All Pages

TitleSectionDateDescription
Value Bet IdentificationCompiled2026-06-09Value bet identification involves finding bookmaker odds that are higher than the true probability warrants; the core skill in profitable sports betting
Sources of Betting Edge (Dispersion, Bias, Forecasting)Compiled2026-06-17A betting edge can only come from a price being wrong in your favour, and there are three distinct sources of that — exploiting price dispersion between books (line-shopping), a systematic directional bias in the consensus, or a genuinely superior forecast — each with different mechanisms, requirements, and the same soft-book/account-limiting trap
Kelly CriterionCompiled2026-06-09Formula for sizing bets to maximize long-term geometric growth rate of bankroll; Kelly bet = (bp - q) / b where b=odds-1, p=win probability, q=loss probability
Expected Value (EV)Compiled2026-06-09Expected value in sports betting is the average amount won/lost per bet if the same wager were placed repeatedly; positive EV (+EV) bets are the foundation of profitable betting strategy
Closing Line Value (CLV)Compiled2026-06-09Closing line value (CLV) measures the difference between the odds placed at and the closing odds before the event starts; key metric for validating predictive model quality
Market Efficiency (Sharp vs. Soft Bookmakers)Compiled2026-06-09Market efficiency in sports betting refers to how accurately odds reflect true probabilities; sharp vs. soft bookmakers, favorite-longshot bias, and implications for finding value
De-vigging (Removing the Vigorish)Compiled2026-06-09De-vigging is the process of removing the bookmaker's overround (vigorish) from betting odds to get fair implied probabilities; essential for accurate EV calculations
Beating the Bookies with Their Own Numbers (Kaunitz et al. 2017)Compiled2026-06-17A landmark study showing football bookmakers can be beaten by exploiting the odds themselves — bet outcomes where one book deviates above the multi-book consensus — proving the market is inefficient; and how bookmakers neutralize winners by limiting and closing accounts
Half-KellyCompiled2026-06-09Half-Kelly staking uses 50% of the full Kelly fraction to reduce volatility and account for model uncertainty while still capturing most of the growth benefit
Line Movement & Steam MovesCompiled2026-06-09Line movement and steam moves track how betting odds change over time; steam moves indicate sharp bettor action and can signal market efficiency shifts
World Cup Market Inefficiencies (Recreational Bias)Compiled2026-06-17International tournament betting markets — especially the FIFA World Cup — are less efficient than top domestic leagues because recreational money saturates the pools, producing favorite over-betting and draw/longshot underpricing; the exploitability is real but noisy and tournament-dependent
Poisson-ELO EnsembleCompiled2026-06-09Ensemble approach combining Poisson goal-scoring models with ELO rating systems via weighted average; captures both team quality and historical performance dynamics
Knockout & Tournament ModellingCompiled2026-06-12Extending a per-match W/D/L model to knockout football: 90-minute vs to-qualify markets, extra time and penalty shootouts, the World Cup 2026 format, and Monte Carlo tournament simulation
Multi-Sport Prediction StackCompiled2026-06-09Unified machine learning architecture for predicting matches across football, NFL, and NBA using shared infrastructure with sport-specific model heads
Poisson DistributionCompiled2026-06-09Discrete probability distribution modeling the number of goals scored in a football match given a known mean rate; foundational model for expected goals prediction
ELO to Match Probabilities (W/D/L)Compiled2026-06-12Converting an ELO expected score into three-way win/draw/loss probabilities: the expected-score identity, Davidson's tie model, empirical draw-rate regression, and ELO-to-goal-expectancy mapping
Bayesian Inference in Sports PredictionCompiled2026-06-09Method of statistical inference updating probabilities as new evidence is acquired; foundational to modern sports prediction models combining prior beliefs with observed match data
Concepts IndexCompiled
Dixon–Coles CorrectionCompiled2026-06-09Statistical model extending independent Poisson distribution to better capture correlations in low-scoring football matches; applies correction to 0-0, 1-0, 0-1, 1-1 scorelines
Massey RatingsCompiled2026-06-09Method for measuring past performance of sports teams using a system of linear equations; designed for order rather than prediction; basis for many computer rankings including BCS
Bradley–Terry ModelCompiled2026-06-09Probability model for pairwise comparisons between items/teams; foundational to ELO rating system; used to estimate team strength from head-to-head results
Fitting Poisson Match Models via MLECompiled2026-06-12The Maher-style joint maximum-likelihood recipe for estimating per-team attack/defense parameters, home advantage, and time-decay weights from historical match data; the core fitting procedure for a v1 Poisson model
ELO Rating SystemCompiled2026-06-09Method for calculating relative skill levels of players/teams in zero-sum games via logistic win-probability function; adapted for football, tennis, and American football
Expected Goals (xG)Compiled2026-06-09Statistical metric assigning a probability to each shot in football that it results in a goal; sum of shot probabilities gives expected goals for a team/player
