Evaluation of Byzantine-Robust Federated Aggregation for Computer Agriculture Pest-Surveillance Consortia Under Poisoned Extension-Device Reports: An Artificial Intelligence Governance Convergence Bound for Outbreak-Alert Reliability

Federated learning consortia that pool pest-surveillance reports from smallholder extension devices face a governance problem that is not captured by standard robust-aggregation theory: the adversary is not an anonymous network intruder but a ministry-level actor whose poisoning budget is set by administrative authority rather than by computational resources. The analysis is a replication-based design evaluation, not an observational study. Under a declared variance structure with between-cluster standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64, power was estimated across 2000 design cells spanning cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. Fifty-four clusters were required for 80 per cent power at a 5-percentage-point effect. The governance convergence bound, not the statistical power curve alone, is the binding constraint on consortium design, and that the two must be evaluated jointly before an outbreak-alert system is certified.