Performance
August 21, 2026
At SignalTurf, we believe an indicator only has value if it can be evaluated.
That’s why the app doesn’t just display predictions. It also publishes statistics that let you observe how the artificial intelligence models behave over the course of many races.
Our goal isn’t to show that a model is infallible, but to provide measurable results so that everyone can assess its performance across a large number of races.
An approach built on transparency
The confidence level and the other indicators shown by SignalTurf come from the analysis of our artificial intelligence models.
Rather than asking users to trust them without evidence, we publish the historical performance observed from predictions actually released before races.
This approach lets everyone verify that the indicators offered are consistent with results observed in the past.
What does performance measure?
The statistics published by SignalTurf are calculated exclusively from analyses actually made available to users.
Depending on the information available in the app, you may be able to check:
- the number of races analyzed.
- the period covered by the statistics.
- the success rate associated with each confidence level.
- the success rate observed when several models converge on the same conclusions.
These indicators let you observe how the models have historically behaved in different analysis contexts.
Success rate by confidence level
The confidence level is designed to help interpret a race.
Historical statistics let you verify how the models have performed when this confidence level was similar.
In other words, they answer a simple question:
Have analyses with a high confidence level actually achieved better results in the past?
The goal is to let everyone evaluate this indicator based on observable results rather than simple claims.
Model convergence
SignalTurf relies on several artificial intelligence models.
When several independent models naturally reach similar conclusions for the same race, this is called convergence.
The app can show the historical performance observed in these particular situations.
These statistics let you study how the models behave when they share the same reading of a race and serve as a complement to the confidence level.
Statistics that keep evolving
SignalTurf’s indicators aren’t built solely from results obtained after the app went live.
They’re first calibrated on an extensive history of races used to train and evaluate the artificial intelligence models.
The statistics shown in the app then gradually grow richer with the results of predictions actually published, allowing for transparent performance tracking over time.
What performance data allows you to conclude
Performance data only describes historical results.
It lets you observe how the models have behaved in the past across different analysis contexts.
However, it never allows you to predict with certainty the outcome of an upcoming race.
Like any sporting competition, a horse race always carries a degree of uncertainty, regardless of the indicators available before it starts.
Go further
Performance naturally complements predictions and the confidence level.
It lets you verify that the indicators shown by SignalTurf are based on observable results rather than simple promises.
The next chapter, Analysis updates, explains why an analysis can change before a race starts and how SignalTurf keeps its information up to date.