Forum

Yes it is difficult to answer that, as it depends on the total number of tokens. What I have found useful to understand is that if an individual token is great than 50%, it pushes the score up, and vice versa. If you want more, you should really look at the underlying code in the extension or in the base code.

I did not write the underlying bayes code, I only tried to show the concept of running % an accurate representation of individual token effects, that sums to the same value calculated by the bayes algorithm. One thing to note is that I first sort the tokens by their absolute distance from 50%, so that you can see which are the most important tokens.

The addon file, which ends in .xpi, is just a zipped version of the source. So if you unzip this you will have the source. You can then zip it back again to reinstall it. There are better ways of doing this if you do development though, see https://developer.mozilla.org/en/Extensions

As for your symptoms, the Junk % in the message list shows the junk % at the time the message was first analyzed, while the Junk Analysis Detail shows the analysis using the current token set. If you as for a message to be reclassified, then the junk % column should match the detail.