So why does http://spamassassin.apache.org/tests.html say that the BAYES_99 test scores 1.886 (net and Bayes) but the value in 50_scores.cf shows 5.400?
the web-site is currently based on SA 3.0 pre-something and is not based on SA 2.63.
One can clearly demonstrate this by noticing that DRUGS_SMEAR1 exists on tests.html, but that rule clearly doesn't exist in any version of SA prior to 3.0.
In general never treat tests.html as documentation of your version of SA. It's documentation of _a_ version of SA, but not necessarily the one you are using.
In particular, as SA gets close to a new release, tests.html, and most of the other pages, will get updated for the new version. Currently SA 3.0 is just about to hit release-candidate stage, probably in the next few days, so the update isn't too surprising.
