DeepRegatta/Guides

Guide · Retrace · ORC

What decides races sailed under ORC?

Boatspeed first — but not as overwhelmingly as under IRC. Across 72 ORC races and 1,032 boat entries, speed factors carry 59% of the model's signal while routing carries 18% — roughly ten times its weight in the IRC sample — with distance lost to missed wind shifts a top-six driver on its own. Under ORC, sailing fast and sailing the right course are both live battlegrounds.

Evidence · Retrace playbook S1b
What drives beating an ORC rating — share of model signal by factor family
SpeedSpeed: 58.8% of model signal58.8%RoutingRouting: 17.7% of model signal17.7%Rating (TCC)Rating (TCC): 10.6% of model signal10.6%Fleet positionFleet position: 5.7% of model signal5.7%WeatherWeather: 4.5% of model signal4.5%StartStart: 2.7% of model signal2.7%share of aggregated driver importance · XGBoost · AUROC 0.786 ± 0.103
View as table
Factor familyShare of signal
Speed58.8%
Routing17.7%
Rating (TCC)10.6%
Fleet position5.7%
Weather4.5%
Start2.7%
Aggregated driver importance by category for the model that predicts top-25% corrected finishes under ORC. Compare the routing share with the 1.7% it carries in the IRC sample.Source: Retrace playbook S1b_orc · 72 races, 1,032 boat entries · data through 16 Jul 2026.

What the evidence shows

The strongest single driver is still average speed versus the fleet, and the rating itself is second. But the routing family behaves differently from IRC: distance lost to missed shifts ranks sixth among all individual drivers. Boats that beat their ORC rating in this sample were disciplined about the wind they sailed in, not just how fast they sailed through it.

How to analyse your own ORC race

Because shift discipline is measurable, it belongs in the debrief. Replay the race with wind-shift scoring and sector analysis to put numbers on what a missed bascule cost — then compare that against the plain speed gap to your rivals. The race replay guide covers a sequence that works.

Methodology

Retrace trains one model per scenario — here XGBoost over 116 features per boat entry — to predict a top-25% corrected result, with grouped k-fold cross-validation so no race leaks between folds. AUROC is 0.786 ± 0.103; importances are aggregated into the families charted above. Full details on the Retrace methodology page.

Limitations

The ORC sample is much smaller than IRC's (72 modeled races against 597), and the wider ± 0.103 fold-to-fold variance reflects that. Of 276 candidate ORC races, only 72 passed the label and track-quality gates for modeling. As always, drivers are correlational: they describe this sample, not a rule of sailing.

Open the ORC playbook on Retrace free · no signup