Too difficult for computers

Here is a directmate in 17 without duals that uses the staircase maneuver and is difficult for most engines:

As an endgame study what matters is the existence of a winning strategy. It's a bit like a maths exam question that asks for a proof of a result. While it might expect the textbook proof, an alternative proof would be both acceptable and not invalidate the question.

I agree.

Wrong, by a mile.

https://www.wfcc.ch/1999-2012/codex/#c1

(3) Studies are unsound if there is a method of fulfilling the stipulation which is different from the author’s solution, and may also be rendered unsound by serious [16A] duals in the main line, but even in the main line many kinds of duals are normally tolerated.

Here, there IS a solution which is different to the composer's, and more than that, it does have many duals.

An endgame study by definition is a type of chess composition where uniqueness of play is a basic requirement. Once a study is published, chess players – as opposed to chess problemists who composed them in the first place – may treat the position as a practical endgame exercise where unique play is not necessary (just like in an actual game, it doesn't matter whether there's only one way to win or a hundred). But this practical function of studies is just a happy by-product. Studies are actually composed for artistic purposes, and unique play is a fundamental part of that. I can guarantee that had Nadareischwili known about the major duals in this study, he would have thrown it in the bin if he couldn't fix these ruinous flaws.

Nope, I disagree. This is unnecessarily restrictive. There is a good reason not to be so restrictive - it would exclude a large fraction of endgame studies!

Uniqueness is often dependent on where a study is considered to end (not something specified). For example, most king and pawn endgame studies end with queening. A bare K+Q versus K has an astronomical number of alternative solutions, mostly of different lengths but some can be the same length. So a solution has to be a route to a position where it is demonstrated that there is AT LEAST ONE way to win (a recursive definition).

The study has one intended way to win and can be solved by showing there is a way to win. It would be nice if this involved 19 unique moves being the only way to mate, but that is not true of most other studies for the reason above.

The points you raise are valid in the sense that anyone doing chess compositions would have to think them over in relation to solving or creating compositions. And after they did that and had their AHA moment they became happy campers and probably continued that way for the remainder of their composition careers.

What you missed, is the proper understanding of the uniqueness requirement. There is only a demand for uniqueness in the lines and moves counting as content of the study - which is more than just the simple solution btw. The remaining non-uniqueness may be replaced with a fixed but uncontentious result such as for the K vs K+Q outcome or that of a tablebase position. This is quite often necessary to complete the result list for the minimax evaluation of the whole study. In order to show the study is a win, giving the unique composer sequences is not always enough. Sometimes you will add that a black defense is refuted by a tablebase win or show that an alternative white attempt fails. These need not depend on unique sequences unless the author considers them thematic. So the author's design concept matters, it's not all technical. But once you dive into ambiguity, there is no coming back. With a big dual on move 2 only move 1 counts as a contributor to the study content!

Btw, this understanding of uniqueness is not overly restrictive and cooks no endgame studies. All composers are aware of the freedom they have in letting go of the non-unique tentacles of their octopussy

To get round your assertion (which is not at all controversial in serious composition circles) you merely need to reclassify part of the solution as proof of the RESULT, rather than how to get to a position with the result.

eg the problem ends with KQ v K and the last part of the composed solution is a way to force getting to this ending. Its lack of uniqueness is no more important than the lack of uniqueness of winning KQ v K, as long as you think of it as part of that final evaluation.

The only motivation for a solution to be 100% unique is in checking if someone has solved a problem. Interestingly it relates to the notion of chess strategies. A pure composition in your sense has a single winning strategy (up to when some unclear criterion of clarity is met), while the sort I am referring to have a winning strategy.

The uniqueness requirement quoted by Mazetoskylo comes from the Codex of the World Federation for Chess Composition. The Codex summarises the basic rules, conventions, and definitions followed by all expert problemists. You really think these study composers agree on a definition that would "exclude" a large fraction of what they create as valid endgame studies? Of course not.

Good job in making the distinction between the solution of an endgame study (which needs to be unique) and the mundane play that follows it when a clearly won/drawn position is reached (which doesn't have to be unique). As I wrote in the blog linked previously: "(Of course, most studies don’t finish with a mate but with White reaching a theoretical won position, one in which the solution must end because unique white play no longer exists.)"

And as I said, the latter part of the solution is simply a way to SHOW it is a theoretical won position (a position that can force a theoretical won position is a theoretical won position).

I am not disagreeing with what the rules chosen by WFCC are, I just disagree that they are ideal choices. I see no problem with having two categories of study.

Similiarly I disagree with the rules chosen by ICCF when engines got strong (chess.com made better choices), but I only object to the exclusion of normal chess by centaur chess, not with the independent existence of centaur chess.

There's a fundamental problem with these "two categories of study", because it's actually a category error to confuse endgame studies (composed positions where duals are eliminated for artistic reasons) and endgame exercises (practical "how to win" endgames where duals don't matter). Wanting the term "endgame study" to encompass the latter group is like wanting the term "centaur chess" to encompass both chess with and without computer assistance. Just as there's an appropriate general term "correspondence chess" to cover both types of chess play, you can use a general term like "endgame puzzles" to cover both endgame studies and endgame exercises.

