Evaluation of Liquidity-Constraint Screening Audits for Artificial Intelligence Agriculture Credit Models Under Cooperative Default-Risk Pooling: A Governance Type-I/Type-II Error Bound for Smallholder Exclusion

Artificial intelligence credit-scoring models increasingly allocate agricultural finance through cooperatives that pool default risk, yet the governance question of how such models should be audited for the exclusion of liquidity-constrained smallholders remains unresolved. The framework distinguishes two error types: a Type-I error, in which a liquidity-constrained smallholder is wrongly flagged as a strategic defaulter and excluded, and a Type-II error, in which a genuinely strategic defaulter passes the screen and draws on the cooperative pool. Under cooperative default-risk pooling, the framework shows that the audit's error bound is not a property of the model alone but of the interaction between the screening threshold, the prevalence of liquidity constraint in the applicant pool, and the pooling rule that distributes losses across members. A governance bound is derived that specifies the maximum tolerable Type-I error rate as a function of the cooperative's risk-sharing obligation and the smallholder's liquidity profile. The analysis is methodological and theoretical: no observational data are used or claimed. The contribution is a governance instrument that converts an abstract fairness concern into an auditable error bound, permitting regulators and cooperatives to specify, ex ante, the exclusion risk they are willing to accept and to test whether an AI credit model meets it.