Puzzle 36287: Broken?

Puzzle 36287: Broken?

http://en.lichess.org/training/36287

After white dodges the check, black makes an odd move that allows a fork check. Without that rook placement (which doesn't seem to have any compelling strategic motive behind it), white can only really gain a pawn and this wouldn't really be worth 1400+. If my assessment is correct, then it's puzzling how this got included as one of the puzzles.

It would be nice if there was a way to specifically flag puzzles for being "broken". The current upvotes/downvotes mechanism just mushes various sentiments (interesting/boring, functional/broken, etc.) into one metric. You can't even make a proper vote when your appraisals aren't in the same vein. (e.g. You might find something boring even if you don't think it is broken or you might like something because it was easy even though it was actually broken.)

P.S. I did search for "36287" before I created this topic and this example hasn't been mentioned anywhere yet.

I wish I had the foresight to use a broader title for this topic. =,=

Anyway, ( http://en.lichess.org/training/44360 ) has a similar issue so I'll just append it here. (I would have EDITed this into the previous post if there was such a feature.)

Black's last move enables a mate-in-3 (which could have been avoided with a different move).

P.S. Since coding a bug report interface into the existing trainer would likely take more effort, maybe lichess can just open a subforum specifically for puzzle issues with the topic titles required to be puzzle id numbers.

If you're unsure about a puzzle continuation, I recommend checking with the analysis tool. After the puzzle is finished, just go back to the starting move, click on "Analyse", enable local computer evaluation and go through the moves.

In the case of puzzle 36287, you will see that Black's ...Rf8 is in fact the best possible reply because all other variations will lead to even bigger material loss (as you can check out in the analysis). The main point is that the rook is undefended on a8 and can be won if White is allowed to take on g7 and then check the king on the 8th rank, whereas in the puzzle continuation the rook is defended by the bishop and knight so that Black "only" loses an exchange.

As for the other puzzle, your point is correct. However, allowing inferior lines to make a puzzle more "understandable" seems to be standard lichess policy, so this puzzle isn't "broken" either, it's just the way lichess puzzles work. (If you're unhappy about that - as I am - you can discuss it in a dedicated post).

@Bonifratz, thanks for your input. :) Coincidentally, I did try the local eval before I created the topic. I don't know if it's because I have a relatively slow, less powerful computer, but I only really get recommended with Re8. It allows the king to hide without losing the rook to a knight exchange. The line that follows is just a pawn gain like I stated. (Do let me know if I'm wrong about this.) =,=

In the case of 44630, local eval recommended Qg5. Being a rook up (as the knight can be defended by that rook after the queen exchange), white has a very good chance of winning. However, it's no longer a quick mate-in-3 as it would be with Kxh8.

Material puzzles end with the computer making its best calculated move. That's why I claimed 44630 to be broken. The computer is supposed to be presenting us with the best defense possible at every point (which is kinda funny how it throws everything sequentially to delay getting checkmated as long as possible) so I considered 36287 to be a malformed puzzle. (Unfortunately, these are pre-computed reactions so even with the server upgrades, we get the same interactions with the puzzler.)

Having an automated opposition makes it convenient, but it does dampen potential lessons, mechanical and mental, that we could be learning. Conventional puzzles required us to think of optimal moves for *both* sides, exploring all relevant branches and lines. On the other hand, older chess puzzle books spoonfed us with explicit goals. (e.g. "white to mate in 3")

Anyway, I am still thankful for lichess having this trainer because I might otherwise not be contemplating (and complaining, haha) on the shortcomings of such a system. ^,^

In regards to 36287: when I let the engine run, it first chooses ...Re8, but quickly changes to ...Rf8. The weird thing is that when I follow the best moves given, the continuation after ...Re8 gives a slightly better evaluation for Black than the other one. So maybe you're right after all that the ...Rf8 line is slightly worse. But this would have to be tested with some deeper analysis. Anyhow I wouldn't call the puzzle "broken" because the lines are both losing for Black and there's not much difference in the evaluation.

As for 44630, again, I think that lichess sometimes chooses these types of losing moves to make it "clearer" why a tactic is winning. I've seen this many times so I'd say it's not an oversight, but lichess policy. It is just one of the problems I have with lichess' tactics trainer which is why I still prefer to train on chesstempo.

@Bonifratz,

36287: "Slightly better" is still better so it should have been the line chosen, but we don't really have measly pawn gain puzzles since that would be too petty for middle game puzzles. Thus, my assertion that this should not have made the cut as a puzzle. ~,~'

44360: I don't think there's a "lichess policy" in play here. If you've read ( http://en.lichess.org/blog/U4sjakQAAEAAhH9d/how-training-puzzles-are-generated ), then you'll notice that mate-in-N and material advantage puzzles were produced by two separate algorithms. This material advantage puzzle overlapping as a mate-in-3 was not intentional. Perhaps some of the recursive evaluations terminated a tad too early and this is one of those.

I'm supposing this trainer was just something (a good novelty overall for reusing resources) Clarkey suddenly came up with one day, fiddled with, tweaked, and implemented and so now we have it. It's amazing how the frustrations can have this much profound impact on some of us though, hahaha.

With my unreliable connection, I naturally gravitate towards puzzle solving (in contrast to those who can even play bullet games). Not only do I sometimes lose clearly winning games, I even time out while resigning. It's pathetic. >,< (The server has single digit pings while mine wavers between the mid-hundreds and even thousands.) Chess is merely a hobby/pastime for me so no, we have no plans for getting better net access for this household for thrice or quadruple the monthly subscription price to have double the speed.

I'm tempted to make a Team for Puzzle Issues, but I'm afraid that despite all these sporadic complaints, people won't be interested enough to join and regularly participate and it will end up being a ghost town. There's also the ethical dilemma about these kinds of discussions being spoilers and we don't have any anti-spoiler features in the lichess forums. Plus, someone will eventually ask "What's the point of this Team?"