Every model, solved and compared.

Every forecast in hand is routed, not just the one you picked. The result panel opens on the consensus track and names the model it came from, and you can step through the others to see what each would have had you do.

When they all put you on the same side of the course, that is a decision you can make quickly and stop thinking about. When they part company, you have found the real question in the race, hours before the start, and you can plan a hedge instead of discovering it at the first shift.

Departure-time scans work the same way: solve the course from up to 24 starts and see which window actually wins.

And the search shows its work. Turn on the isochrone bands and the chart draws the frontier the solver actually reached at each step — you can see why the track bends before you trust it.

Raw global fields run to hundreds of megabytes. What reaches you is the band of ocean your route could cross, at the steps the search needs, compressed — a small fraction of a megabyte for a coastal race.

That is the difference between routing being practical and not, on marina wifi or on a metered satellite connection. Pull it once and keep re-routing against it: different starts, finer resolution, changes to the course, with nothing further to download.

Currents come from the NOAA model that resolves your water — SFBOFS inside San Francisco Bay, the lake models on the Great Lakes, RTOFS offshore — at hourly steps, because a twelve-hour stride does not sample a tide, it misses one.

Every cell of the matrix reports its finish-time delta. That much tells you how wide the spread is. The part that matters is where: for each perturbation that leaves the baseline, the panel names the hour and position the tracks first part company, and draws the diverging track on the chart as a ghost.

A stable route is a decision made. An unstable one hands you the exact question — which side of the headland, at what wind strength the answer flips — while you are still ashore. The matrix runs in your browser against the forecast you already downloaded, so it costs nothing and needs nothing further from the network.

Open the route, take a fix, name the mark you still have to round. The fleet solves the rest of the course from where the boat actually is and answers with a heading to steer now, the tacks and gybes ahead with their times, and your finish. The whole reply is smaller than this page.

It is built for the boat with Starlink or one bar of cell: the solving happens ashore, and what crosses the water is a readout, not a forecast. Re-route from a fresh fix whenever the picture changes, or let it re-route itself on a timer.

Waves never slow the modeled boat — an ORC VPP has already priced wave resistance into the polar, and pretending otherwise would double-count it. They steer the route instead: a hard ceiling the track must respect, and a preference that trades a little time for a calmer sea when you ask it to.

Either way, every candidate reports what it sails through — the highest sea on the track, the average, and the hours spent above two and a half metres — so the comfort of a route is a number you compare, not a guess. Depth works the same way: give the workspace your draft and a margin, and water that fails it is not water the route may use.

Pasted polars take matrix CSV, TSV and .pol files, and Expedition or PredictWind angle-speed rows, and work out which they have been given.

A pasted polar is temporary unless you explicitly save it to your private account. Shared routes carry the track and a digest, never the speed table that produced them.