Mainstream poverty analysis currently renders certain people and degrees of privation more socially legible than others across high-income countries. This article examines how these hierarchies carry through to and corrupt wider social scientific analysis, inscribing differential value to actors and phenomena in ways that undermine social understanding and explanation. First, conventional approaches to poverty analysis and measurement obscure the de facto prevalence of deep poverty, as well as those most subject to its violence. Second, a growing number of hyper-marginalised groups are missing from population income surveys, undermining the accuracy of (deep) poverty estimates and public understanding of both its determinants and dynamics. Third, the inferential and external validity of income surveys is significantly diminished by problems surrounding data quality and coverage. Attempts to address this have principally focused on improving data quality, but as demonstrated in this article, these strategies exacerbate poor representation of the lowest-income groups in distributional analysis. Much more than merely technical or pragmatic, these are theoretical and normative judgements about who counts in welfare policy and politics. Overall, I demonstrate how current data practices occlude some the most violent forms of denigration and exploitation that structure advanced marginality, particularly the gendered, racialised, bordering and ableist practices underpinning state–citizen dynamics. Focusing principally on the UK context, I argue that the epistemic erasure committed features in and systematises a policy blindness to deep poverty for some of the most marginalised social groups making it harder to evidence its effects and address its causes across high-income countries.