Armageddon Openings released

[pohl4711]

Armageddon Openings for Computerchess

No longer any draws in computerchess !!! No boring draw-marathons, no draw-death of computerchess.

https://www.sp-cc.de

Direct download:
https://www.sp-cc.de/files/armageddon_openings.zip


Base idea is from Larry Kaufman (Komodo-Team):
"I have wondered what would be the simplest rule change to chess that would make it a two-result game (win or loss) while still being roughly fair to both players. In the real chess world thisis sometimes done by giving White more time and Black draw odds, but with a few exceptions this has been done only at ten minute or faster time limits. Let's say we won't use time in the rules, equal time is assumed. Here is what I think might be an original, simple, and fair enough solution. White is not allowed to castle long, Black is not allowed to castle short, Black wins draws. That's it. It is a subset of normal chess; if you stipulate the starting sequence 1.Na3 Nh6 2.Rb1 Rg8 3.Ra1 Rh8 4.Nb1 Ng8 you get the above castling status. Since castling short is easier to accomplish than castling long and doesn't require an extra move to "finish" the castle (Kb8) White's advantage is obviously larger than normal, despite the symmetry of the rule, and the fact that same side castling never happens makes the draw odds less significant. But it's far from clear which side has the better chances or the theoretically won position.“

This is a very nice idea, and so I decided to build opening-sets with that starting sequence (1.Na3 Nh6 2.Rb1 Rg8 3.Ra1 Rh8 4.Nb1 Ng8) out of my 4-moves-openings (SuperGM_4moves and IM_4moves), FEOBOS openings and out of a 6 pawnplies-opening set.
The problem of Armageddon, used in computerchess is, that the faster your machine or the longer the thinking-time in your engine-games is, the higher gets the number of draws (a "law" of computerchess). In Armageddon, this means, the whitescore gets lower and the blackscore gets higher, because all draws are counted as a win for black.
Because of this, I decided to build 2 different Armageddon-lines, which give more advantage for white, for higher whitescores. So, depending on the thinking-time and/or the speed of the machine, a higher whitescore-level Armageddon can be chosen. The 3 levels are:

Level 1: Larry Kaufman's Armageddon idea of opposite castling (I call them SALC Armageddon): White can castle short, black can castle long. Line: 1. Na3 Nh6 2. Rb1 Rg8 3. Ra1 Rh8 4. Nb1 Ng8

Level 2: NBC (= No Black Castling): White can castle to both sides, black is not allowed to castle. Line: 1. Na3 Nh6 2. Nb1 Rg8 3. Na3 Rh8 4. Nb1 Ng8 5. Nc3 Na6 6. Nb1 Rb8 7. Na3 Ra8 8. Nb1 Nb8

Level 3: PawnPlus-Armageddon: White is one pawn ahead – black has no pawn on a7. All castlings are allowed for white and black. Line: 1. Na3 a6 2. Nb1 a5 3. Na3 a4 4. Nb1 a3 5. Nxa3 Na6 6. Nb1 Nb8

Each level contains 4 opening-sets: SuperGM_4moves, IM_4moves, FEOBOS, 6 pawnplies. As PGN- and EPD-file. I built opening-books out of the 2 bigger files (IM_4moves and FEOBOS) for Fritz, Arena and polyglot. IMPORTANT: The ShredderGUI has a serious bug, which leads to move-loops in the opening, because, the castling-flags are not set. So, after the moveline 1.Na3 Nh6 2.Rb1 Rg8 3.Ra1 Rh8 4.Nb1 Ng8, the castlings are still allowed and the book would play the same line again. I talked to the author of Shredder (S.Meyer-Kahlen), he confirmed that problem, but did not fix it. If this bug will be fixed in the future, I will release ShredderGUI-books, too.

The download contains a little tool, written by Thomas Zipproth, which can be used, to change all draws in a pgn-file of engine-games to a win for black. But you can use an editor („search & replace“) for that, too. Just replace the string „1/2-1/2“ by „0-1“.

I played 10000 test-games with the Armageddon openings. You find them in the download, too.

