Utility-scale and commercial solar assets increasingly depend on continuous telemetry to catch faults before they cause energy loss, yet conventional rule-based monitoring produces high false-positive rates and misses subtle, gradual degradation. This report presents a per-inverter world model for unsupervised anomaly detection, built on a Joint-Embedding Predictive Architecture (JEPA) that learns the normal operating dynamics of each inverter individually and flags deviations as prediction-error anomalies. Applied across seven inverters at the Sibaya solar site, the approach achieves area-under-ROC scores ranging from 0.69 to 0.98 depending on data availability and fault history, and improves F1 score over a rule-based baseline by roughly 40–260 percent for inverters with sufficient fault history in the test window. We describe the modeling approach, the evaluation methodology, and a field diagnosis that distinguished a data-collection artifact from genuine equipment degradation — without disclosing proprietary model weights, tuned hyperparameter configurations, or internal system implementation details. Positioned within Asoba's broader Intelligence Layer for distributed power operations, this work implements the Orient stage of the underlying OODA loop: transforming raw telemetry into calibrated, per-asset anomaly signals that downstream Decision AI can act on deterministically.