and pawns endgame.

One round to go. KomodoMCTS leads Xiphos by 0.5 points. Booot and Nirvana are safe from relegation, ChessBrain is 0.5 points ahead of Gull in 7th place.

Nirvana - Xiphos: Evals were low in a mostly closed position. The engines exchanged pieces until only RB vs RB were left and the game was adjudicated. With this draw Xiphos ties with KomodoMCTS, its only hope is that KomodoMCTS loses its game against Texel, not likely.

Booot - ChessBrain: The game reached a double rook ending on move 25 and immediately ended in a repetition draw. ChessBrain leads Gull by a point, Gull needs a win against Leela to avoid relegation, also not likely.

Leela - Gull: The game reached an opposite color bishop ending on move 31, Leela wouldn't lower its eval for draw adjudication until move 165. Gull finishes 7th and is relegated to division 3.

KomodoMCTS - Texel: Evals increased from the start. The game reached a RB vs RB position, KomodoMCTS went a pawn up and created a passer in the center. Texel's pieces were paralyzed and KomodoMCTS won the game, though it shuffled for a while and it was not clear what the winning sequence was. KomodoMCTS finishes in 2nd place.

Season 14 division 2 live blog, Wednesday Dec 12th

In this post I'm not following a game. I was away for 2 days and missed all of RR3 of division 2. I'm having a quick look at the archive to see what I missed.

1. First thing that stands out is that Leela lost, Huh? Is this legit? The previous loss was a network failure and the game was manually adjudicated as a win for Leela. What happened here? Round 18 against Booot, the eval graph shows a blunder around move 85. Leela was up a pawn, the engines shuffled for a long time and then Booot pushed a pawn on the queen side and evals started jumping in black's favor. Did Leela blunder? What did SF (bluefish) have to say? The game reduced to a BN vs RB position which Booot won. Looks like a real loss to Leela. I bet the fans went crazy.

2. Leela is sure to promote to division 1, even if still not mathematically certain. After 20 round it had a 5 point lead over 3rd place, 3.5 points to 2nd place.

3. Komodo and Xiphos had a miserable RR, both with 3/6 points. The race for 2nd place is still open.

4. At the bottom of the table Texel is on its way to finish last, but hey - it won a game? Round 18 against ChessBrain? Again round 18 - a cosmic ray hit the server perhaps. The game reduced to a R vs B ending, Texel used its tablebase knowledge to convert the win.

5. At the bottom the race against 7th is very close. Gull (2.5/6) and Chessbrain (1.5/6) had a weak RR. Nirvana (3.5/6) and Booot (4/6) had better results. One point separates these 4 engines with 8 games to go.

Season 14 division 2 live blog, Monday Dec 10th

Division 2 is being played, the engines are stronger and the competition is harder. After 10 rounds it seems that Leela is going to win this division, but the second engine to promote could be any of KomodoMCTS, ChessBrain or Xiphos. The other 4 engines (Gull, Nirvana, Booot and Texel) are all in danger of relegation.

We follow the round 11 game Xiphos - KomodoMCTS, the reverse game was a draw. The book sequence is a QGD semi Slav defense, a popular opening line.

move 10: KomodoMCTS diverges from the reverse game in its first move out of book, choosing a less popular line. Its second move takes the game to a very rare line in human games.

move 15: Xiphos castled long, opposite direction castling may result in a fighting game, evals are low so far. KomodoMCTS opened the center, no pawn pushes so far by both engines.

move 24: Xiphos' eval increased a little, but it doesn't seem to have a plan so far. The engines exchanged all bishops and opened the position. There are many open files and the rooks are in danger of being exchanged soon.

move 31: Xiphos gave a pawn for some initiative on the king side. One pair of rooks was exchanged and only QRN vs QRN remain. I think the game is heading for a draw.

move 40: Now the queens are off, Xiphos is still a pawn down yet its eval is still around 0.4. The engines cleared more pawns and Xiphos is a pawn up, with two pawns on the queen side while KomodoMCTS has one pawn on the king side. KomodoMCTS captures one of the white pawns, Xiphos' eval jumps to around 1, I think it is delusional.

move 47: Xiphos' eval jumps over 2, KomodoMCTS reduces to a tablebase draw.