The ICCF decision doesn't stop anyone from forming another group with no PC-assisted play, but it would be a mistake to combine the two types and allow games between PC-assisted and unassisted players, right? Dualised endgame exercises have little artistic values and that's why they will never be a part of the World Federation for Chess COMPOSITION. The WFCC is not some kind of overlord stopping people from creating endgame exercises, countless examples of which exist in practical endgame manuals. Rather, the WFCC is made up of people who self-impose the unique-play requirement because it generates special, surprising, artistic chess problems.

A typical study involves a paradoxical, unexpected solution that involves, say, a queen sacrifice. How is such a queen sacrifice of artistic value if there is an alternative move that also wins in a much more mundane and obvious way? That, in a nutshell, is why unsound studies with cooks or serious duals (like the OP position) are discarded by the chess problem community.

@Elroch: For instance, imagine that posters would "present" a study like this one with only a mundane dual which looks like the study was incidentally scraped off the seafloor with no visible signs of intelligent design? And imagine that doing that became the norm for posting dualed compositions? Engines would certainly favour those haphazard solutions! Which composer would care to create these ogres? I assume you would insist that the intended thematic lines are always included as well but that obviously won't happen as you have first leveled the playing field. Who decides which solution is "better" and which lines must be included when they all work?

I think you are trying to sell us a study type which already exists - as endgame theory and endgames in actual games where only the result matters. And the opposite of our artistic endgame studies which are !00% about the window views and 0% about the goal which is as exciting as disembarking the bus. Or as wise presidents say it "it's the journey, not the destination stupid!".

Man, somehow solved after 30mins

Note that all claims that this study was intended to have only one winning line but the composer missed alternates are refuted by the 7 alternative mates in 2 at the end.

I presume that the "purists" will say this doesn't really count (for arbitrary reasons).

I know all of the rules that they are talking about, but chess.com is not a "formal" setting where you must have a strict adherence to those rules. The title of your thread is "Too difficult for computers", which for the most part it is if mate in 19 is the objective here, which I believe is the way you meant this to be and I only have maybe 2 engines (out of about 200) that can find a mate in 19. I don't think they mean anything personally by insisting that they are right here, they just seem a little over zealous about a strict adherence as they have scolded me in the past (mainly because I am familiar with the rules) but as you can see it has fallen on deaf ears because to me we are not in an official study and mate solving tournament here the last time I checked.

Not that easy. Rocky64 explained that studies commonly abort before the mate precisely because tail moves are often dualed. How the composer packaged his original study+solution we don't know - this one comes from a book. Books are edited to tailor their content to their public. The poster most likely edited the solution as well - as the chess.com puzzle interface won't be capable of processing much of the book information or the original presentation.

The purists won't be bothered until the author claims valuable content in or after the dualed sections - with the exception of an established list of minor duals which are sometimes penalized by point deductions in composition tournaments. All items on that list have been discussed and agreed and they are not arbitrary interventions by a grudgy judge. Personally, the smallest minor dual I know of is that K vs K+Q has no unique mating sequence

Note: as I mentioned before, aborting a line due to upcoming duality is only permitted when the outcome is no longer in doubt.

You can't "abort before the mate" except by allowing someone to solve it by finding ONE of the mates in order to justify finding the moves up to there. Exactly the same as I stated suffices when there are alternative solutions at any point.

The outcome is not in doubt at the point that the solver has demonstrated a forced win. This technically becomes the initial position once they have found ONE of the possible solution strategies. In most cases, like this, it is very likely to be the intended one (or ones).

That is not only a made-up standard, it is also logically untrue. A line of moves leading to mate is not a proof of anything except that there is a helpmate available. To prove a fighting mate you'd need to provide all of the minimax analysis and not one variation which proves absolutely nothing.

But as said, it is a made-up standard. Most of our human endeavours are supported by packages of knowledge which grow like pyramids. There is no rule (mathematically or otherwise) that prohibits you from using a tablebase to access endgame outcomes directly. The proof that those tablebases are sound was already delivered a decade ago.

You mix up the roles of composition author and solver. The author is fully legitimized to use the knowledge packages available to create a correct composition. He would be insane to walk through the K+Q vs K moves. What a composition solver should do is entirely dependent on the context of the challenge. It is obviously as insane for him as for the author to deliver the full minimax analyses and for him too, one variation proves just a helpmate. What is worse is that you are overlooking the overwhelming presence of a giant mammoth in the room while concentrating on the small fry in the corner. What the solver should really be doing - and is required to do in any decent solving tournament - is to find the black countermoves at every turn. A puzzle interface is not a legitimate opponent in solving a chess problem; it is only there to facilitate the execution of the steps through the solution process. This is in fact the first thing any chess player learns about solving chess problems: you play both sides.

You have an awkward relationship with mathematics in this activity. You think the solver is producing a proof of his solution and consider that he should construct it from a clean slate with only the axioms of chess in his portfolio. If mathematicians had worked that way, math would still be in the stone age now. We are not mad.