How good are these forecasts?

Every forecast scored on seasons the model never saw while its settings were chosen.

Match races (two boats) and fleet races (six to nine AC40s) go into one Plackett–Luce model, so a win in a match race and a place in a fleet race both move a team's rating.

A team's rating is its sailing programme plus a boat term for the season. Programmes carry across cycles where the sailing team did: INEOS Britannia's sailors are GB1, Orient Express is La Roche-Posay, Alinghi Red Bull is Tudor Team Alinghi. The Women & Youth teams are rated on their club's Youth and Women's America's Cup crews.

The 2007 to 2017 cycles (ACC yachts, the 2010 multihulls, AC72s and AC50s) are in the fit too, counting less the older they are. Land Rover BAR is Ben Ainslie's team (GB1 today) and Areva Challenge is K-Challenge (La Roche-Posay today); Oracle Team USA, Artemis, SoftBank Team Japan and the 2007 challengers are rated as their own teams. Each season has its own boat term, so a new class starts every team's boat afresh while the sailing programme carries over. With them the model scores better on the held-back 2024 and 2026 races than on the official feeds alone.

The regatta odds sail every fleet race left with the real points (10 for a win, down one a place), then the top two sail a final race for the win.

The Louis Vuitton Cup odds sail the 2027 format 20,000 times from today's ratings: six fleet races, three round robins of match races (the defender races but its results don't count for the challengers), the top three challengers to the semi-finals and a repêchage among the other three for the fourth place, semi-finals first to five and a final first to seven. Each simulated Cup draws every team's pace afresh with a wide spread (about 110 rating points), because the AC75s are new or rebuilt and the AC40 regattas behind the ratings can't show boat speed. Team Australia has no races in the data and starts 50 points below the other challengers' average.

Emirates Team New Zealand's chance to keep the Cup adds up, over every challenger, the chance it wins the Louis Vuitton Cup times the chance it beats the defender first to seven, from the ratings with the same wide spread of pace.

The 38th America's Cup's regattas (Naples, 2027) aren't in the results feed yet: their dates are kept by hand from the America's Cup's announcements, and the Louis Vuitton Cup and Match odds come from simulating the format, not the feed. The published Conditions don't say how the opening fleet races carry into the standings or how the repêchage is sailed; the simulation uses the fleet races only to break ties, sails the repêchage as a double round robin and pairs the semi-finals first against the repêchage winner, second against third.

The Youth and Women's America's Cups (2027) get odds once their entry lists are out.

The Women & Youth crews race AC40s alongside the principal teams in the preliminary regattas; they are rated as their own teams.

AC75 design (the boat term) and the one-design AC40 regattas share a season's boat term, a simplification: the AC40 boats are identical.

Not included: the AC45 World Series fleet races (2011-2016), whose results survive only as images and standings; the 2000 and 2003 challenger series, published as standings only; and practice races, for which no full finishing orders are published.

Questions and answers

How accurate are PodiumOracle's America's Cup predictions?

On 143 races from held-back seasons (2024 to 2026), which the model never saw while it was tuned, its favourite won 59% of the time in fields of about 4, and it gave the eventual winner 46% on average. Its log loss was 1.085, against 1.147 for treating every team as equal and 1.091 for recent form.