This result keeps the tension at the top of the table, the race for 2nd place is very close.

Season 14 division 3 statistics

A summary statistics table of previous stages and seasons. 

Draw rate, wins


Final draw rate was 45.5%. The draw rate was higher than in division 4 but still quite low.


Game termination

The three most common game termination causes were:
37.5% - TCEC win rule
22.3% - SyzygyTB
18.8% - TCEC draw rule

Hannibal crashed 5 times in the division, Arasan and rofChade crashed once each.

Moves per game


Median= 62.8
Average= 68.9

There were 13 games longer than 100 moves, the longest was 199 moves (Leela - Pedone, round 11, Leela won).

Time per game (hours)


Median= 1:15
Average= 1:14

Openings

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


I'm dropping the opening 'family name' section. I noticed that the current opening naming used by TCEC is not consistent (e.g. "Sicillian", "Sicillian defense" both used, the separator can be ":" or ","). This has probably been true for several past reports.

The engines had a lot of freedom to choose the opening variant. Yet 57.1% of the game pairs repeated the same ECO code twice, and 50% repeated the same opening variant. In 7.1% of the game pairs the ECO first letter was not repeated twice.

Reverse pairs, wins


Reverse pairs, same moves 


Pairs of reverse games diverged very quickly, 50% diverged immediately out of book, 83.9% of the pairs diverged at most after 1 move. The longest repeated sequence of moves was 8 plys (rofChade - Hannibal, rounds 6 and 13, KGD Keene's defense, Hannibal crashed as white, otherwise two draws)

Season 14 division 3, final results

Final crosstable:
 


Hannibal was disqualied due to crashes. The table is not accurate since all Hannibal's games were ignored. However, the promotions and relegations do not change. Leela and KomodoMCTS were far better than the other engines in this division. The bottom of the table was more crowded, in the end Nemorino was relegated back to division4.

The bottom of the table with 3 rounds left was:

rofChade 9.5, 2 games left
Arasan 9, 3 games left
Pedone 9, 2 games left
Vajolet 8.5, 3 games left
Nemorino 8, 2 games left

At this stage Nemorino was most likely to be relegated. Vajolet was still in danger despite having more games to play, it still had to play both table leaders.

Pedone - rofChade: Pedone pushed pawns on the king side, exposing its king while attacking rofChade's king. Pedone exchanged Q for RN, opened the king side and created an advanced passer. rofChade lost its queen for a rook to stop the passer, leaving Pedone a knight and pawn up in a winning RBN vs RB ending.

KomodoMCTS - Arasan: Arasan survived a long game. After 50 moves Arasan gave a rook for a knight and pawn, KomodoMCTS' eval jumped to around 1. Arasan exchanged down to a R vs N endgame, it used a passer on the 2nd rank to hold the draw.

Vajolet - Leela: Vajolet was up a pawn with a queen side passer, it held the position until move 48. Leela pushed pawns on the king side and (negative) evals jumped over 2. The game reached a RB vs RR ending with a black pawn on the 2nd rank, Leela converted the win.

Two rounds to go. Vajolet is still 0.5 points ahead of Nemorino, both engines with two games left to play. The other engines increased their gap to at least 1.5 points, almost surely safe.

Vajolet - Pedone: Vajolet was up a pawn early in the game, Pedone did not find compensation and evals slowly increased. Vajolet created a queen side passer and exchanged down to a RB vs RN position. Pedone lost a knight for the passer and Vajolet won.

Arasan - Nemorino: The position was more or less static from move 27, on move 53 only the major pieces remained on the board and the engines shuffled until draw adjudication.

Last round: Pedone and Arasan are safe, Vajolet and rofChade are 1 point ahead of Nemorino. Nemorino's last game is white against Leela, and it needs a win, highly unlikely.

rofChade - Arasan: The game reached a double rook ending on move 29. Both engines pushed passers, the game ended in a perpetual check. rofChade is safe.

