Equity-Audit Inference for Computer Agriculture Credit-Scoring Models Under Unobserved Land-Tenure Documentation: An Artificial Intelligence Fairness-Generalization Bound

Credit-scoring models deployed in computer agriculture routinely condition on land-tenure documentation that is systematically missing for smallholders, and the resulting fairness audits are themselves biased because the audit sample is a non-random subset of the scoring population. The bound is derived from a before-after-control-impact design with a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64) and evaluated across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points.54 clusters are required to detect a 5-percentage-point equity gap at 80% power, and that the audit's coverage of the fairness-generalization interval degrades predictably as cluster count falls below this threshold. The bound separates the audit's sampling uncertainty from the documentation-induced selection bias, yielding an identifiable interval for the equity gap that remains valid when tenure documentation is missing not at random. The contribution is a formal inferential object that lets auditors state what an equity audit can and cannot establish about a deployed credit-scoring model, rather than a point estimate whose validity depends on an untestable missingness assumption.