#26 #27 #28 I'm slowly making progress on atomic syzygybases:
github.com/ddugovic/tb

My plan is that after 6-man bases are generated, then have Stockfish use them; and only then start defining other endgame evaluations such as you discuss. (And it's OK if Stockfish has a nonzero evaluation for a position drawn with perfect play, if the position offers some winning chances. I know that isn't the case in #28 but in other endgames it may be the case.)

@31

Nice.

Regarding ending position knowledge with pieces more than 6, I think strong human players know some of these positions that are drawish.

Atomic stockfish analyzer is now very strange.

en.lichess.org/tgwqWpha/white

1. Nf3 f6 2. e3

Look at suggested sequence: 2. ... d5 ?! Blunder, best move was e6

(2... e6 3. b3 Nc6 4. Bb5 d5 5. d4 Nh6 6. c3??? Ng4 7. h3 e5??? 8. dxe5 Nh2)

forgot to mark this:
8. dxe5??? Nh2???

all the moves marked with "???" are BLUNDERS.

6. c3 turns ~draw position to "losing in 2".
7. ... e5 is somewhy played instead of 7. ... Nxf2#
8. dxe5: black didn't explode white king, so let's give a second chance to black to play Nxf2# !
8. ... Nh2: again, black doesn't want to blow up white king. Probably white will give to black a third chance...

WTF?

@ixdon2: This has already been fixed, but has not been merged yet.

@ubdip
Is this really fixed?

Look at this game:
en.lichess.org/hItuGM1d/white

1. Nf3 f6 2. e3 e5?? {Blunder...} 3. Ng5 c6?? { (3.74 → 7.20) Blunder. Best move was Bb4. } (3... Bb4 4. c3 fxg5??? 5. Qh5+ g6 6. Qg4 Kf8 7. Qxd7 Nf6 8. f3 Nc6 9. g4 Rd8 10. d3)

after. 4. ... fxg5 in suggested sequence black gets mated in 3.

@ixdon2: What I meant is that it has been fixed in the repository where it is developed (github.com/ddugovic/Stockfish), but has not been merged yet into the repository that it used for lichess (github.com/niklasf/Stockfish).

@revoof: Since the difference due to the reintroduction of atomic SEE is several hundred Elo, could you merge it soon?

@ubdip: merged & released. lichess cluster will be updated within 12h :)

Thanks @revoof.

Thanks for the help fixing this... just now I've found time to start looking into how to duplicate this, only to discover it's the same issue/fix "Reintroduce atomic SEE" I merged earlier this week.