The 4 golden and IMPORTANT rules for using the Armageddon Openings:

1) Not made for playing no Armageddon - so it is absolutely necessary to set all 1/2-1/2 results (draw) of the played games to 0-1 (win for Black), when an engine-tournament or testrun is finished.

2) Not made for playing versus other books or opening-sets: In an engine-tournament or testrun ALL engines must use the same Armageddon openings-set (or book).

3) Only the endpositions of the PawnPlus-Armageddon openings sets are checked by Komodo. But not the moves, leading to these endpositions. In that movelines, strange or bad positions can be found. So, never cut the openings or books before they are finished - that will make them useless and broken.

4) The faster your machine or the longer the thinking-time in your engine-games is, the higher gets the number of draws (a "law" of computerchess). In Armageddon, this means, the whitescore gets lower and the blackscore gets higher, because all draws are counted as a win for black. To avoid too low whitescores, use a higher Level (1-3) of my Armageddon openings: The higher the level, the higher the advantage for white in the endpositions of the openings -> the higher the whitescore.


Here the results of my testings:

Test-environment:

Quadcore (Stockfish around 1.8 MN/s (singlethread)), 256 MB Hash, no bases, no ponder, 3 games running simultaneously.
Stockfish 190728 (Contempt= +25) vs Komodo 13.01 (Contempt= +20).
Because Stockfish is so much stronger than Komodo, Stockfish gets only 50% of the thinking-time of Komodo:
Stockfish 60''+500ms, Komodo 120''+1000ms (= 2'+1''). Average game-duration: 5 minutes
Each testrun: 500 games (16 testruns = 8000 games), cutechess-cli, Errorbar of each run around +/-11 Elo

Standard openings, for comparsion:
IM_4moves:
White Wins: 143 (28.6 %), Black Wins: 85 (17.0 %), Draws: 272 (54.4 %), White Score: 55.8 %, Black Score: 44.2 %
1 SF 190728 (half time) : 3448 500 (+148,=272,- 80), 56.8 %
2 Komodo 13.01 : 3400 500 (+ 80,=272,-148), 43.2 % (Elo-spreading: 48 Elo)

SuperGM_4moves:
White Wins: 152 (30.4 %), Black Wins: 72 (14.4 %), Draws: 276 (55.2 %), White Score: 58.0 %, Black Score: 42.0 %
1 SF 190728 (half time) : 3454 500 (+150,=276,- 74), 57.6 %
2 Komodo 13.01 : 3400 500 (+ 74,=276,-150), 42.4 % (Elo-spreading: 54 Elo)

FEOBOS (final 20.1-Version, contempt 3):
White Wins: 139 (27.8 %), Black Wins: 85 (17.0 %), Draws: 276 (55.2 %), White Score: 55.4 %, Black Score: 44.6 %
1 SF 190728 (half time) : 3461 500 (+155,=276,- 69), 58.6 %
2 Komodo 13.01 : 3400 500 (+ 69,=276,-155), 41.4 % (Elo-spreading: 61 Elo)

6 PawnPlies:
White Wins: 146 (29.2 %), Black Wins: 103 (20.6 %), Draws: 251 (50.2 %), White Score: 54.3 %, Black Score: 45.7 %
1 SF 190728 (half time) : 3470 500 (+174,=251,- 75), 59.9 %
2 Komodo 13.01 : 3400 500 (+ 75,=251,-174), 40.1 % (Elo-spreading: 70 Elo)


Armageddon openings
Level 1 (SALC)

SALC_Armageddon_IM_4moves:
White Wins: 294 (58.8 %), Black Wins: 206 (41.2 %), Draws: 0 (0.0 %), White Score: 58.8 %, Black Score: 41.2 %
1 SF 190728 (half time) : 3504 500 (+322,= 0,-178), 64.4 %
2 Komodo 13.01 : 3400 500 (+178,= 0,-322), 35.6 % (Elo-spreading: 104 Elo)

SALC_Armageddon_SuperGM_4moves:
White Wins: 285 (57.0 %), Black Wins: 215 (43.0 %), Draws: 0 (0.0 %), White Score: 57.0 %, Black Score: 43.0 %
1 SF 190728 (half time) : 3527 500 (+337,= 0,-163), 67.4 %
2 Komodo 13.01 : 3400 500 (+163,= 0,-337), 32.6 % (Elo-spreading: 127 Elo)

