Resolving to Directionally Consistent: Replicating a Capability-Practice Divergence in AI-Drafted Assessment Feedback, and What Became of the Null Result