EGTB Online distributed kbpkn.tbs stats file is incorrect
Posted: Sun Feb 17, 2008 5:52 am
This is mostly directed at Kirill, but other people will likely want to be aware of this as well. The stats file kbpkn.tbs that is distributed on Kirll's EGTB Online web page is incorrect. It's packaged in the file all-5-men-32p-tbs.zip.
I discovered this in the process of working on my revamped version of Nalimov's tbgen code (henceforth Kominek's tbgen2), wherein I found it wise to generate all 3-5 man tables as a regression test. In this one case my generated stats file disagreed with the reference. Faced with such as conflict my usual response runs along the lines of "oh bother, what have I mucked up now?" Looking again, however, my stats agree with those generated by FEG, a valuable double-check. So I dug out the precompiled tbgenm.exe that was included in Nalimov's release (*) and generated kbpkn from scratch. Two days later, the verdict: all three programs are in agreement.
This signature of the the bad stats file are dtm values over 100.
In fact, the max dtm is 100, which is consistent with Kirill's longest checkmates page, found at http://kirr.homeunix.org/chess/longest- ... ngth.shtml.
Interestingly, the wtm lost and btm mate values are correct, but the wtm mate and btm lost values are all wrong. I can't say I know how this could have happened with such a behind-the-times 5-man file. It isn't obviously mixed up with another one.
I've attached the kbpkn.tbs file from my running of the official, as it were, tbgen program.
john
(*) Possibly compiled by Dann Corbit?
I discovered this in the process of working on my revamped version of Nalimov's tbgen code (henceforth Kominek's tbgen2), wherein I found it wise to generate all 3-5 man tables as a regression test. In this one case my generated stats file disagreed with the reference. Faced with such as conflict my usual response runs along the lines of "oh bother, what have I mucked up now?" Looking again, however, my stats agree with those generated by FEG, a valuable double-check. So I dug out the precompiled tbgenm.exe that was included in Nalimov's release (*) and generated kbpkn from scratch. Two days later, the verdict: all three programs are in agreement.
This signature of the the bad stats file are dtm values over 100.
Code: Select all
---- from the bad kbpkn.tbs
wtm: Mate in 105: 2
wtm: Mate in 104: 2
wtm: Mate in 103: 5
wtm: Mate in 102: 5
wtm: Mate in 101: 1
wtm: Mate in 100: 2
Interestingly, the wtm lost and btm mate values are correct, but the wtm mate and btm lost values are all wrong. I can't say I know how this could have happened with such a behind-the-times 5-man file. It isn't obviously mixed up with another one.
I've attached the kbpkn.tbs file from my running of the official, as it were, tbgen program.
john
(*) Possibly compiled by Dann Corbit?