Each year, I group the top 10 models by MAE — which is a more reliable way to measure long-term performance than number of correct tips — into a special bucket called s10. And that’s the new benchmark for the next year, which powers the headline prediction.
This leverages a “wisdom of crowds” effect to produce tips that tend to be slightly better than those of its contributors. For example, in 2026, s10 beat 26 out of 28 models on MAE; the previous year, it beat 20 out of 25; the year before, 20 out of 27, and so on, pretty reliably back to 2021.
Qualifying for s10 is cut-throat, with some models falling out each year and new models breaking in. This year, five models successfully defended their place and will return for 2027:
- Wheelo Ratings (1st model in 2026 by MAE), lining up for its 5th consecutive year in s10. Wheelo has been absolutely killing it the last few years, and is the only model to beat s10 in three different years. Which slightly undermines the whole “wisdom of crowds” thing.
- Drop Kick Data (4th), returns for a third consecutive year.
- AFLalytics (2nd) was bumped out in 2025 but has otherwise made s10 every year since 2020.
- Winnable (5th) returns for a second consecutive year.
- Glicko Ratings (8th) returns for a second consecutive year and fourth year in total.
Five new models enter:
- plusSixOne (3rd) and Stattraction (6th) both return for the first time since 2021.
- HBin (7th) and Holy Grail Ratings (9th) go straight to the top after successful debut seasons.
- FootyCharts (10th) enters s10 for the first time since its 2024 debut.
Congratulations! My own in-house model missed for the second year in a row after a glorious 8-year stretch, and breaking back in looks daunting. The young models these days, they’re pretty good. In 2027, my model will be renamed from “Squiggle” to “Spline” to better clarify when a tip is coming from the Squiggle platform — and is a combined prediction of many different models — and when it’s coming from my own model.