Version 0 sets out the logical scaffolding of the full article. The article addresses a question that policy success scholars have identified as open: what makes a criterion of policy evaluation methodologically sound if the criterion itself is formed within the very social and institutional process that the framework is meant to evaluate. The proposed answer consists in introducing a methodological layer over the existing framework that converts the problem of criteria from an epistemological question into a systematic analytical instrument. This layer operates not on policy itself but on its regime of recognition, and makes it possible to distinguish evaluative systems that preserve a society’s long-term capacity for self-correction from evaluative systems that secure institutional entrenchment while diminishing that capacity. The article develops the methodological layer through six analytical categories that specify the meta-layer across the four dimensions of the existing framework, and through a conception of society as composed of subjects who simultaneously occupy multiple social roles. This conception turns the process dimension of the framework into a testable property: a process is sound to the extent that it brings the full set of the subject’s roles back into the analysis. In the context of the AI transition, this layer specifies engineering requirements for the design of sociotechnical systems. When a framework becomes the operational logic of automated decision loops, the soundness of criteria becomes a matter of structural safety of democratic governance.
Dmitrii Krivosheev (Mon,) studied this question.