In the marine-hull research programme, a simulated autonomous wing-sail boat gains about 5 points of progress over simply following the wind headers, but only if its forecast is skilful at 15-minute scale for 1 to 2 hours. We asked whether archived HRRR wind forecasts (3 km, issued 3 h before use) can supply that skill. They cannot, at least as tested.
- Against 12 station and 9 to 10 open-water buoy records for January and July 2024, raw HRRR lands 0.8 to 6.5 points below the header-follower.
- Trailing-hour bias correction, a longer window, a fade and a learned station bias recover about 2 points at most.
- A nowcast (buoy level plus HRRR's change since the replan) only ties persistence.
- Correcting buoy anemometer heights to 10 m shrinks the speed bias but not the gap: direction error (24 to 30 degrees) and day-to-day shifts carry it.
Limits: two months, every fourth day, swept anemometer heights, a 3 h latency. Hourly-cycle HRRR is untested. Details and result files: research/marine-hull/LOG.md, milestones 5h to 5m.