Low-rated problems with only one move

[drahacikfm]

Some of the very low-rated problems are actually fairly difficult.  They are low rated only because they are marked correct after only one move.  If the user had to actually show the idea, the problems would have much higher ratings.  A couple examples:

5732:  Marked correct after 1.Rxc6+.  But after 1...Kxc6, it's a King and pawn ending with 3 pawns for each player.  The interesting part is 2.f5 gxf5 3.g5! queening a pawn.  The user should have to show that idea, not just that he can trade a rook.  This problem is not 951 Standard!

4602: Marked correct after 1...Rxc5.  But after 2.a6, White is queening his pawn.  2...Rxc2 3.a7 Ra2 4.a8=Q Rxa8 5.Bxa8 d3 6.Kf2 b4 7.Bd5 I don't see any win for Black.  Maybe there is a better move for Black earlier, but that is not the point.  The point is this is a hard problem, and should not be rated 955 Standard!  But it is because it requires only one move.

I've seen many more problems like this, where there is a simple capture of material, and the problem is marked correct.  But then the opponent has some interesting counterplay that takes real thinking to deal with.  The difficulty in dealing with this automatically in the generator is that the computer sees how to deal with the counterplay, and thinks the problem needs only one move because it's just a hanging piece.

[newlook]

Maybe there could be a tag or a checkbox for "needs more moves." If a threshold is met, then the computer adds one more move from its stored line.

[richard]

After the next problem set update in the next couple of weeks the "need more moves" issue is certainly the area in which the problem set needs most work, unfortunately it is also fairly hard to handle automatically in the generator.  Users have given me a few ideas that are worth investigating , but ultimately it is still a difficult problem.  It may be easier in the long run to provide a mechanism for users to suggest better lines.

Newlook, your idea of a checkbox would only work for some problems, many of these problems have short sequences because - as Drahacik point out - the computer essentially sees these positions as hanging pieces due to having seen that the counterplay doesn't ultimately help so doesn't include the counterplay line, these are the positions that users describe as "computerish defences".  Humans would often try the counterplay lines even if they see them as futile, hoping that their human opponent doesn't have computer like tactics ability. There are some positions where the checkbox would help, I attempt to detect where a position has gone quite to avoid pointless moves at the end of a tactical sequence, but sometimes I get this wrong (especially in pawn promotion tactics) and showing more of the computer line would be useful in these situations.

I also plan to make more use of the quality ratings for problem set pruning in the future, at the moment I only use the stars as a guideline for looking at areas I can improve the generator but eventually I will start to use these to explicitly shape the problem set.  The main barrier at the moment is that the problems need more exposure to higher rated players.  Lots of 1 star problems are rated 1 star because the user made a blunder and mistakenly blamed the problem (there are also legitimate one star problems [most commonly "more moves", but also some 'winning move' threshold issues] but at the moment I think that the majority are still based on users not understanding the problem).

Regards,
Richard.