Limitations and Deployment

Future work should include a full randomized clinical trial with CONSORT-style criteria, prespecified endpoints, preregistered analysis, and independent research organization involvement. The study does not establish real-world safety, effectiveness, workflow impact, or health-equity outcomes. The specific implementation is not open-sourced because it depends on internal Google infrastructure and safety concerns about unmonitored medical AI deployment. Chat-based consultations limit non-verbal cues, dynamic visual assessment, and physical examination. Validation on private or prospectively collected datasets remains important because some public artifacts may have been encountered during pretraining.