This paper argues that the meaning of a clausal ellipsis site can only be recovered from a syntactically derived question, regardless of whether this question is explicitly uttered or is merely pragmatically inferred. This entails that the meaning of a clausal ellipsis site cannot be recovered from an inferred question q in a language L if q is syntactically ill-formed in L. I demonstrate that this restriction on recoverability can account for Merchant’s ( 2001; 2004) Preposition-Stranding Generalisation and for the observation that fragments appear to be sensitive to syntactic islands ( Merchant 2004; Abels 2011; Barros et al. 2014; 2015) without any mention of whether remnants of clausal ellipsis themselves undergo movement. Because there is no need to stipulate that remnants themselves undergo (often exceptional) movement under this approach, a theory of clausal ellipsis modelled on Cable’s ( 2010) Q-based analysis of wh-questions is developed that permits non-pronunciation “around” designated phrases. This approach is shown to be preferred on many occasions to the predominant movement-based analysis ( Merchant 2004), which is too restrictive and must frequently resort to the notion of ellipsis repair.