Here's a story from Al which illustrates the advantage of property rationalization and the dangers of overtraining extremely well:

"The log P value for phosphonic acid, 2-oxopropyl, dimethyl ester was entered as 0.86 when it should have been -0.86. CLOGP gets -0.84. Priscilla Jow's notebook had the correct value. I must have entered it wrong and she didn't catch it in proofreading. It has the negative sign in my old copy of June 1987, and it still hadn't been corrected in our 'Exploring QSAR' book.

You can find it by SMILES as: CC(=O)CP(=O)(OC)OC

Syracuse Research's KOWWIN quotes our 1987 Issue (+0.87) and comes up with a 'correction factor' for the string:

and a calculated log P of +0.65."