Probably this has already been considered, but it's something that's bothering me a bit, so I want to mention it explicitly.

If I understand things correctly, the idea here is that the user will be given a certain position which can be won (or drawn) by the application of a certain method or technique, and their task will be to play out the win (or draw) against a strong engine. Great idea, but I wonder if there might be a problem.

In another thread I gave this position as a simple case where the Greek gift is correct. The trouble is that if the engine used is anything like mine, then after 7. Bxh7+ Kxh7 8. Ng5+ Kg8 9. Qh5 it is likely to play 9. ...Qxg5. The computer sees that it's going to have a massive disadvantage and so throws its queen onto the fire as the least horrible of all its options. The problem, of course, is that this isn't what a human is likely to do, and the solver doesn't really learn much about the Greek gift this way.

It would be more instructive, I think, to have the computer play 9. ...Re8 and see if the solver can find 10. Qxf7 etc. Perhaps, then, some combination of pre-entered defences (so the user is sure to see the relevant ideas) and an engine (so moves keep coming if the user steps off the main path) is the right way to go?

Sorry if I'm stating the obvious and/or missing the point :)

Valid points it seems to me. My first thought is: well, it's not as though the feature we're working on has to accommodate all positions. Perhaps for exactly the reasons you've given, the Greek gift position you've given is not suitable for this feature. ("this feature" needs a name, IMO).  It would be an excellent position for the Chess Mentor, however. 

We should gather up the low-hanging fruit first perhaps before concerning ourselves with the tougher bits.

I think a lot of endgame's are "low-hanging fruit" -- B and N mate. B and Bishop mate. King and pawn. Opposition, distant opposition. 

I think it's a bit a of challenge, an interesting one, to find the positional themes that are suitable for testing against an engine.

That said, obviously, if there's a way of solving the "engine sees too deeply to  play futile defensive moves" problem (perhaps forcing a certain limited ply on the engine would cause a more human-like approach to the Greek-gift) position) then let's solve it. No immediate solution jumps out at me.

I think this is a question for the application development team, with some communication with the position development people. One solution is that each problem would come with a move tree of limited size, like a computer's opening book. As long as the user stays in "book" the computer plays "book" moves. The engine is relied on when the user strays from the standard solution or once the solution line runs out of moves.

 

It would be nice if multiple defenses were possible. So the user would play out the above position and then would be able to go play it again where the computer tries a different defense.

we are not going to be able to do a move-tree unfortunately. :( that is why we are going to have to test them all to make sure they are good positions for this application. it's a bummer that computers play this way...

Ah, fair enough. So long as we know :)