A quantitative sports-forecasting and stake-governance framework — how it prices games, how it sizes risk, and why it refuses to give advice until the data earns it.
Blue Joker does not assume it can out-predict the market from fundamentals alone. Its working edge hypothesis is market execution latency: retail sportsbooks sometimes lag sharp-market line movements by 2% or more in implied probability. The models exist to estimate fair prices well enough to recognise that gap — and the governance layer exists to prove, or disprove, that the gap is real.
Removes bookmaker margin from quoted odds using Shin's insider-trading model, giving cleaner implied probabilities than proportional normalisation.
Time-decayed bivariate Poisson goal model with a low-score correlation correction for 0-0, 1-0, 0-1 and 1-1 results.
Backs out implied scoring rates from market totals so model and market can be compared on the same scale.
Models dependence between parlay legs (e.g. run line and total in the same game) instead of assuming independence.
Simulates 72-hole outcomes from off-the-tee, approach, around-the-green and putting components, with course-surface variance.
A regularised accumulator that combines ratings, market signals, weather, travel and officiating features into a single win probability.
Stakes are sized with analytical Quarter-Kelly (0.25 × the Kelly fraction). Full Kelly is optimal only when probabilities are known exactly; with estimated parameters it overbets and drifts capital downward. Shrinking to a quarter trades a little growth for far lower variance and drawdown. Multi-leg portfolios are solved as a constrained Kelly problem with total exposure capped at the same 0.25 fraction.
A companion monograph, Non-Commutative Probability in Competitive Dynamics, explores a speculative quantum-probability framing of wagering (including a "Quantum Kelly" idea). It is exploratory theory, not a validated component of the forecasting engine.
Designed and directed by Jakub Rybaczuk (Windsor, Ontario), built in collaboration with AI co-authors Gemini Spark, Claude, Grok and OpenAI — a multi-agent cluster working in symphony.
If this research helped you or sparked an idea, please consider a small tip. It keeps the work going (and probably buys a few baseball cards). Thank you, with love — Jakub.
12hSpkTxK3xSrrWtosct6kgEk7JmmYTpgS