By Max Gurgel, Vmax Yachting.
I am summarising the key technical points below for the attention of the ITC, the Management Committee and the congress members:
- Class 40 vs XR-41 (2022 vs 2025)
In the 2022 ITC Minutes, the Class 40 was highlighted as proof that the new ANN model improved the agreement with CFD and justified its introduction:
However, as the ITC 2025 minutes now shows (see attached scatter plot extract), the Class 40 was never within the training range of the ANN data.
Despite that, it was used as validation.
Now, three years later, the XR-41 — located in the same region of the parameter space — is labelled as “out of range” and adjusted by blending and hard limits.
This shows that the model was never truly validated: its claimed accuracy was based on incomplete data, and when the gaps became visible, the response was not to retrain or extend the dataset, but to apply ad-hoc parameter limits and blending terms.
Such a process cannot be considered scientific. It undermines ORC’s promise of fairness and transparency. - Instability within the ANN model
The attached Fn-scan, taken at the centre of the ANN parameter space, shows that even there, the three ANNs used by the ITC return significantly different Rr multipliers — one approaching zero at Fn ≈ 0.32.
This indicates that the issue is not limited to extrapolation near boundaries but occurs within the core of the model itself. - Cross-check within the training domain (GS44 vs XR-41 vs X-41)
The attached Excel (“X41_GS_XR scratch.xlsx”) lists time allowances for XR-41 Formula X, GS44 Windwhisper, and old X-41 Dixi under the 2025 VPP. Using the old X-41 Dixi as the reference, two patterns emerge from 10 kts TWS and above:
In VMG upwind, the (older) X-41 must give time to both GS44 and XR-41 – not only to the XR-41!!
On reaching (TWA ≥ 52°) and running, GS44 and XR-41 are faster almost throughout (one GS44 exception at 24 kts).
Since GS44 lies inside the ANN training range (except a few points at very high Fn), the same “apparent anomaly” exists within the data domain. The proposed solution (parameter limits + blending to 2013 outside the domain) therefore does not affect GS44 and cannot address the underlying inconsistency. This strongly indicates that the issue is not an out-of-range problem, and a deeper diagnostic is required before adopting changes. - Warnings since 2023
These exact issues were already raised in my presentation at the ITC meeting in Hamburg (October 2023).
At that time, I explicitly warned that “with the introduction of ANNs we’re leaving the solid ground” and stressed the importance of populating the entire parameter space with valid data to ensure model stability and avoid extrapolation.
The ITC, however, noted in its 2023 minutes that it did not share the concerns raised and expressed confidence that the ANN Rr model was performing well and would continue to be used.
Unfortunately, the very weaknesses I highlighted then — insufficient coverage of the parameter space and instability in the ANN behaviour — are precisely what we are seeing now in the current results.
Perhaps this is the right moment to revisit those warnings before another season passes with the same unresolved systematic problems instead of applying the quick-fixes presented now. - Magnitude and implications of rating changes
Below is an Excel summary of the ITC’s rating deltas for the XR-41 shows very large shifts in VMG upwind and downwind:
We see upwind differences averaging 14 seconds per mile across all wind speeds, and 7 seconds per mile downwind — meaning the XR-41 is effectively penalised by about 10 seconds per mile overall.
For a typical course of one mile with two laps, that’s at least 40 seconds per race.
For a W/L course of 1.5 miles with two laps, thats one minute!
Looking at these numbers doubt that the offshore races have been correctly re-scored.
Interpretation of the results clearly depends on how the list is read:
If the Worlds had consisted only of up-and-down races, and excluding the final race where the GS44 sailed a match race against Dixi, then the GS44 — designed by ITC member Matteo Polli — would have won its fifth consecutive World Championship, with 7 points to 10 for Formula X. - Need for independent review and a reliable long-term solution
Given the complexity of the ANN approach, we strongly recommend an independent scientific review involving external ANN and data-science experts.
Only very few within the ITC — Marcus Mauleverer and Jason Ker, whose pioneering work I sincerely acknowledge — fully master this domain.
Such a review would align with the ORC’s core values of scientific integrity, transparency, and fairness.
Above all, the goal must be clarity and long-term stability, not a short-term “quick fix” that creates new inconsistencies.
I write this with great respect for the ITC’s achievements, but the current evidence calls for a more fundamental and transparent scientific process before further adoption of the ANN-Rr model.
With best regards,
Max