DeepRegatta/Guides

Guide · Retrace · Mini 6.50

What wins Mini 6.50 races — in Série and in Proto?

In the one-design Série fleet, results are almost pure sailing speed: 88% of the model's signal across 40 modeled races. In the development Proto fleet the picture opens up: speed drops to 62% and routing rises to 16%, with routing boldness — how far a boat dares to sail from the fleet's route — the second-strongest individual driver. Identical boats isolate the sailor; open rules reward the navigator too.

Evidence · Retrace playbooks S3a
Mini 6.50 Série vs Proto — share of model signal by factor family
Série (identical boats)Proto (open rule)SpeedSérie — Speed: 87.9%87.9%Proto — Speed: 62.2%62.2%RoutingSérie — Routing: 7.7%7.7%Proto — Routing: 16.2%16.2%WeatherSérie — Weather: 3.7%3.7%Proto — Weather: 11.1%11.1%Fleet positionSérie — Fleet position: 0%0%Proto — Fleet position: 8.9%8.9%HandlingSérie — Handling: 0%0%Proto — Handling: 1.6%1.6%share of aggregated driver importance · XGBoost, grouped k-fold
View as table
Factor familySérieProto
Speed87.9%62.2%
Routing7.7%16.2%
Weather3.7%11.1%
Fleet position0%8.9%
Handling0%1.6%
Aggregated driver importance by category for the models that predict top-25% finishes in each fleet. Solo sailors on identical Série boats are the closest thing in the study to pure skill, isolated.Source: Retrace playbooks S3a_mini_serie (40 races, 1,437 entries) and S3a_mini_proto (45 races, 636 entries) · data through 16 Jul 2026.

What the evidence shows

The Série model is one of the most confident in the whole Retrace study (AUROC 0.959 ± 0.021 over 1,437 boat entries): with the boat held constant, average speed versus the fleet dominates everything else. The Proto model is just as accurate (AUROC 0.974) but distributes its signal — routing boldness is its second driver, and weather and fleet-position factors appear where the Série fleet shows none.

Reading the contrast

The pairing is a natural experiment in what open rules buy. Same courses, same solo sailors' world — but where design freedom exists, willingness to leave the fleet's line becomes a scoring factor. For a Série campaign the message is blunt: boatspeed hours beat routing hours. For a Proto, the routing desk earns its keep.

Methodology

Retrace trains one model per fleet — XGBoost over ~123 features per boat entry — to predict a top-25% finish, with grouped k-fold cross-validation. Série: 40 modeled races, AUROC 0.959 ± 0.021, High confidence. Proto: 45 modeled races, AUROC 0.974, Medium confidence. Importances are aggregated into the families charted above; details on the Retrace methodology page.

Limitations

The Proto sample is smaller per race (636 entries across 45 races) and carries Medium confidence. Solo offshore tracker data is noisier than fully crewed racing. Drivers are correlational and describe these samples — the boldness signal in Protos says bold routes distinguished leaders, not that any bold route pays.

Open the Mini 6.50 playbook on Retrace free · no signup