Lichess puzzles problems

Lichess, which I thoroughly regret to have been introduced to only a few months ago, is a definitely wonderful chess-site, all the more remarkable that it is open to all including the most cash-strapped players – thank goodness.

I train with Lichess puzzles to improve my chess level, and they generally work very well. From time to time however, I do experience frustration to a degree. After noting the clearest occasions for several weeks, I found three types of anomalies. Here are three cases in point, one of each type.

Type I
Some puzzles just jam. 50854 for instance :
Dc4+ d5 And then no further move is allowed. I tried all of them regardless of their value, and they all get the failure answer.

Type 2
Some puzzles lend your virtual opponent incredibly bad moves. 2350 is such a case :
1 NxBf5 (only move allowed by the puzzle). And then :
… dxCe5 ? instead of BxDd1, obvious move, thus forsaking your Queen.
And what follows is nearly as incredible :
2 QxBg4 Nc2+ forking the King and the Rook – ok, but
3 Kd1 Nd4 ? instead of NxRa1
The puzzle ends with a victory message. Some victory.

Type 3
Some claimed failures are just equivalent to official solution – let alone better solutions. Number 29626 :
Te2 is the book's solution but two other moves are better though they are counted as failures :
a.- KxNf8 and nothing will prevent black Rook b2 from causing a check by the Bishop and threatening an unprotected white Rook at the same time ;
b.- Rb5+, securing white's a5 Rook as well as c Pawn next move, before it becomes really dangerous.
Both solutions come at no cost and are rated as failures.

I was lucky enough to run into a comment by (apparently) a Lichess administrator explaining two things : one was that in early stages, puzzles were generated by a second-choice software and that removing older puzzles was more or less impossible. I'm in no position to comment on that, much regretting to be thoroughly unable to make any suggestion.

The second comment gave some useless hope : users rate puzzles, and badly rated puzzles can just be ignored. There are two flaws, and the most obvious is that when you randomly call a puzzle, it pops up without showing its rating.
The less obvious is that an easy, ready-to-use rating should give plus and minus numbers as ratios of number of plays.
Then rises the matter of having players' opinions in the first place, if possible in significant numbers. Maybe with a comment of the type : "This puzzle has not been tested often enough yet to be rated ; please test and rate it" ?

I dearly hope a solution will come soon to clear Lichess of its puzzles problem. Keep up the good work – and thank you very much indeed.

Type I — read up on en-passant capture. Puzzle 50854 — Qc4+, d5, exd6+ etc.

Type 2
1...Bxd1 2.Bxf7#;
3...Nxa1 4.Bb5+ Qd7 6.Qxd7#

Indeed Knives'moves should be played. But the puzzle doesn't play them, and that's the whole problem.

Indeed, see post #2 in regard to your Type I problem...it's a nonexistent problem, you just don't understand en passant. As for Type II those are indeed silly when the puzzle opponent just makes bad moves, but it's fairly rare. Type III is the most frustrating in the few puzzles it happens.

However rare the problems may be, I've been on about this for weeks now due to the overwhelming number of these posts every week.

I really don't know enough about how the puzzles development worked, but I know that many of the oldest puzzles are inherently flawed. If they indeed puzzles that have been generated cannot be removed, then that is a HUGE design flaw in this (I've used the word a lot) experimental system for generating tactics puzzles.

I think the puzzle system is great when it works. But when it doesn't work there are no good options for us except to downvote the puzzle. But the entire system of ratings and up/down votes is flawed. I honestly think that someone should just take the time to fix the algorithms used to generate puzzles, and regenerate a whole new batch of puzzles with the updated engine with a bit more tolerance for suboptimal solutions (for instance, mate in 3 vs. mate in 2 should give "good move but you can do better" rather than an outright fail, as should equivalent moves that miss mate but win so much material that a win is still guaranteed, etc.) and then I think everyone would be happy with the up/down voting and ratings systems that are in place because the puzzles would have far less flaws.

The problem is, I don't think any developer cares enough to do this, and if there is "no way" to remove the puzzles, then I suppose that would also create a huge problem to regenerating them all. I find it hard to believe though that there is absolutely no way to remove them. How absurd. It's all being stored in a database isn't it? o.0 Just flush the database and start over with a few tweaks to how they deal with a couple problem situations. I've yet to have any developer respond to this idea, though.

As regards the type 1 problem, I just hadn't noticed that the last move offered a "prise en passant" (I've known en passant for over 40 years thanks). But there were two similar cases of no move allowed by the puzzle, one I couldn't find back in my list, and the other that I obviously misspelled when writing its number.
In the messages about puzzles problems, someone seemed to be offering some help.

The chap who seemed to offer some help is "game_spectator" in
"38554 and lots of badly broken ones" subject, in a message dated yesterday.

What I quoted to achja in another thread: "Moves that result in checkmate in the minimum amount of moves are also solutions; moves that result in checkmate in 1 or 2 extra moves are considered retries, meaning the user can make another attempt to find the correct solution." ( source: http://en.lichess.org/blog/U4sjakQAAEAAhH9d/how-training-puzzles-are-generated )

It's almost trivial to implement a boolean sieve to filter out broken puzzles. Maybe they can't remove specific puzzles because of how too tightly integrated the existing player and puzzle ratings are though the puzzle ratings aren't really reliable due to two main reasons.

1) Unreplayability. The more games two players play against each other, the more the win/lose/draw ratio will reflect their relative strength. However, with puzzles, you only play against it once. Your success or failure might even be a fluke.

2) Indirect statistics. How two players fare against each other affects their ratings. Puzzles aren't competing against other puzzles. The players' one-time trials serve as fuzzy mediums for determining the puzzle ratings.

#5 Fenris1066,

If someone does make a new batch of puzzles with an improved system with better ratings determination for players and puzzles, no spoonfeeding via auto-promotion of pawns ( e.g. http://en.lichess.org/training/28533 ), and requiring players to move for both sides while exploring all relevant solution branches, etc., lichess can opt to retain the current training feature as "old training" for those who wish to continue wading through its frustrations and surprises while a "new training" can become available for those who want it. People will have old training ratings *and* new training ratings.

#1 NickelAnge,

Wow! You've already gone through a tenth of all the puzzles. You sure are tenacious. :3