SALC_Armageddon_FEOBOS:
White Wins: 263 (52.6 %), Black Wins: 237 (47.4 %), Draws: 0 (0.0 %), White Score: 52.6 %, Black Score: 47.4 %
1 SF 190728 (half time) : 3481 500 (+307,= 0,-193), 61.4 %
2 Komodo 13.01 : 3400 500 (+193,= 0,-307), 38.6 % (Elo-spreading: 81 Elo)

SALC_Armageddon_6pawnplies:
White Wins: 221 (44.2 %), Black Wins: 279 (55.8 %), Draws: 0 (0.0 %), White Score: 44.2 %, Black Score: 55.8 %
1 SF 190728 (half time) : 3515 500 (+329,= 0,-171), 65.8 %
2 Komodo 13.01 : 3400 500 (+171,= 0,-329), 34.2 % (Elo-spreading: 115 Elo)

Level 2 (NBC)

NBC_Armageddon_IM_4moves:
White Wins: 296 (59.2 %), Black Wins: 204 (40.8 %), Draws: 0 (0.0 %) White Score: 59.2 %, Black Score: 40.8 %
1 SF 190728 (half time) : 3477 500 (+304,= 0,-196), 60.8 %
2 Komodo 13.01 : 3400 500 (+196,= 0,-304), 39.2 % (Elo-spreading: 77 Elo)

NBC_Armageddon_SuperGM_4moves:
White Wins: 284 (56.8 %), Black Wins: 216 (43.2 %), Draws: 0 (0.0 %), White Score: 56.8 %, Black Score: 43.2 %
1 SF 190728 (half time) : 3451 500 (+286,= 0,-214), 57.2 %
2 Komodo 13.01 : 3400 500 (+214,= 0,-286), 42.8 % (Elo-spreading: 51 Elo)

NBC_Armageddon_FEOBOS:
White Wins: 287 (57.4 %), Black Wins: 213 (42.6 %), Draws: 0 (0.0 %)m, White Score: 57.4 %, Black Score: 42.6 %
1 SF 190728 (half time) : 3490 500 (+313,= 0,-187), 62.6 %
2 Komodo 13.01 : 3400 500 (+187,= 0,-313), 37.4 % (Elo-spreading: 90 Elo)

NBC_Armageddon_6pawnplies:
White Wins: 277 (55.4 %), Black Wins: 223 (44.6 %), Draws: 0 (0.0 %), White Score: 55.4 %, Black Score: 44.6 %
1 SF 190728 (half time) : 3473 500 (+301,= 0,-199), 60.2 %
2 Komodo 13.01 : 3400 500 (+199,= 0,-301), 39.8 % (Elo-spreading: 73 Elo)

Level 3 (PawnPlus)

PawnPlus_Armageddon_IM_4moves:
White Wins: 300 (60.0 %), Black Wins: 200 (40.0 %), Draws: 0 (0.0 %), White Score: 60.0 %, Black Score: 40.0 %
1 SF 190728 (half time) : 3454 500 (+288,= 0,-212), 57.6 %
2 Komodo 13.01 : 3400 500 (+212,= 0,-288), 42.4 % (Elo-spreading: 54 Elo)

PawnPlus_Armageddon_SuperGM_4moves:
White Wins: 292 (58.4 %), Black Wins: 208 (41.6 %), Draws: 0 (0.0 %), White Score: 58.4 %, Black Score: 41.6 %
1 SF 190728 (half time) : 3483 500 (+308,= 0,-192), 61.6 %
2 Komodo 13.01 : 3400 500 (+192,= 0,-308), 38.4 % (Elo-spreading: 83 Elo)

PawnPlus_Armageddon_FEOBOS:
White Wins: 301 (60.2 %), Black Wins: 199 (39.8 %), Draws: 0 (0.0 %), White Score: 60.2 %, Black Score: 39.8 %
1 SF 190728 (half time) : 3478 500 (+305,= 0,-195), 61.0 %
2 Komodo 13.01 : 3400 500 (+195,= 0,-305), 39.0 % (Elo-spreading: 78 Elo)

