Problem 16737

[drahacikfm]

Another high-rated problem that is high-rated only because it is ambiguous.

Rybka gives these first moves:

1.Nxe6 +3.67
1.Kh1  +3.32
1.Ngf3 +3.03

A difference of only 0.35 between the best move and the second best move is clearly ambiguous.  This problem should be disabled.

[richard]

Toga liked the Knight move a lot more than Rybka, at depth 16 it gives it a +6

There is always going to be differences between engines, I'd probably tend to trust Rybka (based on reputation) over Toga..do you have fritz or some other engine for a third opinion?

Of course my too large "winning alternative" test here is not helping either, the bug I mentioned in the other thread is also in play here with the pre-tactic evaluation comparison confusing the generator (toga would have removed this as having an alternative winning line, as while it did like Nxe6 enough to deem it tactic worthy it also liked the next best move enough to be deemed winning if it were not for my "compare to pre-tactic" evaluation bug.

Regards,
Richard.

[drahacikfm]

I think Toga +6 is probably close to Rybka's +3.6, because as mentioned earlier, Rybka gives much lower numbers than other engines, often only half as much.  For this problem it's not so much the actual number, but the very small difference between best and second best moves.

Sorry for all the posts today.  I don't mind getting lots of high-rated problems wrong.  But it seems that most of the ambiguous problems in the problem set are pushed up into the very high ratings, which I am getting a lot of now! 

[richard]

Toga does see a big advantage to the second move though, the toga evals (after depth 18 and looking at over half a billion positions) are:
Nxe6 +6.53
Ngf3 +3.92
Kh1 +3.61

So Toga clearly thinks it is seeing something here that Rybka isn't (given that the other two moves are relatively close to Rybka's analysis).  I've become a bit mistrustful of Toga in multipv mode, if you or anyone else has Fritz I'd be very interested in knowing its idea of this.

In any case this is all a bit of a mute point, I should probably have seen the +3.92 as good enough to trigger a "winning alternative" evaluation.

No need to apologize for the posts, it helps draw my attention to the worst offenders.  I'm not going to disable these manually for now as I need to add some extra code for manual disabling not to impact the next verification run  (not a lot of code, but I'd rather spend that time on the generator which can fix this across all problems instead of just problem by problem - if the next run still has issues I'll start the manual disabling).

I'm also hoping that as the higher rated users get more of a look at the harder problems that the illegitimately hard problems will move to a higher rating level thus hiding them a bit more than is currently the case.

Apologies for the annoyances.

Regards,
Richard.