Can someone explain this? Looks like at least half a dozen tests should have hit it. Furthermore, if the score is 0.0, why wasn't it autolearned? Thinking that maybe SA was just to busy at the time and took a break, I ran it through manually. Same thing, but it got autolearned as ham.
Autolearning is an opportunistic process. It can, should, and will fail to happen sometimes.
Basically only one process can have write access to the bayes database at a time. If a message comes it that would qualify for autolearn, but the database is already write-locked it will skip the auto learn (reasons for the existing write lock include expiry and other autolearning runs).
The alternative is to force SA to wait for the lock to be released, but that would result in rather slow performance.
