In British English, non-finite do can precede the site of traditional VP-ellipsis: Tom has written a paper and Emma has done too. Interestingly, the presence of do has puzzling consequences for both A- and Ā-extraction possibilities. Namely, local subject wh-extraction, Quantifier Raising and topicalization are all acceptable, but object wh-extraction is disallowed; A-extraction of derived subjects is allowed with unaccusative and raising verbs, but not passive be. This paper shows that accounting for these facts and more requires that (i) non-finite do is inserted to host stranded non-finite inflectional affixes (i.e. it is an extension of traditional do-support (Ramchand 2018)) and (ii) do-ellipsis involves deletion of the full verbal phase (Boskovic 2014). It is shown that the extraction possibilities out of do-ellipsis corroborate Boskovic's (in press) claim that local wh-subjects move to a lower position than wh-objects. Further, given that do-ellipsis involves deletion of the entire verbal phase, the paper re-examines which projection should delimit this domain. It is shown that while the progressive layer always delimits the verbal phase when present (Harwood 2013), the perfective layer may delimit the verbal phase under certain conditions; namely, only with auxiliary been.