Evaluation of Sortition-Weighted Jury Learning for Artificial Intelligence Agriculture Subsidy Adjudication Under Ministry Capture: A Governance Legitimacy Bound

Ministry capture threatens the legitimacy of artificial intelligence systems that adjudicate agricultural subsidies, because a captured ministry can shape training labels, feature selection and appeal outcomes without ever issuing a visible instruction. The design is formalised as a legitimacy bound: the probability that a captured ministry can secure a favourable adjudication falls as the jury's sortition weight rises, but the bound tightens only when jury competence and independence are jointly maintained. Because no observational data on captured ministries exist, the paper characterises the design analytically rather than empirically. A replication-based evaluation of the before-after-control-impact estimator was conducted under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64), with power estimated across 2000 design cells. The analysis indicates that 52 clusters are required to detect a 5 percentage-point effect at 80 per cent power. Sortition-weighted jury learning can bound capture-induced legitimacy loss, but only under identifiable conditions of jury independence, label auditability and appeal reversibility, and it specifies those conditions as the governance contribution.