Turning it into a pick
The World Cup path and honest limits
5 min
The model handles club leagues and the World Cup a little differently, and it's worth being clear about both — and about where the whole approach is weakest.
The World Cup fallback
National teams aren't in the club-football CSVs, so there's no local history for them. For these fixtures the model falls back to API-Football: it pulls each team's recent results and the head-to-head, and — within a strict per-run budget — fetches per-fixture statistics to recover corners and xG, caching them so later runs reuse them for free. Once enriched, World Cup predictions run through the same goal, corner and Poisson machinery as the leagues, plus the secondary API prediction. Before that first enrichment, corners can show as no data.
The honest limits
No model beats the nature of the sport, and football is hard:
- It's low-scoring and high-variance. Two or three goals decide a match, so a single deflection or red card can overturn a sound read. The model is right about probabilities over many games, not about any one result.
- Corner coverage is patchy. Outside the top European leagues, the free history often has no HC/AC, so corner predictions are simply absent rather than wrong.
- Club xG is thin. The free CSVs carry goals and corners but not xG, so the xG sharpening mostly helps where per-fixture stats were collected — not yet across all club football.
- Daily, pre-game only. The free budget means one read per day and no live updates. A late injury or lineup change after the prediction is made isn't reflected.
The takeaway
This is a transparent, free-data model that's honest about its gaps. It gives you a reasoned pre-game read, not a crystal ball — use it as one input, and lean on the responsible-betting habits in the football betting track.
The model's job is to be calibrated and honest, not infallible. In a sport this random, knowing the limits is part of using it well.