Plausibly, how much is at stake in some salient practical task can affect how generously people ascribe knowledge of task-relevant facts. There is a metaphysical puzzle about this phenomenon, and an empirical puzzle. Metaphysically: there are competing theories about when and how practical stakes affect whether it is correct to ascribe knowledge. Which of these theories is the right one? Empirically: experimental philosophy has struggled to find a stakes-effect on people’s knowledge ascriptions. Is the alleged phenomenon just a philosopher’s fantasy? I propose a new psychological account of when and why people’s knowledge ascriptions are sensitive to stakes. My hypothesis is motivated by empirical research on how people’s judgements are sensitive to their social context. Specifically, people’s evaluations are sensitive to their ‘psychological distance’ from the scenarios they are considering. When using ‘fixed-evidence probes’, experimental philosophy has found that what’s at stake for a fictional character in a made-up scenario has little or no effect on how participants ascribe knowledge to them. My hypothesis predicts this finding: the scenarios are too ‘psychologically distant’ to participants. Our empirical puzzle is resolved: the stakes-effect often present in the wild won’t be present in vignette studies. (This illustrates a widespread problem with X-phi vignette studies: if people might judge differently in other social contexts, we can’t generalize from the results of these experiments. That is, vignette studies are of doubtful ‘external validity’.) The hypothesis also resolves our metaphysical puzzle. It predicts that people do not ascribe knowledge in a way deemed correct by any of the standard philosophical views, namely classical invariantism, interest-relative invariantism, and contextualism. Our knowledge ascriptions shift around in the way that’s most useful for social beings like us, and this pattern in our judgements can only be endorsed by a genuinely relativist metaphysics for knowledge.