Race, retrain, repeat: the data loop behind digiSail’s personalised polar

 In March, Joss Creswell wrote about what he and digiLab's R&D team, theLab, were building ahead of the Solo Guy Cotten: a personalised polar trained on his boat, his sailing and the conditions he'd actually be racing in. Not a generic factory polar built in ideal conditions, but a personalised model that learns how Joss really sails his Figaro 3.

The Solo Guy Cotten has now been raced. Here's how the personalised polar performed, and what it means for the rest of the 2026 Classe Figaro Bénéteau season – including the one Joss is really building towards: La Solitaire du Figaro Paprec.

"I'm really excited to be working with digiLab for another racing season," Joss says. "It's great news that digiLab is growing fast at the cutting edge of UK personalised AI technology, and that as they grow, the scale and ambition of our racing program is able to grow in step with that."

A one-design class, where the edge comes from the sailor

The Classe Figaro Bénéteau is offshore racing's most famously ruthless arena. It's a one-design championship – every boat in the fleet is identical, down to the hull, sails and foils – so results come down entirely to navigation, strategy, endurance and decision-making. Almost every Vendée Globe winner since 1992 has come through it.

Which means the only edge available is the one the sailor builds for themselves. 

That's where digiLab comes in.

Building the first personalised polar: last season’s data

Every boat comes with a factory polar – the manufacturer's theoretical prediction of how fast the boat should go in any given wind condition. It's a useful baseline, but it's generic. It doesn't know how Joss actually sails this boat.

Having followed Joss through last season's racing, theLab was able to use the data we'd gathered to train the first version of his bespoke polar – a model that learned to predict his real boat speed across the conditions he'd encountered. The improvements over the factory polar were substantial:

• Average speed prediction error fell by roughly a third.
• The model explained around 75% of the real variation in performance, up from 44% – meaning it captured far more of what was actually happening on the water.

That's the difference between a polar built for an idealised sailor in idealised conditions, and one trained on how Joss actually sails.

The honest test: pushing the model outside what it knew

A data-driven model learns patterns from the conditions it has seen before. When it encounters conditions outside that range, its accuracy naturally declines. When we applied that first polar to Joss's March training session – in wind conditions outside the range it had learned from – the size of the improvement narrowed. It still beat the factory polar, by around 7%, but the margin was much smaller than under known conditions. The model was being asked to extrapolate beyond its experience, and the gains shrank accordingly.

That’s not a flaw – it’s a feature of how data-driven models work, since they are most accurate within the conditions represented in their training data.

That’s why continuous data collection matters.

The data loop in action: retraining after Solo Guy Cotten

The first race of the 2026 season, the Solo Guy Cotten, gave us fresh data to fold in. After retraining, the results jumped sharply:

• Average error on Solo Guy Cotten data dropped from 2.91 knots to 0.43 knots.
• The model now explains over 93% of the variation in real boat speed within that regime.

This is also where digiLab's signature uncertainty quantification earns its keep. Alongside each prediction, the model produces a confidence range – an honest estimate of how certain it is. In testing, those ranges held up nearly 95% of the time, meaning Joss can trust not just the predictions, but the uncertainty around them. That's the difference between a number on a screen and a tool you can actually race with.

Looking ahead to the Solitaire du Figaro

The retrained polar is being used to prepare for this weekend's race, and the loop continues. Every race adds new data; every new dataset has the potential to sharpen the next prediction. The closer Joss gets to the Solitaire du Figaro, the more racing experience the model will have to draw on.

For Joss, the value is clear. "What's exciting about working with digiLab is developing tools that help me explore different scenarios ahead of time, so when conditions change at sea I'm reacting from experience rather than guesswork."

As Joss heads into the ultimate solo challenge, he does so with a model that's growing smarter race by race. It's also a good illustration of what digiLab does at the cutting edge of UK sovereign AI: turning real-world complexity into structured, defensible decision support – in offshore racing, and in any domain where the cost of getting it wrong is high.