PawnPlus_Armageddon_6pawnplies:
White Wins: 296 (59.2 %), Black Wins: 204 (40.8 %), Draws: 0 (0.0 %), White Score: 59.2 %, Black Score: 40.8 %
1 SF 190728 (half time) : 3492 500 (+314,= 0,-186), 62.8 %
2 Komodo 13.01 : 3400 500 (+186,= 0,-314), 37.2 % (Elo-spreading: 92 Elo)

*************************************************************

Test-environment for 4 long thinking-time testruns with the IM_4moves-sets, 7x more time, than the testruns above:
Stockfish 420''+3500ms (= 7'+3.5''), Komodo 840''+7000ms (= 14'+7''). Average game-duration: 35 minutes
Each testrun: 500 games (4 testruns = 2000 games), cutechess-cli, Errorbar of each run around +/-11 Elo

IM_4moves (default, for comparsion):
White Wins: 119 (23.8 %), Black Wins: 57 (11.4 %), Draws: 324 (64.8 %), White Score: 56.2 %, Black Score: 43.8 %
1 SF 190728 (half time) : 3483 500 (+146,=324,- 30), 61.6 %
2 Komodo 13.01 : 3400 500 (+ 30,=324,-146), 38.4 % (Elo-spreading: 83 Elo)

SALC_Armageddon_IM_4moves (Level 1):
White Wins: 274 (54.8 %), Black Wins: 226 (45.2 %), Draws: 0 (0.0 %), White Score: 54.8 %, Black Score: 45.2 %
1 SF 190728 (half time) : 3526 500 (+336,= 0,-164), 67.2 %
2 Komodo 13.01 : 3400 500 (+164,= 0,-336), 32.8 % (Elo-spreading: 126 Elo)

NBC_Armageddon_IM_4moves (Level 2):
White Wins: 295 (59.0 %), Black Wins: 205 (41.0 %), Draws: 0 (0.0 %), White Score: 59.0 %, Black Score: 41.0 %
1 SF 190728 (half time) : 3502 500 (+321,= 0,-179), 64.2 %
2 Komodo 13.01 : 3400 500 (+179,= 0,-321), 35.8 % (Elo-spreading: 102 Elo)

PawnPlus_Armageddon_IM_4moves (Level 3):
White Wins: 310 (62.0 %), Black Wins: 190 (38.0 %), Draws: 0 (0.0 %), White Score: 62.0 %, Black Score: 38.0 %
1 SF 190728 (half time) : 3519 500 (+332,= 0,-168), 66.4 %
2 Komodo 13.01 : 3400 500 (+168,= 0,-332), 33.6 % (Elo-spreading: 119 Elo)



Base idea of Armageddon openings: Larry Kaufman (Komodo-Team)
Developed and constructed by Stefan Pohl (SPCC), except analyzing FEOBOS-endpositons: by Hauke Lutz

(C) FEOBOS openings: Frank Quisinsky and Klaus Wlotzka
(C) 6-pawn-plies openings: Hauke Lutz

(C) Armageddon Openings: Stefan Pohl (SPCC) 2019

[Nordlandia]

I discovered this position couple of years ago.

For two pawns white got open files for the rooks and black can't castle either sides.

[d]rnbqkbnr/pppppppp/8/8/8/8/1PPPPPP1/RNBQKBNR w KQ - 0 1

[lkaufman]

I was pleasantly surprised to see how you developed my idea, I know you put a lot of time into it. Some comments:
1. The comments about draws increasing with more time and faster hardware are accurate for normal chess, where the proper result is almost certainly a draw. However if you start the game with one side having a theoretically won position, then the opposite is true; the draw percentage should drop with more time and faster hardware. Since we are aiming to start the game with a position right on the edge of a win or a draw, it is not at all clear whether the draw percentage should go up or down with more time. Based on your statistics, the Black percentage (which is mostly draws) goes down with more time for Level 1, but goes up with more time for Level 2 and 3. This suggests (but does not prove) that my first idea (Level 1) is theoretically drawn (counting as a Black win), but that my second idea (Level 2) is theoretically won for White. The pawn up idea is almost certainly won for White and the stats are too favorable for White to be very useful, in my opinion, as I expected. Perhaps it could be made more equal by requiring White to open 1.a3, which will usually have to be played at some point anyway to free the rook. Then book moves would be played with colors reversed, as Black is playing first except with the White pawn on a3.
2. I was rather surprised to see White clearly ahead with the I.M. and G.M. books even with my first (Level 1) idea, since my own tests using Lc0 with no book moves suggested a slight Black edge. But the explanation is simple; in normal openings Black aims to castle short quickly, but if this is illegal then he is making inferior opening moves; he should be aiming to castle long right from the start. I'm not yet sufficiently familiar with your other opening books to comment on them, I'll check them out. Theoretically the best way to handle this problem would be to have a bunch of top engines play a massive round robin from the start position, with MP and/or variable time to produce variety, and then pick out as many differing opening sequences as needed. But I know this would be a lot of work, and getting enough variety might require some trial and error.
3. So far your data suggests that my initial idea (Level 1) might be more balanced than Level 2, but in view of the above comments that isn't clear yet.
Anyway, all very interesting with lots to investigate!

[lkaufman]

I reviewed the openings in the various books you use, and some things are becoming clearer. The FEOBOS book is too deep for this purpose, almost every move played with nonstandard castling rights is likely to be inferior, so the lines should be as short as possible. The superGM and IM books are short enough, but have a large White bias since far more normal Black moves are wrong than White moves with the SALC restriction. The 6pawnplies book, which seems rather pointless to me for normal chess, may be the best one for this purpose, because there is no obvious bias in favor of moves that help with kingside castling. Aside from the question of which book is best for engine testing, I think that only the 6pawnplies book gives us reliable stats regarding whether White or Black has the advantage with these rules. It shows that Level 1 (SALC) clearly favors Black, but Level 2 (NBC) favors White by a much larger margin. So the evidence does indicate that my first idea, "SALC", is probably the best one, and it is not clear which side is theoretically winning or which side benefits more from having more time. Perhaps you could run your longer time control test again on Level 1 with the 6pawnplies book to try to get an answer to this important question.

[pohl4711]

I think, your focus is too much on the whitescore/blackscore. For engine-testings and tournaments, the Elo-speading is much more important. Because, the wider the Elo-spreading, the lower the number of games is, which have to be played in order to get results out of the errorbar (and get a statistically reliable result) and get a reliable ranking of engines.
And for answering the question, which Armageddon-Level is best or which is working and which is not working, the Elo-spreading is the parameter of interest (or should be...), not only the whitescore.
Because, if the whitescore gets too high in Armageddon, the Elo-spreading gets (of course) lower, because in my testings, the stronger engine (Stockfish) and the weaker engine (Komodo) play the same amount of games with white and with black. Think of a whitescore of 100%. That would mean (of course) a Elo-spreading of 0 (because engine head-to-head points are 50%-50%).
So, lets take a closer look at the Elo-spreading:

(IM_4moves:)
Short time:
SALC (Level 1): 104 Elo
NBC (Level 2): 77 Elo
PawnPlus (Level 3): 54 Elo

Long time:
SALC (Level 1): 126 Elo
NBC (Level 2): 102 Elo
PawnPlus (Level 3): 119 Elo

What we see, is, that the Elo-spreading improves with more thinkingtime on all 3 levels - very nice!
But, what is much more interesting is, that from short to long time, the increase of the Elo-spreading is +22 on Level 1, +25 on Level 2, but +65 (!) on Level 3. And that is exactly, what I have expected: The difference between Level 1 and 2 is very small, because these both Levels are quite similar. And Level 3 improves its result-quality (and that means Elo-spreading, not whitescore!) more than the other levels with more thinkingtime. So, it is clear to me, that your opinion "The pawn up idea is almost certainly won for White and the stats are too favorable for White to be very useful", is definitly wrong. The tests are showing, that the main idea of my 3-level-concept is correct:
With short thinkingtime (or not so beefy harware), the Level 1 (SALC) gives the best results (= the widest Elo-spreading (104 Elo)) and Level 3 the worst results (=lowest Elo spreading (54 Elo)), but the longer the thinkingtime gets (or the faster the hardware gets), the better get the results of Level 3. So, for the future, Level 3 (PawnPlus) is very promising. And that was the goal of development.

