Skeptical invariantism is the view that the meaning of ‘know’ is invariant, and so strong that it is very difficult or even impossible to know anything. A well-known difficulty for invariantism is that whether people make or accept a knowledge claim varies from context to context without any evident change in the subject’s belief, its truth, or its justification. This has led to (indexical) contextualism, on which an instance of ‘S knows p’ expresses a different relation between S and the proposition that p, and thus has different truth conditions, in different contexts of use.

I will examine another leading alternative to skeptical invariantism: John MacFarlane’s assessment relativism, which consists of a relativist semantics for ‘S knows p’ and a set of norms. I will argue that the assessment-relative semantics has little explanatory value, is subject to some of the objections to contextualism, and does not provide monadic truth conditions even when conjoined with an assessment-relative semantics for truth. What is relative to contexts of assessment, I suggest, are conditions of appropriateness. Like many other terms with strict and invariant meanings, loose uses of ‘know’ are appropriate in some contexts but not others.