EDIT: I am actively debugging this, will post results shortly.
This string will cause the bug to appear:
"This is a test EN! This is a testEN. This is a testEN? This is a testEN, This is a test; This is a testJP！ This is a testJP。 This is a testJP？This is a testJP、 blah"
I did this testing with the default autoDelay settings left in place, and tested with Japanese punctuation marks as well as English ones. I am wondering whether there is something that is interpreting/consuming the exclamation mark in SuperTextMesh logic before a given character can be assigned the auto-delay in the preprocessing step.
And again, this only appears when the ruleset it set to STMAutoDelayData.Ruleset.FollowedBySpace.