Regard - Stefan (SPCC)

[lkaufman]

I think, your focus is too much on the whitescore/blackscore. For engine-testings and tournaments, the Elo-speading is much more important. Because, the wider the Elo-spreading, the lower the number of games is, which have to be played in order to get results out of the errorbar (and get a statistically reliable result) and get a reliable ranking of engines.
And for answering the question, which Armageddon-Level is best or which is working and which is not working, the Elo-spreading is the parameter of interest (or should be...), not only the whitescore.
Because, if the whitescore gets too high in Armageddon, the Elo-spreading gets (of course) lower, because in my testings, the stronger engine (Stockfish) and the weaker engine (Komodo) play the same amount of games with white and with black. Think of a whitescore of 100%. That would mean (of course) a Elo-spreading of 0 (because engine head-to-head points are 50%-50%).
So, lets take a closer look at the Elo-spreading:

(IM_4moves:)
Short time:
SALC (Level 1): 104 Elo
NBC (Level 2): 77 Elo
PawnPlus (Level 3): 54 Elo

Long time:
SALC (Level 1): 126 Elo
NBC (Level 2): 102 Elo
PawnPlus (Level 3): 119 Elo

What we see, is, that the Elo-spreading improves with more thinkingtime on all 3 levels - very nice!
But, what is much more interesting is, that from short to long time, the increase of the Elo-spreading is +22 on Level 1, +25 on Level 2, but +65 (!) on Level 3. And that is exactly, what I have expected: The difference between Level 1 and 2 is very small, because these both Levels are quite similar. And Level 3 improves its result-quality (and that means Elo-spreading, not whitescore!) more than the other levels with more thinkingtime. So, it is clear to me, that your opinion "The pawn up idea is almost certainly won for White and the stats are too favorable for White to be very useful", is definitly wrong. The tests are showing, that the main idea of my 3-level-concept is correct:
With short thinkingtime (or not so beefy harware), the Level 1 (SALC) gives the best results (= the widest Elo-spreading (104 Elo)) and Level 3 the worst results (=lowest Elo spreading (54 Elo)), but the longer the thinkingtime gets (or the faster the hardware gets), the better get the results of Level 3. So, for the future, Level 3 (PawnPlus) is very promising. And that was the goal of development.

Regard - Stefan (SPCC)

The increase in the elo-spread for PawnPlus with long time is indeed impressive, although the actual number is still lower than the SALC number. If we make the speculative assumption that the trend will continue with even more time, then yes, PawnPlus will show better resolution with more time. But I am thinking more about the chess factors than just the math. For the Armageddon Openings to be used in actual tournaments (not just private testing) the games need to be reasonably close to even for good spectator interest. The PawnPlus showed White winning 60% at fast level and 62% at long level, both of which I consider to be too unequal for a sporting competition. Moreover the trend is bad, there is no reason to think that the game is not won for White with proper play, though this can't be proved. I think that almost any GM would agree with me that with both the first move initiative and an extra pawn, White must be theoretically winning. If we want the results to have as much as possible in common with normal chess, it is better that the initial position of Armageddon be theoretically drawn (since normal chess presumably is), but close to the winning line so as to make Armageddon workable and the Elo spread high. It appears to me that SALC is probably not quite over the win/draw line, which I would say is ideal. This is also critical if the idea is to be possibly used in human tournaments, either as tiebreakers or even as the main event. If the initial position is too favorable for either side, it just won't work.
It's also not at all obvious to me why PawnPlus shows so much more elo spread with more time. It may be something very specific to the two engines in question. Converting an extra pawn to a win and converting a safer king (typical of SALC) to a win may be very different and that could explain the result. Anyway, my main point is not that your pawn plus idea is bad, just that the exact choice you made is too clearly on the winning side to be typical of normal chess. If you could get good elo-spreading with a modified pawnplus that gave close to 50-50 results I would be all for it. Anyway I'm happy with the SALC 6pawnplies book and plan to try it out with Komodo soon.

