After my posts earlier this month, I decided to dig a bit more into the “state of the art” in FRC match prediction. As I started to write up a blog post, I realized this would be longer than I had anticipated – so this first part focuses only on Elo.
My main takeaways:
- Statically embedding LaTeX into GitHub Pages is way harder than it should be.
- Roughly speaking, Elo prediction accuracies across a season are in the 60%-70% range.
- Predictions on a single, specific event can be much lower – sometimes worse than a coin flip.
- On balance, Elo predictions don’t get more accurate later in the season (though it did in 2017 and 2019).
You can read it here.
P. S. Questions and constructive feedback are always welcome. I’m not infallible. I’ve tried to be thorough and meticulous – but if there are any mistakes I’ve made, please let me know so I can address them.