DeepRegatta/Guides

Guide · Retrace · Class40

What separates the leaders in a Class40 fleet?

Speed — then the weather you sail into. In 14 modeled Class40 races (179 boat entries), speed factors carry 47% of the model's signal, but weather carries 28% — the highest weather share of any scenario in the Retrace study — with wind strength during the race the single strongest driver of all. Class40s reward crews who position for pressure, not only those who sail fast.

Evidence · Retrace playbook S3b
What separates Class40 leaders — share of model signal by factor family
SpeedSpeed: 47.5% of model signal47.5%WeatherWeather: 27.8% of model signal27.8%RoutingRouting: 13.2% of model signal13.2%Fleet positionFleet position: 8.3% of model signal8.3%Race contextRace context: 1.8% of model signal1.8%StartStart: 1.5% of model signal1.5%share of aggregated driver importance · LightGBM · AUROC 0.949 ± 0.062
View as table
Factor familyShare of signal
Speed47.5%
Weather27.8%
Routing13.2%
Fleet position8.3%
Race context1.8%
Start1.5%
Aggregated driver importance by category. The strongest individual driver is wind strength during the race; wind luck (better or worse wind than the fleet) ranks fifth, position relative to the pack sixth, and side-of-course choice seventh.Source: Retrace playbook S3b_class40 · 14 races, 179 boat entries · data through 16 Jul 2026.

What the evidence shows

Within a box-rule fleet, relative speed still leads — but it explains less than half the signal. The individual driver list is telling: wind strength during the race is first, wind luck — getting better or worse wind than the fleet — is fifth, position relative to the pack sixth, and which side of the course was chosen seventh. In offshore one-design racing, leverage and pressure-hunting are measurable, repeatable differentiators.

What that means for a campaign

Strategy preparation earns a bigger share of the budget than it does under handicap rules. Studying how past editions of your target races rewarded positioning — and replaying them boat against boat — is not a luxury; in this fleet it is where a quarter of the outcome signal lives.

Methodology

Retrace trains one model per scenario — here LightGBM over 127 features per boat entry — to predict a top-25% finish within the Class40 fleet, validated leave-one-race-out because the race count is small. AUROC is 0.949 ± 0.062. Importances are aggregated into the families above; full details on the Retrace methodology page.

Limitations

Only 14 of 125 candidate Class40 races passed the label and track-quality gates, so this playbook carries Medium confidence — the model is strong on the races it saw, but the sample is thin. Weather-heavy signal can also reflect the specific editions modeled. Correlational, sample-specific, not a rule of sailing.

Open the Class40 playbook on Retrace free · no signup