[pohl4711]

The increase in the elo-spread for PawnPlus with long time is indeed impressive, although the actual number is still lower than the SALC number. If we make the speculative assumption that the trend will continue with even more time, then yes, PawnPlus will show better resolution with more time. But I am thinking more about the chess factors than just the math. For the Armageddon Openings to be used in actual tournaments (not just private testing) the games need to be reasonably close to even for good spectator interest. The PawnPlus showed White winning 60% at fast level and 62% at long level, both of which I consider to be too unequal for a sporting competition. Moreover the trend is bad, there is no reason to think that the game is not won for White with proper play, though this can't be proved. I think that almost any GM would agree with me that with both the first move initiative and an extra pawn, White must be theoretically winning. If we want the results to have as much as possible in common with normal chess, it is better that the initial position of Armageddon be theoretically drawn (since normal chess presumably is), but close to the winning line so as to make Armageddon workable and the Elo spread high. It appears to me that SALC is probably not quite over the win/draw line, which I would say is ideal. This is also critical if the idea is to be possibly used in human tournaments, either as tiebreakers or even as the main event. If the initial position is too favorable for either side, it just won't work.
It's also not at all obvious to me why PawnPlus shows so much more elo spread with more time. It may be something very specific to the two engines in question. Converting an extra pawn to a win and converting a safer king (typical of SALC) to a win may be very different and that could explain the result. Anyway, my main point is not that your pawn plus idea is bad, just that the exact choice you made is too clearly on the winning side to be typical of normal chess. If you could get good elo-spreading with a modified pawnplus that gave close to 50-50 results I would be all for it. Anyway I'm happy with the SALC 6pawnplies book and plan to try it out with Komodo soon.

All what you say could be true. At the moment, the Level 1 (SALC) is definitly the best solution. Nobody can really be sure about the future development of hardware and engines (some years ago, nobody saw Lc0 coming...). So nobody can say, if Level 2 or 3 will work better than Level 1 in the future or not. And nobody can say, if engines will be so strong, that a theoretically won position will always lead to a win (the opponent will also be strong).

When I am doing so much work on new opening-sets, I want to build openings, too, which I believe to work properly in the future. The Armageddon Openings contain 12 different opening sets, so I hope, some of them will work in the future, too. But nobody can be sure about that.

It would be nice, if you share your testings with my openings with us (me? -> PM).

Regards - Stefan

[lkaufman]

The increase in the elo-spread for PawnPlus with long time is indeed impressive, although the actual number is still lower than the SALC number. If we make the speculative assumption that the trend will continue with even more time, then yes, PawnPlus will show better resolution with more time. But I am thinking more about the chess factors than just the math. For the Armageddon Openings to be used in actual tournaments (not just private testing) the games need to be reasonably close to even for good spectator interest. The PawnPlus showed White winning 60% at fast level and 62% at long level, both of which I consider to be too unequal for a sporting competition. Moreover the trend is bad, there is no reason to think that the game is not won for White with proper play, though this can't be proved. I think that almost any GM would agree with me that with both the first move initiative and an extra pawn, White must be theoretically winning. If we want the results to have as much as possible in common with normal chess, it is better that the initial position of Armageddon be theoretically drawn (since normal chess presumably is), but close to the winning line so as to make Armageddon workable and the Elo spread high. It appears to me that SALC is probably not quite over the win/draw line, which I would say is ideal. This is also critical if the idea is to be possibly used in human tournaments, either as tiebreakers or even as the main event. If the initial position is too favorable for either side, it just won't work.
It's also not at all obvious to me why PawnPlus shows so much more elo spread with more time. It may be something very specific to the two engines in question. Converting an extra pawn to a win and converting a safer king (typical of SALC) to a win may be very different and that could explain the result. Anyway, my main point is not that your pawn plus idea is bad, just that the exact choice you made is too clearly on the winning side to be typical of normal chess. If you could get good elo-spreading with a modified pawnplus that gave close to 50-50 results I would be all for it. Anyway I'm happy with the SALC 6pawnplies book and plan to try it out with Komodo soon.

