The field of Adult Literacy and Numeracy (ALN) has an evidence problem. This paper analyses that problem through the lens of three overarching types of evidence available to researchers and policymakers: Type 1 evidence, which provides descriptions of the scope of social problems; Type 2 evidence measuring the impacts of programmes addressing those problems; and Type 3 evidence of the mechanisms through which programme impacts are realised. Whereas ALN was once characterised by a lack of Type 1 evidence of ALN problems and Type 2 evaluative evidence of programme impacts, there is now an abundance of these evidence types. However, ALN finds itself in an 'evidence impasse': it repeatedly appears that programmes aimed at improving ALN skills do not do so. This impasse may be the product of evaluation failure, not programme failure. I argue that evaluations should place greater emphasis on the production and analysis of Type 3 evidence of the mechanisms through which adult basic skills may be improved over time. This paper outlines a strategy for doing so, with a particular emphasis on long-term impact of literacy and numeracy practices on basic skills.