How this is built: for each regatta, every school's
predicted skipper(s) and their current Plackett–Luce rating are
pulled from the lineup predictions. Each simulated race perturbs every
skipper's rating with independent Gumbel(0,1) noise and sorts descending
— the Gumbel–max trick, which produces an
exact draw
from the Plackett–Luce distribution those ratings were originally
fit under, so the simulation uses the same generative model as the
ratings themselves, not a separate approximation. Each school's team
score in a simulated race is the sum of its skippers' finishing
positions (standard low‑point team scoring: lower wins). Repeated
5,000 times per regatta; win / top‑3 / average‑rank are the
fraction of simulations each school actually won, placed top 3, or its
mean finishing rank across all of them.
Singlehanded and keelboat regattas use ratings fitted to those
disciplines (the sailor's main rating plus their own results; for keelboats also the crew aboard and the school's
keelboat program), calibrated on seasons the model had not seen. A singlehanded school is its sailors, one per boat;
a keelboat school is one boat per crew of the size that event used before. See the
singlehanded and keelboat selectors on the
Power Rankings.