Problems #11523, #25045, #20625

[roq]

Hi Richard

What do you think about this kind of problem?

CT #11423

The solution is the simple 1...Rxg2 [end of solution]. However, at this point the combination hasn’t really happened and after 2.Kxg2 white is ahead in material for the moment. The best continuation for black isn’t trivial and seems to be: 2...Bd5+ 3.Kh3 Qe4 4.Qc8 4...Be6+ Line, but white has other defensive options that drop material.

CT #25045

This is a more borderline case. You could argue that anyone who plays 1. Bxf6 will probably have seen that if black takes back with the bishop then 2.Nd5 is going to win material (although I don’t necessarily believe that argument). I think though that the solution would be better if it made this explicit. The difficulty is that black retaking with the bishop is probably not the computer’s idea of black’s best defence – but of course the fact that taking back with the bishop loses material is the reason that the combination works in the first place!  

CT #20625

This one is a bit different. In the solution line, but after 4...Kd7 (instead of 4...QxN?!) the line below ends after 8.Qd4+ with white having gained the exchange and a pawn (although now, not surprisingly since the king is exposed, a very long mate has appeared in the position). However, 1.Qd5+ wins a piece (the same material value) immediately with a crushing advantage (after this the best move for black is probably 3...resigns!).

1.Qh7+ [1.Qd5+ Be6 2.Nxe6 Qxd5 3.cxd5] 1...Kf8 2.Qh8+ Ke7 3.Qg7+ Kd8 4.Nf7+ Kd7 (...QxN?! is solution)  5.Nxd6+ Kc6 6.Nxe8 Qxe8 7.b5+ Kb6 8.Qd4+ Line

I think one of the the issues here is that the computer’s evaluation of 1.Qh7+ is based on factors that are not going to be apparent to human players at the moment that it curtails the line.

---------

Currently I’ve marking these 2* (dubious) and using the comment: “needs more moves” for the first 2 and the last one has both “early sac of material” and “ambiguous tactic” issues.

Problems of the first kind are very prevalent on CTS, but some get through on Tempo as well. I assume that the generator sees multiple winning moves or multiple defences close together in evaluation and terminates the solution - since any further moves would be ambiguos. I suspect that the only real way of fixing some of these kind of issues would be by human editing.

It occurs to me, though, that what these three positions have in common is that when the solution ends there is at least one continuation that has not yet reached a quiescent state. In the case of the first two the solution line has not reached a material resolution at the time the solution ends - the position is still up in the air. In the last problem its not the solution line but an alternative defence that hasn’t been resolved. Aesthetically, It’s a bit like hearing the end of a symphony where the conductor has forgotten the coda or failed to resolve one of  the minor themes.

[richard]

Unless I've missed something I think you might have the wrong problem number for your first example.  Could you double check that one please.

#25045 is certainly one I have to look carefully at to see the point (or have the computer look carefully for me :-)  ).  As you point out the computer isn't going to give me the retake move.  I think this one is going to annoy anyone that doesn't take the time you have to find out why the retake isn't great. I think that probably the more skilled users might be more forgiving on this one, I'm actually a little surprised its rated so low at the moment, over time it should drift a bit higher. The main problem with positions like this one is that its not going to be that instructive to a lot of users.  I'd be happy to leave it in if its rating does go higher.

#20625 is an interesting one.  The engine I'm currently using to feed me position evaluations (Toga 1.3) sees the 4..Kd7 move as around 8 pawns worse than the 4.. QxN move.  It is not obvious why as at this point it still hasn't seen the mate coming up at the end of that line. The only thing I can think of is that the lack of king safety for black with the rook and queen still on the board is evaluated very harshly by Toga.  What does Fritz see as the best move if you leave it running for a couple of minutes at 4..  ?

I'd probably be tempted to leave #25045 in (it seems justifiable, although I didnt look at the computer evaluation), but remove #20625. #20625 doesn't seem to be a generic problem with the generator but rather a quirk of the computer's evaluation that is being provided to the generator. 

Regards,
Richard.

[roq]

CT #11523
Yes, you are right - I put the correct number in the title, but then botched it in the body (-: So the number should be #11523 not #11423.

CT #20625

On #20625 Fritz wants to play 4...Kd7 for about a minute before switching to 4...QxN The difference in evaluation slowly climbs until after about 5 minutes Fritz shows about 8 pawns difference the same as Toga. It’s difficult to believe that Fritz’s programmers would ever give as high a penalty as 8 pawns for something as insubstantial as king exposure, but Fritz doesn’t see an actual mate until 8.Qd4+ is played when it finds mate in eleven in about 30 seconds or so. So I’m not sure what is happening here either. This is what Fritz is looking at when 4.Nf7+ has been played on the board: 

1.Qh7+ Kf8 2.Qh8+ Ke7 3.Qg7+ Kd8 4.Nf7+ Kd7 5.Nxd6+ Kc6 6.Nxe8 Qxe8 7.b5+ Kb6 8.Qd4+ c5 9.bxc6+ Ka6 10.Qc5 bxc6 11.Rb2 Qh5 12.Qxc6+ Ka5 13.Rb5+ Line

Note that 13...Rxh5 wins the black queen, but Fritz won’t admit to having seen that at the stage when 4.Nf7+ has been played.

I’m being strict in marking the problems, because I feel that a good number are already up to the kind of standard that you get in a tactics book. Since you have so many potential problems available it seems to me that it might be feasible to have all the problems up to that standard eventually - so this approach might be more helpful in the long run.

I'm actually a little surprised its rated so low at the moment

Whilst I agree with you about #25045 (but it would greatly benefit from editing to add an extra move), the issue is that it’s possible to use a simple algorithm to solve quite a large percentage of the problems (even more so on CTS), very quickly and with minimal chess benefit: 1) Select all possible captures and checks as your candidate moves 2) If any candidates are checks play the most forcing check, else play the most forcing capture.
 
Using the above algorithm what would one play in #25045 and #11523? Hmmm, thought so. That’s why this kind of position *needs* the additional moves that justify the obvious first move. IMOP, with the prevalence of this issue on CTS, anyone there with an accuracy stat below about 70% is likely to be making almost exclusive use of the kind of algorithm outlined above and this may explain why there are often plaintive remarks on the message board from people whose practical play has not been improved, despite their prowess on CTS.

My own experience on these two was that although I was able to mentally analyze #25045 out to quiescence quite quickly, I couldn’t see the justification of #11523 (lots of branching) in a reasonable time and had to resort at least partially to the guessing game, but of course it did produce the right answer here!

[richard]

Hi Roq,

Guessing is certainly a problem with some of these. I take some solace in knowing that CTS is even more abundant with these types of positions :-)  I'm glad you are rating these as there will definitely be a stage where I start to manually prune the problem set and user ratings will be very important for this, partly due to the number of problems to go through, but also due to my own poor tactical vision :-)

Regards,
RIchard.

[roq]

Hiya

my own poor tactical vision

What’s that expression they use at Microsoft? Something like errrhm “eating your own omelette”? No that’s not quite it.... 

[richard]

Hiya

my own poor tactical vision

What’s that expression they use at Microsoft? Something like errrhm “eating your own omelette”? No that’s not quite it.... 

:-)  My Omlette's probably taste like dogfood!