v1 classification model trained and validated on FathomNet imagery. Hits 95%+ accuracy on known species. Anomaly detection flags unknowns for human review and also self resolves with VLM integration. Future iterations will expand taxonomy coverage and improve edge-case handling.