Built on weather inputs that are ~1.5–2.5 hours old at the most, far fresher than the ~6–12 hour lag typical of ECMWF/GFS. By ingesting more recent observations, the model provides a near real-time view of evolving conditions, significantly reducing forecast staleness.