> Using scoreset 1 I get I get 2.12.
This still begs the question of why the subject is being re-written for some messages if the score is below my threshold.
It may have been scanned twice, and got marked the first time (perhaps even by a different system in a different network). The second scan will over-write the spam-status header, reflecting a lower score.
Try customizing your spam tags (and make them different for each server you run if you have multiple).
Bonus points if you use the add_header feature to create a secondary X-Spam-Status header that is X-Server1-Spam-Status:
