Competition
Copa Libertadores predictions
This is the competition our model handles worst: 23% tip accuracy across 13 graded picks, against an always-pick-home baseline of roughly 47%. We publish it rather than quietly dropping the tournament.
| Measure | Libertadores | All competitions |
|---|---|---|
| Graded picks | 13 | 157 |
| Tip accuracy | 23% | 43.3% |
| Draw rate | 46% | — |
| Home wins | 23% | — |
| Value bets | 5 | −14.76 P/L |
Nearly half these matches finished level
A 46% draw rate is roughly double a typical domestic league, and draws are precisely what a goals-and-results model predicts worst. Our Pi layer caps its peak draw probability at 0.34 because pushing it higher starts tipping draws outright and loses more accuracy than it gains. In a competition where draws are the modal outcome, that cap is a structural handicap.
Why the format defeats the model
- Two-legged ties change the objective. A side that wins the first leg away is often content to control the second rather than win it. The model predicts a football match; the teams are playing a tie.
- Altitude and travel. A fixture at 2,850m in Quito is a different sport from one at sea level, and a side may have flown five hours to reach it. Nothing in a goals model sees that.
- Squad rotation across competitions. Clubs balance the Libertadores against domestic leagues, so the eleven that produced the form the model fitted is often not the eleven playing.
- Cross-league comparison. Ratings are fitted within a competition. A Brazilian side and a Chilean side arrive with ratings calibrated against different opposition pools.
What we are not doing about it. We are not hiding the competition, and we are not adding a fudge factor to make the number look better. The honest position is that this is a market where our model has no demonstrated edge, and 13 picks is too small a sample to conclude much beyond that. If you were going to weight our Libertadores calls, weight them low.
What would actually help
Not a parameter tweak. Modelling two-legged ties needs tie state as an input — aggregate score, away goals where they apply, which leg this is. That is a different model from the one that runs here, and until it exists these predictions should be read as weak.
Libertadores fixtures appear on the board when the competition is in season. See how every competition compares.
Figures from the forward log, current as of 27 August 2026.