Blackmageddon Openings released

[lkaufman]

2. Some openings don't make sense with a2 missing. For example, you include the Ruy Lopez (Spanish) with 3...a6? 4.Ba4? but both of those moves are silly with a2 missing, since a6 is pinned.

Maybe these things aren't so important for the purpose of using it just to test engines, but I don't think that the missing a2 Armageddon game is a playable version for actual tournaments (whether human or engine) as it is too favorable for White without forcing inferior openings on White.

And what do you think, how many opening-lines are nonsense, if black is not allowed to castle???
A King on e-line should make a lot more lines wrong, than a missing pawn on a2. In most openings, a castled black king is an important part of the opening.

In Blackmageddon Komodo checked all endpositions of the lines, so that all these endpositions are playable.

What was your definition of "playable", given that these are armageddon games? I suppose something like a -0.9 to -0.5 eval perhaps? This would explain why the lines mostly feature bad white opening play, because with good White opening play the eval will be too close to even.
Regarding no-castle, yes, the normal openings are nonsense, which is why Kai made a book based on playing with these rules, not based on normal chess openings.

Perhaps a look in rhe pgn would help. The Komodo Eval and the Eval-limits are in the TAGs.
I analyzed more than 40000 lines. And then splitted the lines in 10 centipawn-eval Intervals from 0 to 120 (0-10, 11-20, 21-30 and so on).. And the lines, that were finally chosen, were from the middle of that and there (in the middle) were the most lines. So the chosen lines are most “normal“ lines out of the 40000. In the opinion of Komodo.
Look, how small the eval-interval of the 10k pgn file is. And still 10000 lines out of little bit more than 40000: nearly 25% of all openings in such a small interval.

OK, now I understand, you chose openings that resulted in a score of -.59 or -.60 for the small set, and as close as possible to that range for the larger sets. This teaches us something very interesting; since you got just a slight plus score for White with the Armageddon rule, a slightly worse initial score, perhaps -.62 or -.63, would have been perfectly balanced. I had normally estimated that the break-even score for Armageddon was 0.70 for Komodo (or -.70 with White winning the draws), so based on your data I should reduce that by about 10%, although we only know this definitively for the specific case of a rook's pawn advantage for one side. So thanks for that info! My main issue with "Blackmageddon" is that the initial position with a2 gone is not nearly a .62 advantage for Black, so White has to play some poor moves to get to -.60 (or -.62 ideally). If we're going to have to play some poor moves for one side anyway, why not just use normal chess and only choose Black openings that give White about a 0.60 (or 0.62) advantage? These can be found in actual master games, though not generally in grandmaster games. Wouldn't this be much more like normal chess than "Blackmageddon"? I actually made such a book myself for testing a while ago, but it has just 25 openings so not much use for serious testing, just for getting a quick idea about some new engine. But aside from that I imagine that Blackmageddon will be useful for engine vs. engine testing, even though the start position is unsuitable for engine events that people would want to watch, as it is too favorable for White.

[pohl4711]

The Komodo Eval and the Eval-limits are in the TAGs.

Did Komodo know that a draw would count as a white win? Otherwise, those evals can't be used because black would be fine with a 0.00 because all the other moves were in white's favor, except if the 0.00 leads to a white draw wins the eval makes no sense.

I.e. as black Komodo would choose a 0.00 forced draw (with black losing) over a 0.01 move because it doesn't know it's not chess.

All I can say is, that all my testings (more than 4000 games) show, that Blackmageddon works fine with engines, which dont know, that they play Blackmageddon.
Alternativly you can use my advanced scoring system/tool in which a white-win is 2 points for white, a draw is 1 point for white and a win for black is a point for black.

[pohl4711]

So thanks for that info! My main issue with "Blackmageddon" is that the initial position with a2 gone is not nearly a .62 advantage for Black, so White has to play some poor moves to get to -.60 (or -.62 ideally). If we're going to have to play some poor moves for one side anyway, why not just use normal chess and only choose Black openings that give White about a 0.60 (or 0.62) advantage? These can be found in actual master games, though not generally in grandmaster games. Wouldn't this be much more like normal chess than "Blackmageddon"? I actually made such a book myself for testing a while ago, but it has just 25 openings so not much use for serious testing, just for getting a quick idea about some new engine. But aside from that I imagine that Blackmageddon will be useful for engine vs. engine testing, even though the start position is unsuitable for engine events that people would want to watch, as it is too favorable for White.

I had this idea too. But the problem is, that each engine has its own eval. Stockfish eval for example differs very much from Komodo's eval. So, I believe, a material advantage is a better way. I used the Komodo-eval not as an absolute value - i chose Komodo-evals around -0.6, just because I had the best whitescore/blackscore balance in my alpha-testings. Another engine would mean complete other values of eval could or should work.
So, the material advantage works for each engine, because "a pawn is pawn, no matter, how an engine evaluates it".
Next problem is, without a pawn more for white, human-openings (played on a good human Elo-level) with an Komodo eval of around -0.60 are very rare - why should white play so bad moves... So, it would be a big problem, to get enough of these openings for a big openings-set. And these openings would be very strange, I believe. With a pawn more and "normal" openings pasted, white have not to play so bad (only "suboptimal"). And so, we got much more openings with a Komodo-Eval of -0.60. And more "not too strange" openings.
My 2cents.

