The author built eleven models to forecast the 2026 World Cup, covering every major chapter of a machine‑learning textbook. Using the same 358 historic matches—from the 2010‑2022 World Cups and two European Championships—the suite feeds a vectorized tournament simulator that runs 20,000 iterations per model. The approach replaces a single‑number forecast with a spectrum of opinions.

Three rating systems (Elo, Colley, PageRank) generate team strengths, two goal‑distribution models (Poisson and Negative Binomial) produce expected scores, and five classifiers—logistic regression, K‑Nearest Neighbours, random forest, XGBoost and a small neural net—predict win/draw/loss probabilities from three features. Colley solves a parameter‑free system; PageRank uses a damped walk on Elo scale. The betting market serves as a benchmark, and all models share a calibrated draw curve.

When the eleven engines run the 48‑team draw, they crown four different champions, exposing how model choice swings outcomes by double‑digit percentages. The author argues that this disagreement, not consensus, offers the most actionable insight for bettors and analysts. The Elo‑plus‑Poisson variant emerges as the cleanest single‑forecast option. Its probability outputs integrate seamlessly with the simulator, making it a practical reference for quick predictions.