All what you say could be true. At the moment, the Level 1 (SALC) is definitly the best solution. Nobody can really be sure about the future development of hardware and engines (some years ago, nobody saw Lc0 coming...). So nobody can say, if Level 2 or 3 will work better than Level 1 in the future or not. And nobody can say, if engines will be so strong, that a theoretically won position will always lead to a win (the opponent will also be strong).

When I am doing so much work on new opening-sets, I want to build openings, too, which I believe to work properly in the future. The Armageddon Openings contain 12 different opening sets, so I hope, some of them will work in the future, too. But nobody can be sure about that.

It would be nice, if you share your testings with my openings with us (me? -> PM).

Regards - Stefan

Well, we can be pretty sure that if the initial position used is not objectively winning, but close to it, then it should work much better in the future for Armageddon than normal chess with time adjustment, since the value of time adjustment fades as the play approaches perfection. Perhaps some deep analysis on the SALC start position with the top engines might shed some light on whether the White position is likely winning or not. One problem for testing is that current engines don't recognize the armageddon rule, so it's a bit "wrong" to test them with that rule without making any changes. But probably testing will still work well, especially at longer time limits where the draw problem is normally severe.

[pohl4711]

Well, we can be pretty sure that if the initial position used is not objectively winning, but close to it, then it should work much better in the future for Armageddon than normal chess with time adjustment, since the value of time adjustment fades as the play approaches perfection. Perhaps some deep analysis on the SALC start position with the top engines might shed some light on whether the White position is likely winning or not. One problem for testing is that current engines don't recognize the armageddon rule, so it's a bit "wrong" to test them with that rule without making any changes. But probably testing will still work well, especially at longer time limits where the draw problem is normally severe.

It would be easy, to include a uci-paramater in a chess-engine: Armageddon true/false, which sets the value of a draw to win for black internally...(from 0.00 to -99.99 or something like that).
But even without such a parameter, the engines should see a clear advantage for white, when they start playing out of my Armageddon openings. So, this is a high contempt for white (and a negative contempt for black), which makes black trying to reach a draw and white trying to avoid it. So, I believe, games with Armageddon openings are fine, even when the engines not know, that they are playing Armageddon...

[lkaufman]

Well, we can be pretty sure that if the initial position used is not objectively winning, but close to it, then it should work much better in the future for Armageddon than normal chess with time adjustment, since the value of time adjustment fades as the play approaches perfection. Perhaps some deep analysis on the SALC start position with the top engines might shed some light on whether the White position is likely winning or not. One problem for testing is that current engines don't recognize the armageddon rule, so it's a bit "wrong" to test them with that rule without making any changes. But probably testing will still work well, especially at longer time limits where the draw problem is normally severe.

It would be easy, to include a uci-paramater in a chess-engine: Armageddon true/false, which sets the value of a draw to win for black internally...(from 0.00 to -99.99 or something like that).
But even without such a parameter, the engines should see a clear advantage for white, when they start playing out of my Armageddon openings. So, this is a high contempt for white (and a negative contempt for black), which makes black trying to reach a draw and white trying to avoid it. So, I believe, games with Armageddon openings are fine, even when the engines not know, that they are playing Armageddon...

Yes, most of the time the Armageddon openings will work well without telling the engines the rules, but "most of the time" is not ideal, so we should really do as you say with the UCI-parameter if this turns out to be really worthwhile for testing. Even so, a lot of the chess knowledge will be wrong, for example bishops of opposite color endgames won't be "drawish", but usually lost for White. Humans will adapt to the rule much better than engines will, unless programmers put in the long hours needed to teach the engine how to modify strategy. Actually I think that setting Contempt to a reasonably high value for White and a high negative value for Black will work better than the UCI option. It is vital that even if the opening edge vanishes, White should still try to avoid trades and other drawish ideas as he will lose draws. It may not even require any programming for Komodo, I just have to verify that "White Contempt" works properly in actual game play. Of course it would require some testing to determine the proper value for Contempt.