Like all modern organizations, modern states are subject to a “data imperative”: a mandate to mine data and to decide what to manage based on what can be measured. In service of data-hungry machine learning techniques, the state (and its contractors) find themselves compelled not only to seek and demand new kinds of data, but also to mine it in a somewhat agnostic fashion to find the relations that stick. It is no longer necessary to flatten society to make it legible (as high modernism required); instead, ubiquitous data capture means that categories emerge inductively from regularities observed in the data.