[lkaufman]

So thanks for that info! My main issue with "Blackmageddon" is that the initial position with a2 gone is not nearly a .62 advantage for Black, so White has to play some poor moves to get to -.60 (or -.62 ideally). If we're going to have to play some poor moves for one side anyway, why not just use normal chess and only choose Black openings that give White about a 0.60 (or 0.62) advantage? These can be found in actual master games, though not generally in grandmaster games. Wouldn't this be much more like normal chess than "Blackmageddon"? I actually made such a book myself for testing a while ago, but it has just 25 openings so not much use for serious testing, just for getting a quick idea about some new engine. But aside from that I imagine that Blackmageddon will be useful for engine vs. engine testing, even though the start position is unsuitable for engine events that people would want to watch, as it is too favorable for White.

I had this idea too. But the problem is, that each engine has its own eval. Stockfish eval for example differs very much from Komodo's eval. So, I believe, a material advantage is a better way. I used the Komodo-eval not as an absolute value - i chose Komodo-evals around -0.6, just because I had the best whitescore/blackscore balance in my alpha-testings. Another engine would mean complete other values of eval could or should work.
So, the material advantage works for each engine, because "a pawn is pawn, no matter, how an engine evaluates it".
Next problem is, without a pawn more for white, human-openings (played on a good human Elo-level) with an Komodo eval of around -0.60 are very rare - why should white play so bad moves... So, it would be a big problem, to get enough of these openings for a big openings-set. And these openings would be very strange, I believe. With a pawn more and "normal" openings pasted, white have not to play so bad (only "suboptimal"). And so, we got much more openings with a Komodo-Eval of -0.60. And more "not too strange" openings.
My 2cents.

No, you misunderstand. I propose looking for normal openings with a PLUS 0.60 score for White, with Black getting the draw odds, as in normal Armageddon games. These are not nearly so rare as -.60 openings, and I think you would find enough such openings in games by players with ratings in the 2100 to 2400 range, especially older games before everyone checked engine scores. Lots of dubious Black gambits and systems that don't fight for space will qualify. Black only needs to make one or two dubious moves in the opening to be near the draw/loss line. I don't think that the difference between SF eval and Komodo eval matters so much, it's just that SF evals are roughly 4/3 as large, so 0.6 for Komodo is like 0.8 for Stockfish. Anyway all that matters is that you pick a value that results in roughly a 50% score with the Armageddon rule.

[pohl4711]

So thanks for that info! My main issue with "Blackmageddon" is that the initial position with a2 gone is not nearly a .62 advantage for Black, so White has to play some poor moves to get to -.60 (or -.62 ideally). If we're going to have to play some poor moves for one side anyway, why not just use normal chess and only choose Black openings that give White about a 0.60 (or 0.62) advantage? These can be found in actual master games, though not generally in grandmaster games. Wouldn't this be much more like normal chess than "Blackmageddon"? I actually made such a book myself for testing a while ago, but it has just 25 openings so not much use for serious testing, just for getting a quick idea about some new engine. But aside from that I imagine that Blackmageddon will be useful for engine vs. engine testing, even though the start position is unsuitable for engine events that people would want to watch, as it is too favorable for White.

I had this idea too. But the problem is, that each engine has its own eval. Stockfish eval for example differs very much from Komodo's eval. So, I believe, a material advantage is a better way. I used the Komodo-eval not as an absolute value - i chose Komodo-evals around -0.6, just because I had the best whitescore/blackscore balance in my alpha-testings. Another engine would mean complete other values of eval could or should work.
So, the material advantage works for each engine, because "a pawn is pawn, no matter, how an engine evaluates it".
Next problem is, without a pawn more for white, human-openings (played on a good human Elo-level) with an Komodo eval of around -0.60 are very rare - why should white play so bad moves... So, it would be a big problem, to get enough of these openings for a big openings-set. And these openings would be very strange, I believe. With a pawn more and "normal" openings pasted, white have not to play so bad (only "suboptimal"). And so, we got much more openings with a Komodo-Eval of -0.60. And more "not too strange" openings.
My 2cents.

No, you misunderstand. I propose looking for normal openings with a PLUS 0.60 score for White, with Black getting the draw odds, as in normal Armageddon games. These are not nearly so rare as -.60 openings, and I think you would find enough such openings in games by players with ratings in the 2100 to 2400 range, especially older games before everyone checked engine scores. Lots of dubious Black gambits and systems that don't fight for space will qualify. Black only needs to make one or two dubious moves in the opening to be near the draw/loss line. I don't think that the difference between SF eval and Komodo eval matters so much, it's just that SF evals are roughly 4/3 as large, so 0.6 for Komodo is like 0.8 for Stockfish. Anyway all that matters is that you pick a value that results in roughly a 50% score with the Armageddon rule.

OK. That could be a possible approach. But I doubt, that there will be many lines. Especially, when they are so short as they are in Blackmageddon (10 plies). Perhaps they had to be much longer for a high number of lines. And for this, we would have to evaluate and filter a lot, lot, lot of lines. That could take a lot of time...