How good are these forecasts?
Every forecast scored on seasons the model never saw while its settings were chosen.
Each GP's classification (first to last of nine or ten teams) goes into one Plackett–Luce model. A team's rating is the team plus a term for the season, which takes up a change of pilots: teams keep their pilots through a season but swap between them, and each team races a woman and a man who share the driving.
The boats are identical RaceBirds, so the rating is the pilots and the team's set-up and strategy rather than the boat.
The title odds race the rest of the season with the points a place has scored on average this season (E1 adds bonus points, for qualifying, fastest lap and the group races, on top of the table), and the points so far are the championship's own.
Wind is tested as an input: the RaceBird foils a metre above the water, and a strong breeze brings chop that can take a boat off its foils, which makes a GP more of a lottery. The weight was fitted on the tuning season, and it is kept only if it helped there.
Very little data: three seasons and fewer than twenty GPs. Treat every number here as rough.
The heats (group races, play-off, place race and final) aren't published as data, only each GP's classification and points, so each GP is one result.
Pilots aren't rated on their own: a pilot who moves team takes nothing with them in the model.
No live timing feed is published; results arrive with the classification after each GP.