We will be available for live Q&A on September 16, 2020 at 7 am Pacific / 4 pm Central European time. If you want to attend, please email Pranav at panand@ucsc.edu for the meeting details. We cannot host it during the conference due to a time conflict. If you'd like to chat but but can't make it to the Q&A, let us know. Comments and suggestions are most welcome. Abstract: Starting with Sæbø (2009), recent years have seen a lot of interest in the so-called subjective attitudes: English find and its counterparts in, e.g., French, German, or Norwegian. Unlike vanilla doxastics (i.e. think), find-verbs have been argued to only allow matters of opinion, rather than fact. In this paper, we consider one underanalyzed class of expressions in find-complements: epistemic modals. Must and might have featured prominently in the literature on judge-dependence and assessment sensitivity. Those modals exhibit such hallmarks of subjective expressions as faultless disagreement (Egan et al. 2005; MacFarlane 2014; Stephenson 2007), which makes them prime candidates for embedding under find. However, epistemics are prohibited in subjective attitude complements (first observed by Sæbø 2009). The find+must ban has been attributed to must not being subjective in the right way (Coppock 2018; Sæbø 2009). We argue instead that the real culprit is a matter of evidence: find-verbs require their subject to have direct evidence for the complement (first proposed in Stephenson 2007), while must and its counterparts in other languages require a lack of direct evidence (von Fintel and Gillies 2010, 2018; Lassiter 2016; Mandelkern 2019). Therefore, the find+must combination yields an evidential clash. We support our claim by novel data on find-verbs and a range of indirect expressions, and analyze the find+must ban as a semantic contradiction, couched in frameworks of von Fintel and Gillies (2010) and Anand and Korotkova (2018)