TL;DR: The accuracy of your GPS system directly impacts the reliability of an automated start system and the fairness of its OCS calls. Simply put: the more accurate the GPS, the fewer incorrect start decisions the system will make over time. Better accuracy leads to better and more fair decisions—and fewer mistakes across a series. Accuracy matters.
We have spent the last three years working with both eager early adopters and reluctant skeptics as we have developed our automated race starting system. Some people are happy to embrace any solution that ends repeated general recalls and black flags. Others need to see 50 starts without an observable flaw before they will consider automation.
There are some parallels with the adoption of self-driving cars. You could make the argument that we should adopt self-driving cars as soon as they are statistically more safe than human driven cars. But "statistically more safe than a human driven car" just doesn't cut it when a self-driving car makes an obvious error that results in a serious injury or death.
Statistics are abstract. Easily observable anecdotes are far more tangible.
In sailboat racing, the idea of automation replacing the black flag sounds great until you get called over and the boat next to you is declared clear, despite being obviously more advanced. This is why we believe the standard for adopting an automated starting line system must be zero observable errors.
How is GPS Accuracy Defined?
GPS accuracy refers to how closely a measured position matches the true location of an object on Earth. Several factors can affect this accuracy, including the positions of satellites, atmospheric conditions, and the quality of the GPS receiver. The limitations of standard precision GPS are well established. The FAA, self-driving cars, and land surveyors agree that standard precision L1/L5 GPS is not accurate enough for them.
When the US government designed the Global Positioning System (GPS) their target error rate before taking into account device specific variables like antenna size or obstructions of the sky was 2 meters or less. When using this technology, the US government has tested cell phone GPS accuracy to 4.9 meters, and by 2021 had improved dedicated GPS receivers to 0.64 meters. You can read their reports here. When it comes to the GPS receivers used on sailboats, we have observed error rates in this range: between 4.9 meters and .64 meters. Later in this article we will dig into how often a GPS receiver could have no error, and how often the error could be large. For now, let’s go with this quote from gps.gov:
Recent FAA data shows their high quality, single-frequency GPS receivers attaining horizontal accuracy of ≤ 1.82 meters, 95% of the time.
It is a common reaction to these facts about accuracy to assume that all devices in an area are in error by the same amount and to conclude that if all devices are in error by the same amount, the performance of a GPS race management system would be unimpacted. Unfortunately, GPS errors are not the same from one receiver to another. GPS errors are random and uncorrelated because at any given time, of the 20–30 GNSS satellites in view, each device is using a random subset of these satellites to come up with its solution. As a result, even two stationary receivers a few meters apart will see errors in different directions.
The FAA, NASCAR, land surveyors, SailGP and the America’s Cup have deployed Real Time Kinematic (RTK) GPS as a response to this error rate because the 2cm positional error of RTK GPS is 2% of the error rate of conventional GPS.
Normal Distribution and GPS Accuracy
If you log a stationary GPS receiver for 24 hours and plot the results, you won’t get a single dot—you’ll get a scatter plot. In your scatter plot there will be dots that are very close to the true position (0 error), there will be others quite a ways from the true position (say 5m error), and many of them clustered around the true location.
These two dimensional scatter plots can be transformed into radial distributions which can then be graphed as a one dimensional distribution curve. GPS errors follow a normal (Gaussian) distribution, meaning most position readings cluster around the true location, with fewer—but still notable—outliers. For standard GPS systems using L1/L5 bands, this usually results in an error range of ±1 meter. That level of accuracy works fine for many everyday applications, but in the high-stakes OCS or CLEAR world of sail racing, even a one-meter error can make or break a day.
Because of this bell-shaped error curve, most readings are close—but not identical—to the true position, and occasional larger deviations are inevitable. In the curve above, for L1/L5 GPS with 1 meter CEP50, 50% of the time readings are outside of 1 meter of true position and the 95% confidence interval is a radius of roughly 2.5m. In the case of a popular One Design like the Melges 15, 1m is 22% of a boatlength and 2.5m is 55% of a boatlength. For race officials and sailors who value transparency and fairness, that level of uncertainty will generate observably inaccurate outcomes which may not be acceptable.
RTK GPS solves this problem. By adding correction data from a single source of truth into the position calculation for a network of devices, it reduces positional error to the centimeter level, making position readings far more consistent and reliable. This dramatically improves confidence in automated start-line decisions—and helps avoid questionable or outright incorrect OCS calls.
At the Resolute Cup, our RTK system enabled 22 starts for 21 boats—all under a P-flag, with zero general recalls. The outcome was even starts, competitive first beats, and tight top mark roundings, highlighting the impact of accurate technology on race fairness and competitiveness.
What is Circular Error Probable (CEP)?
Circular Error Probable (CEP or CEP50) is a common way to describe GPS precision. It refers to the radius of a circle around the true location where 50% of recorded position points fall. In other words, there’s a 50% chance that any given GPS reading will land outside that circle.
