Losing matters more than winning
Scoring with Scorito
On Scorito you fill in World Cup pools. This project turns that around: lose on purpose, but with scorelines that still look realistic — otherwise it’s obvious you’re wrong.
The idea is simple. First I predict who would win according to statistics: a Poisson model based on FIFA rankings calculates the most likely scoreline. Then I flip it. If the model says 2-1, I predict 1-2. The goal count stays believable; only the winner is wrong on purpose.
Technically I fetch matches, teams, and players via Scorito’s APIs — no scraping, just finding the JSON endpoints through DevTools and wiring them up. Then a pipeline runs: convert rankings to expected goals, pick the best non-draw score, flip it, export to CSV for entry. Champion and top scorers follow the same philosophy: underdogs and anti-favourites.
Much of the boilerplate — CLI, client, tests, project structure — was built with AI. Handy, but it only works if you understand what’s happening yourself. Poisson distributions aren’t magic, and you don’t paste an endpoint from the Network tab blindly without knowing which ID belongs to which tournament. Statistics and APIs — you can’t AI your way around those.