It seems like a lot of people have recently been having trouble with getting their comments rejected as spam. Going through trac, I selected some random tickets and tried to re-train our bayesian filter. In the course of doing so, I noticed that every single Conch-related ticket had a starting score of nearly 100% spam. After repeatedly training 30 or 40 messages, it became clear that the training database was insistent upon Conch-related tokens being extremely spammy. As a result, I've re-set the training database, which means we're starting from scratch. Trac admins: when doing spam monitoring, remember that "delete as spam" means "selected messages are spam" and "delete as ham" means "selected messages are NOT spam". It seems like someone (possibly me!) accidentally flagged a whole bunch of Conch-related messages as spam, and this was causing a lot of problems. Thanks, -glyph