The Judgment Infrastructure Gap: A Standard Operating Framework for Making Judgment Delegation Observable and Auditable in AI-Mediated Systems

This working paper formalises the Judgement Infrastructure Gap: a structural failure mode in AI-mediated decision workflows where judgement is silently delegated across system layers without observability, auditability, or responsibility traceability. As AI systems increasingly operate as intermediaries in high-stakes institutional, organisational, and governance contexts, decision failures can no longer be adequately explained by model accuracy, alignment objectives, or interface usability alone. Drawing on a longitudinal, cross-domain corpus of empirical case analyses and system-level studies, the paper reframes judgement not as a cognitive attribute or ethical add-on, but as an infrastructural function that requires explicit representation. It demonstrates how judgement delegation currently occurs implicitly across instruction intake, generative execution, and delivery layers—producing recurring failure patterns such as false completion, artefact mismatch, cross-model convergence with institutional disagreement, and format-fit illusions. The paper introduces a Standard Operating Framework for Judgement Infrastructure, specifying core structural primitives, judgement telemetry signals, and responsibility-flow representations that make delegation events observable and auditable without modifying underlying models. Rather than proposing new optimisation targets or alignment metrics, the framework positions judgement as a control plane that must be governed at the system and organisational level. Positioned deliberately at the boundary between academic research and operational deployment, this work contributes to AI governance, systems engineering, and science and technology studies by naming a class of failures that existing disciplinary frameworks cannot fully address in isolation. It is intended both as a methodological anchor for scholarly inquiry and as an operational reference for organisations seeking to manage decision risk, responsibility distribution, and governance integrity in AI-mediated workflows.