Analytics & Forecasting
How do you know a forecast is any good?
Anyone can publish a prediction. The only honest test is whether it holds up when you grade it in public, against a fair benchmark, over a lot of events. So I built one in the open: a model for the 2026 World Cup, scored match by match against the betting market and the big commercial forecasters.
Data Sciencing the World Cup: expected goals, match simulations, and honest predictions, updated live.
The honest scoreboard
Every match, each forecaster's probability gets a Brier score, where lower means sharper and better calibrated. Here's where my model (DSWC Pro) sits against the betting market and the commercial services after 28 graded games.
Snapshot from the live board. My model lands above the betting market and two commercial services, and runs close to the sharpest models in the world. The ranking is earned on graded results, not asserted, and that's the part that matters.
What's under the hood
The dashboard is the visible tip of a full forecasting pipeline, the same kind of system I build for teams that need to predict demand, churn, or revenue.
- Expected goals & team strength from match data, not vibes
- Monte Carlo match simulations that roll up to bracket and tournament odds
- Calibrated probabilities, so a stated 40% wins about 40% of the time
- Live benchmarking against the market, Kalshi, Opta, Dimers and more
- Automatic grading every match, so the model can't hide from its misses
- A decision-ready dashboard that turns the model into something you can read at a glance
The World Cup project is where I keep myself honest: a forecast you can't grade is just an opinion. Live dashboard at worldcupdata.pages.dev, write-ups on Substack. Built by Paul Thibodeau.