The strongest action any matching rule demanded (block > sanitize > warn > allow), or the policy default when nothing matched.
Content classes recognised in the message.
Reason code(s) from the matched rule(s), for the log + finding.
OptionalsanitizedFor a "sanitize" decision: the cleaned message JSON (a deep clone; the original is never mutated). Present only when sanitising actually changed something.
The outcome of evaluating one message against the policy.