For more detail, please refer to www.paulgraham.com. JunkBroom implements many of the suggestions made by Paul Graham, with some minor tweaks added.
Spam filtering is basically a classification problem. The task is to find out whether a certain message is Spam or not. This is done by calculating probabilities that relevant tokens (words) of an email are indicators of Spam; these probabilities are gathered from existing Spam and non-Spam messages.
Basically, if a word appears only in Spam messages, it is a very good indicator of spam; if it appears only in good messages, it is a very good indicator of a good message. Everything else is in between.
This is why JunkBroom needs to know about the Spam probabilities of tokens: in order to classifiy future tokens.
JunkBroom's "View" menu gives you the ability to display a window showing token/probability mappings (probabilities are from 0 to 1, 0.99 meaning 99 percent). These aren't all tokens, only those considered relevant by the algorithm.
A token is something very much but not quite like a word. You can simply consider it as a word, although "0000FF" hardly qualifies as one. Words from your messages are mostly single tokens.
This is an indication that the algorithm is working as expected. "0000FF" is the RGB value for red, a color spammers often use in HTML messages (to gain attention).