The word “know” is revealed as vague, applicable to fallible agents, factive, and criterion-transcendent. It is invariant in its meaning across contexts and invariant relative to different agents. Only purely epistemic properties affect its correct application—not the interests of agents or those who attribute the word to agents. These properties enable “know” to be applied correctly—as it routinely is—to cognitive agents ranging from sophisticated human knowers, who engage in substantial metacognition, to various animals, who know much less and do much less, if any, metacognition, to nonconscious mechanical devices such as drones, robots, and the like. These properties of the word “know” suffice to explain the usage phenomena that contextualists and subject-sensitive invariantists invoke to place pressure on an understanding of the word that treats its application as involving no interests of agents, or others. It is also shown that the factivity and the fallibilist-compatibility of the word “know” explain Moorean paradoxes, the preface paradox, and the lottery paradox. A fallibility-sensitive failure of knowledge closure is given along with a similar failure of rational-belief closure. The latter explains why rational agents can nevertheless believe A and B, where A and B contradict each other. A substantial discussion of various kinds of metacognition is given—as well as a discussion of the metacognition literature in cognitive ethology. An appendix offers a new resolution of the hangman paradox, one that turns neither on a failure of knowledge closure nor on a failure of KK.