Hello! I'd like to say that in general the training feature is nice, and I do acknowledge that it's problematic to gather puzzles automatically so that they would be always pleasing to human players.
However, I'd still like to discuss about opportunities for improvements!
1) The problem of multiple solutions
Example from puzzle 47008 (en.lichess.org/training/47008): 3r4/2k5/4Q2p/1p3pp1/2pP2P1/2q2b1P/B4P2/R4KR1 b - - 7 35
My idea was Qd3+ followed by Qxd4, when the double threat of Qd1+ mating and Qxa1+ leaves white helpless. The problem is that white can give up his queen and the forced mate is two moves slower than with the faster Qxa1+ Qe1 Qxe1+ Kxe1 Rxd4.
A while ago there was a similar situation without a forced checkmate, where I saw two winning lines and picked the one that was +5.9 instead of +6.5 and lost the puzzle. Unfortunately I can't find that puzzle anymore.
It is of course debatable whether it really is failing not to play the engines top choice, or the absolutely fastest mate, but I would enjoy the puzzles more if winning would be enough.
Suggested improvements:
All clearly winning moves should be accepted. It is up for debate what computer eval corresponds to clearly winning, but it shouldn't be impossible. Maybe the "Good move, but there's a better one" -feature could be expanded?
All mating sequences that don't repeat and transpose to another line should be accepted. It happens too often that an easy mate is not accepted because it can be delayed a couple moves with computer desperation.
2) Puzzles where everything wins
This is maybe a smaller issue for me, but there are some puzzles (like this: en.lichess.org/training/4204) where the most obvious moves just win without any hustle. Essentially they are "White to move and mate in four moves" or similar, but just doesn't specify it. I think they should, or they should be excluded. This might be just me though, as I tend to lose motivation if I have to calculate unnecessary tactics.
Cheers!
However, I'd still like to discuss about opportunities for improvements!
1) The problem of multiple solutions
Example from puzzle 47008 (en.lichess.org/training/47008): 3r4/2k5/4Q2p/1p3pp1/2pP2P1/2q2b1P/B4P2/R4KR1 b - - 7 35
My idea was Qd3+ followed by Qxd4, when the double threat of Qd1+ mating and Qxa1+ leaves white helpless. The problem is that white can give up his queen and the forced mate is two moves slower than with the faster Qxa1+ Qe1 Qxe1+ Kxe1 Rxd4.
A while ago there was a similar situation without a forced checkmate, where I saw two winning lines and picked the one that was +5.9 instead of +6.5 and lost the puzzle. Unfortunately I can't find that puzzle anymore.
It is of course debatable whether it really is failing not to play the engines top choice, or the absolutely fastest mate, but I would enjoy the puzzles more if winning would be enough.
Suggested improvements:
All clearly winning moves should be accepted. It is up for debate what computer eval corresponds to clearly winning, but it shouldn't be impossible. Maybe the "Good move, but there's a better one" -feature could be expanded?
All mating sequences that don't repeat and transpose to another line should be accepted. It happens too often that an easy mate is not accepted because it can be delayed a couple moves with computer desperation.
2) Puzzles where everything wins
This is maybe a smaller issue for me, but there are some puzzles (like this: en.lichess.org/training/4204) where the most obvious moves just win without any hustle. Essentially they are "White to move and mate in four moves" or similar, but just doesn't specify it. I think they should, or they should be excluded. This might be just me though, as I tend to lose motivation if I have to calculate unnecessary tactics.
Cheers!