Here is an excerpt from a conventional L1-L5 chipset specification. This specification comes from the product brief for the Telit SE868SY which is representative of a conventional non-RTK multi-frequency and multi-constellation positioning receiver module:
Horizontal Positional Accuracy, L1+L5 (G3BQ): - CEP50 < 1 m
The smaller the CEP, the more precise the system. High-precision systems—like RTK GPS—can achieve a CEP of just a few centimeters. By contrast, standard GPS without corrections may have a CEP of several meters, which makes it far less reliable for things like start-line decisions in racing, where random errors of meters can cause incorrect line calls.
CEP is commonly cited because it produces the smallest radius. But a 50% confidence level is a coin toss. Sailors demand better odds when approaching a start line. Because GPS position stats follow a normal distribution, 65% and 95% confidence intervals can be cited easily, but at the expense of a much larger radius—roughly 2.5 times larger than CEP for R95. The positional accuracy of the Velocitek RTK Puck is 1.8cm CEP and 4.5cm R95.
The RTK Puck: Accuracy in a Single Package
To bring the power of RTK GPS to race committees and sailors, we’ve developed the RTK Puck—a compact, high-accuracy GPS receiver designed for race management applications. By leveraging RTK corrections applied equally to every receiver in a fleet, the RTK Puck provides 1.8 centimeter accuracy, ensuring fair, and reliable start line calls. No more “gaming the system” or second-guessing positions—just pure racing accuracy.
The Impact of GPS Error on Sailor Behavior
As sailors and race committees grow increasingly reliant on GPS starting systems, they are developing habits that influence their decision making at the starting line. Sailors are adjusting their racing strategy to compensate for L1/L5 GPS and starting system errors, often holding back by 3-4 meters to avoid being called OCS due to positional uncertainty. Instead of advancing the sport, this forces competitors to act defensively to compensate for technological limitations rather than racing.
With RTK GPS, and the RTK Puck, sailors don’t have to second-guess their position or back off the line to de-risk GPS error. They can push hard with confidence, knowing that their position is being measured with real accuracy. This leads to fairer, more competitive racing where skill—not uncertainty—determines the outcome.
Implications for Automatic Starting Systems
It’s important to note that everything we’ve discussed so far—the capabilities and limitations of GPS systems—assumes a controlled environment. As soon as you mount the technology on a boat, things change.
On the water, there are a number of new challenges: local magnetic interference from metal hardware, multipath errors caused by reflections off masts, booms, and rigging, and RF interference from carbon fiber sails and other equipment. All of these can degrade signal quality.
That means a system that is “good enough” in the lab might not be good enough in real-world racing conditions—especially in the critical final seconds of a start sequence, when accuracy matters most.
That’s why we designed the RTK Puck as an external antenna, so it can be mounted away from rigging and obstructions with a clear view of the sky. This placement helps preserve the centimeter-level accuracy needed to make confident, reliable automated start calls—even in the most complex race environments.
Final Word: Accuracy Matters.
If we want race management to be automated, trusted, and fair, then accuracy isn’t optional—it’s foundational. The Velocitek RTK Puck is the first product built to meet the standard of zero observable errors.
Ready to see it in action? Get in touch.
Appendixes
Appendix A: Other specifications for GPS Accuracy
| Term | What it Means | Confidence Level | How to Think About It |
| CEP50 | Circular Error Probable - the radius within which half of the measurements fall | 50% | Like a coin toss - half the positions fall outside the circle |
| RMS | Root Mean Square - a calculated average error distance |
63-68% |
About 2 out of 3 points fall inside this radius |
| 2DRMS | Two times the RMS - a larger circle that catches nearly all points | 95-98% | Almost all points land inside this circle |
| R95 | Radius where 95% of points fall | 95% | A tighter statistical version of 2DRMS |
Appendix B: Monte Carlo Analysis to Model GPS Error
To quantify how GPS accuracy impacts start line decisions, we ran a Monte Carlo simulation. This statistical technique uses repeated random sampling to model uncertainty and estimate the probability of different outcomes.
In the case of GPS accuracy, a Monte Carlo simulation can generate thousands of position readings based on a known error distribution. By running these simulations, you can analyze how often a boat's reported position deviates and becomes an incorrect start line call.
We modeled 1000 race starts for a 25-boat fleet of 7-meter One Design boats, with a 210-meter starting line.
We simulated two GPS scenarios:
- L1/L5 GPS with 1m CEP50
- RTK GPS with 2cm CEP50
We included:
- Final Distance to Line modelled off distribution from historical start data and sampled randomly
- Slightly higher density of boats near each end of the starting line
Results:
| L1/L5 GPS (1m CEP50) | RTK GPS (2cm CEP) | |
| Total Number of Errors (per 25,000 start decisions) | 1919 | 1 |
| Fair Start Rate (0 errors in a given start) | 12.0% | 99.9% |
Here are plots of the simulations:
These numbers plots speak for themselves. RTK accuracy doesn’t just improve fairness—it enables trust in the system.