Thursday, July 19, 2007

CT-Art coming apart at the seams

Ok, I'll say it. The makers of CT-Art did not put in the necessary effort to make problems at the highest difficulty they intended. They have flat out failed to make the problems on the last level correct. I'm not just blowing gas here either, I'll give you two very concrete examples -- and I promise the last levels are riddled with this kind of thing.

First, a simple example. A position from problem 1178.
White to move

White has crashed through on the kingside and just needs to polish off the mate. The first thing that came to mind was f7. This threatens mate on g8 and forces Bg7 (other bishop moves allow Qxh6). Then white will play Ne4 planning Nf6 or Ng5 threatening Qh7 or getting the queen to the h-file (e.g. Nf6 Bxf6 Qxh6 or Ng5 hxg5 Qh5).

With those ideas, we think, what if white plays Ne4 first. Black does not seem to have any real productive moves. Certainly not any more productive than he would play after Ne4 in the previous line. If you're really keen you'll realize that the move order with Ne4 first leaves the f7 square open for the knight, e.g. Ne4 c5 Ng5 threatening Nf7 (if hxg5 Qh5 leads to mate quickly).

The CT-Art solution was f7, with no credit for Ne4. The best line for black after f7 is Bg7 Ne4 c5 (CT-Art gave Rd5 falling for mate in 2) Nf6 (or Ng5) Be4 Nxe4 (Notice f7 is unavailable for the knight!) e5. White is ahead 2 pawns and has a much safer king, but the move order with Ne4 first was clearly better.

Ok, that was the simple example. At least the solution given by CT-Art was actually winning. Problem 1177 was riddled with even worse errors. This takes some time to explain the multitude of errors, feel free to skip to the end of the rant.
Black to move
The first move, Bxh3, is difficult to calculate all the consequences of, but it's easy to see the white king position will be exposed and there will be chances for black. White's response gxh3 is forced. Now CT-Art completely misses the win. Qxh3 threatening Nf3 mate. Be2 is forced. Black then has Nd5 and after white moves the queen away, Qa3 Ng4 Bxg4 Qxg4+ Kh1 is forced.Black continues, Nxe3 fxe3 (forced) Qh3+ Kg1 Qg3+ Kh1 Rd5 Threatening mate.

White has to block the rook from going to h5 using either the knight or rook on f5. This gives a piece back to black after Rf5 Rxf5 Nxf5 Qh3+ Kg1 Qxf5. This leaves black ahead to pawns and the white king remains exposed.

This line starting with Qxh3 is not the solution given. Instead, the program gives what is actually a move that does not lead to an advantage. After Bxh3 gxh3 from the starting position, CT-Art gives Rxd4 Bxd4 Nf3+ Kg2 Nf5.

Position A, White to move

And now a poor move is played by white to allow black to win. CT-Art doesn't even consider Rfd1 which allows white all the wiggle room he needs to stay alive. The program gives Be3 as the main line. This allows N5h4 Kh1 Qf5 and the threat of Qxh3 is too strong.

The program also gives two other lines from 'Position A'. First, Bf6 gxf6 Rh1

And now the baffling Re8. There is no commentary or further moves after Re8 except that they give it an '!'.

The second line from 'Position A' is Rg1 N5xd4 Bd3 (another questionable move from white) Nxg1 (Qe5 was much better) Qxd4 Nxh3 Rh1 Qd5+ Qxd5 Nf4+ Kf3 Nxd5. After all that, black has a two pawn advantage.

Labels: , ,


At 10:38 AM, Blogger Blue Devil Knight said...

Nice examples. Even in CTB I've noticed that a couple of 'solutions', to work, rely on suboptimal play from the opponent. Not a good habit to get into!

It is really quite ironic that they obviously didn't computer check these programs.

At 1:04 PM, Blogger Loomis said...

Unfortunately, there are errors like this on nearly every problem at this level. The problems are complex and this leads to multiple winning moves before the very end of the problem. Convekta seems to have missed many of the ways to finish off a position. I think so close to the end, I'll finish off the problem set even though the problems are disappointing me.

At 10:43 PM, Blogger Underpromoted Knight said...


Sometimes CTA mentions that the puzzle was taken from a particular game, but a few of the pieces were moved around--AND THOSE SMALL CHANGES SIGNIFICANTLY CHANGE THE PUZZLE.

I have not yet decided what to do about these frustrating "incorrect" problems. One solution is to skip those problems entirely, as you can still get in 1000 problems (surely fewer than 209 are wrong.) Another solution is to consider only the variations in which there *is* a clear winning path. This is practical; for, in real OTB games, you don't always have time to calculate everything, so you make your best effort, *especially* if/when you are losing and need to create head-splitting complications to see if your opponent will crack under the pressure.

At 5:20 PM, Blogger transformation said...

good to know, good to have the heads up.

thank you.



Post a Comment

<< Home