Keith DeRose and Stewart Cohen object to the fallibilist strand of pragmatic invariantism regarding knowledge ascriptions that it is committed to non-cancelable pragmatic implications. I show that this objection points us to an asymmetry about which aspects of the conveyed content of knowledge ascriptions can be canceled: we can cancel those aspects that ascribe a lesser epistemic standing to the subject but not those that ascribe a better or perfect epistemic standing. This situation supports the infallibilist strand of pragmatic invariantism (...) according to which knowledge semantically requires absolute certainty but this claim is often pragmatically weakened: it turns out that exactly those aspects of the conveyed content are cancelable that this view claims are pragmatic. I also argue that attributor contextualism and relativism do not have an alternative explanation of this phenomenon. (shrink)