Here is how the multistage Monte Carlo router works. Multiple paths are propagated using polars and GRIB wind data. In this case there are three cycles of 48 hours each. At the end of each cycle there is a crude pruning based on the simple distance to the finish, and then a fancy alpha shapes pruning is done to eliminate near-equivalent cases and also routes terminating internal to the fur ball. When you have gone as many cycles as you care to, examine the isochronic tracks that end closest to the finish. And then tell yourself, “no way am I going that far North”.
The next step would be to take the wobbly best paths and do some sort of simulated annealing on them to refine the results, but no time, no time!