Nemorino - Leela: Leela gave two pawns and attacked the uncastled white king. Nemorino exchanged pieces, the game reached a rook ending. Leela regained the pawns, the black pawns were much better with a connected pair of passers on the king side. Leela won the game easily. After this loss Nemorino is relegated, Vajolet is safe.

Season 14 division 3 live blog, Thursday Dec 6th

Division 3 is being played, 3RRs done and one to go. The qualifiers for division 2 will be KomodoMCTS and Leela, they have a huge lead over the other engines in the division. They are figthing for first place with Leela ahead by 0.5 points, Leela is also leading 1-0 in the head-to-head score. The bottom of the table is less clear. Hannibal is last, disqualified due to crashes though it is still playing. Nemorino is in 7th place, 0.5 points below Vajolet  and 1.5 points below Arasan.

Currently Pedone is playing against KomodoMCTS at the start of round 22. The reverse game was a draw.

move 10: Pedone chose a less popular sideline in the Sicilian opening, still played at the top GM level. Pedone castled long, KomodoMCTS hasn't castled yet. The black queen side does not look safe for castling, KomodoMCTS may castle short and white can attack with its pawns.

move 15: KomodoMCTS castled short and Pedone is pushing its king side pawns. Evals are close to 0, Pedone is not attacking hard, it seems it doesn't have a clear plan.

move 21: Pedone exchanged pawns in the center, opening the position. It is a pawn up and has a bishop pair but evals remain close to 0. The rooks could be exchanged soon, will this be a quick draw?

move 25: Pedone seems to be preparing to push its g pawn. Meanwhile KomodoMCTS pushed a pawn on the queen side, Pedone's evals turned slightly negative after a long think.

move 30: Pedone continues to be slow on the king side and negative evals are increasing. After exchaging a pair of rooks KomodoMCTS' eval was over 1, the white pieces are on the king side and the white king does not look safe.

move 35: KomodoMCTS is threatening mate on the queen side, its eval jumping. Pedone was too slow and now it seems too late. Most of the white pieces are too far to help the king, now it seems black will win soon.

move 40: Now Pedone pushes the g pawn in desperation. KomodoMCTS ignores this and continues with its queen side attack. It captures a pawn, its eval over 5.

move 48: Pedone loses its queen for a bishop and a knight, game over.

Season 14 division 4 statistics

A summary statistics table of previous stages and seasons. 

Draw rate, wins

 
Final draw rate was 34.1%. The low draw rate reflects the wide range of engine strengths in this division. In particular ScorpioNN lost 43/44 games and KomodoMCTS won 32/44 games (and lost 1).


Game termination

The three most common game termination causes were:
43.9% - TCEC win rule
17.8% - SyzygyTB
12.9% - TCEC draw rule

ScorpioNN crashed 8 times in the division. It was allowed to continue playing since it lost all its games anyway, it had no effect on the table. Also of note are the 23 games that ended with mate on the board, with ScorpioNN the loser in almost all.

Moves per game


Median= 58.5
Average= 65.1

There were 27 games longer than 100 moves, the longest was 183 moves (Rodent - Tucano, round 37, draw).

Time per game (hours)
  
Median= 1:12
Average= 1:12

Openings

There were 2-move book openings in this stage chosen 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 - 26 times
QGD - 14 times
French - 13 times

The engines had a lot of freedom to choose the opening variant. Yet 56.8% of the game pairs repeated the same ECO code twice, and 50.8% repeated the same opening variant. In 10.6% of the game pairs the ECO first letter was not repeated twice.

Reverse pairs, wins


Reverse pairs, same moves 


Pairs of reverse games diverged very quickly, 43.9% diverged immediately out of book, 79.5% of the pairs diverged at most after 1 move. The longest repeated sequence of moves was 10 plys (ScorpioNN - pirarucu, rounds 2 and 13, KGD Keene's defense, pirarucu won both games)