The Volador 1.0 preprint introduces a Swin-Transformer with Mixture-of-Experts and cross-grid attention for full air-sea coupling in the South China Sea, claiming RMSE/MAE parity with REDOS V2.0, GLORYS12 reanalysis, and ROMS on 0-72 hour upper-ocean forecasts while resolving internal waves via expected turbulence cascades. Methodology relies on three-month hindcasts plus 15-day real-time runs without disclosed training dataset size or independent validation splits, a common preprint limitation that invites overfitting concerns. This approach extends patterns seen in earlier data-driven models like FourCastNet but adds latent-space flux exchanges absent from uncoupled regional systems; however, the arXiv submission (May 2026) lacks peer review and real-time buoy or satellite ground truth beyond reanalysis benchmarks. Ablation tests correctly highlight coupling gains, yet broader adoption for shipping and fisheries hinges on multi-year operational trials the authors themselves flag as future work. Related ROMS evaluations in the same basin show similar submesoscale fidelity only after extensive parameterization tuning, suggesting Volador's speed advantage may trade long-term stability.