Never seen this happen before...
X-Spam-Status: Hits=2.7 Required=5.0 Tests=STRONG_BUY=2.7 Autolearn=ham
SA 2.63, threshholds are 1.0/12.0. WTF? ;-) \
The score used to determine autolearning is NOT the same score as hits.
The score used is the score the email would have gotten if bayes was completely disabled. This includes using a different scoreset.
The autolearner saw the score of this email as zero, since set1 was used for autolearning scores.
Look at the scores of strong_buy:
score STRONG_BUY 2.799 0 2.800 2.700
