Show HN: PlayAiOdds – Football probabilities from AI and mathematical models (www.playaiodds.com)

PlayAiOdds (Show HN) publishes weekly “tickets” that pair AI and mathematical models to produce match-by-match probability estimates and suggested multi-leg bets for upcoming football fixtures. Each ticket lists individual picks (win, double-chance 1X/12), modelled probabilities for each outcome (typically 50–72% per pick), the bookmaker-style decimal odds, and aggregated metrics: combined ticket probability (about 8.6–9.1%), composite odds (≈14.8–16.4×), a €10 stake and potential payouts in the €148–€164 range. The site also reports portfolio-level stats such as the probability at least one leg hits (24.25%) versus none (75.75%). For the AI/ML community this is a concrete, production-style use of probabilistic forecasting: ensembles of statistical (Poisson/ELO-like) and ML models appear to be used to estimate event-level win probabilities, which are then aggregated into compound outcomes. Key technical implications include calibration and sharpness of probability estimates, model ensemble and feature engineering (injuries, lineups, form), handling of dependent events across legs, and risk/EV assessment versus bookmaker margins. The project highlights practical concerns—backtesting, Brier/Log scores for evaluation, explainability of model drivers, and ethical considerations around automated betting recommendations and responsible gambling.