International Football ModellingCompiled2026-06-12What changes when the model targets national teams instead of clubs: neutral venues, sparse and heterogeneous match data, match-importance weighting, and cross-confederation identifiability
Player-Level Team StrengthCompiled2026-06-12Estimating team-level attack/defense parameters by aggregating individual player contributions; roadmap for a v4 (post-tournament) player-level Poisson upgrade
Glicko-2 Rating SystemCompiled2026-06-09Improved rating system over ELO that introduces rating deviation (RD) to measure uncertainty and rating volatility; used in chess, Dota 2, Counter-Strike, and other competitive games
International Results DatasetsCompiled2026-06-12The free GitHub datasets the World Cup model trains on: martj42/international_results (all internationals since 1872, with neutral-venue flag) and jfjelstul/worldcup (deep World Cup history); schemas, traps, and preprocessing decisions
SportradarCompiled2026-06-09Professional-grade sports data provider covering official statistics, live odds, player tracking, and event data for 80+ sports; enterprise pricing typically $1k+/month
OddsJamCompiled2026-06-09Sports betting odds API focused on market efficiency and value identification; $199/mo plan with full historical odds, line movement, and sharp book comparisons
Football-Data.co.ukCompiled2026-06-09Free historical football results, match statistics, and closing betting odds for 22+ European leagues spanning 25+ seasons; the gold standard for free backtesting data
Pinnacle APICompiled2026-06-09Pinnacle Sports API provides sharp bookmaker odds for major sports; free API requires funded account and approval, historically the benchmark for market efficiency
The Odds APICompiled2026-06-09Aggregated sports betting odds API covering 80+ bookmakers, multiple sports, free tier with 500 credits/month, paid plans from $99/mo
API-FootballCompiled2026-06-09Comprehensive football REST API covering 1,200+ competitions, fixtures, match statistics, lineups, player data, and odds; plans from $19/mo
PolymarketCompiled2026-06-09Prediction market for sports, politics, and world events built on Polygon blockchain; public API for market prices, volumes, and settlement
Walk-Forward ValidationCompiled2026-06-09Walk-forward validation is a chronological train/test split where a model is trained on one period and tested on the next; mimics real deployment and prevents look-ahead bias
Brier ScoreCompiled2026-06-09Brier score is a strictly proper scoring rule measuring mean squared error between predicted probabilities and actual outcomes; lower is better, 0 is perfect
Overfitting in Sports ModelsCompiled2026-06-09Overfitting occurs when a sports prediction model learns noise in training data rather than true patterns; prevention requires walk-forward validation, regularization, and small parameter counts
Backtesting FrameworkCompiled2026-06-09Backtesting framework for sports betting models including walk-forward validation, CLV tracking, ROI metrics, and bet-level simulation across historical World Cups
Log-Loss (Cross-Entropy)Compiled2026-06-09Log-loss (cross-entropy) is a scoring rule that heavily penalizes overconfident wrong predictions; used for optimizing probabilistic classifiers
Calibration PlotsCompiled2026-06-09Calibration plots (reliability diagrams) visually compare predicted probabilities vs. actual outcomes; a well-calibrated model shows points on the diagonal
scikit-learn Probability Calibration DocumentationSourcesOfficial scikit-learn documentation on probability calibration; covers calibration_curve function, CalibrationDisplay, and post-hoc calibration methods (Platt scaling, isotonic regression)
Poisson DistributionSourcesDiscrete probability distribution modeling the number of events occurring in a fixed interval, widely used for expected goals prediction in football/soccer
Glicko-2 Original Paper (Mark Glickman)SourcesMark Glickman's original Glicko-2 paper (PDF); complete mathematical specification including the iterative volatility (sigma) estimation algorithm
ML-KULeuven soccer_xg — Open-Source xG ImplementationSourcesML-KULeuven's Python package for training and analyzing xG models using SPADL event stream data; open-source implementation with ML pipelines and feature experimentation
Pinnacle No-Vig Price (NVP) — De-vigging GuideSourcesPinnacle Odds Dropper guide on removing vig from Pinnacle odds; covers the NVP (No-Vig Price) concept and four de-vigging methods including the recommended Power method
SportradarSourcesProfessional-grade sports data provider covering official statistics, live odds, player tracking, and event data for 80+ sports; enterprise pricing typically $1k+/month
Walk-Forward ValidationSourcesWalk-forward validation is a chronological train/test split where a model is trained on one period and tested on the next; mimics real deployment and prevents look-ahead bias
Brier ScoreSourcesBrier score is a strictly proper scoring rule measuring mean squared error between predicted probabilities and actual outcomes; lower is better, 0 is perfect
OddsJamSourcesSports betting odds API focused on market efficiency and value identification; $199/mo plan with full historical odds, line movement, and sharp book comparisons
Value Bet IdentificationSourcesValue bet identification involves finding bookmaker odds that are higher than the true probability warrants; the core skill in profitable sports betting
FiveThirtyEight NBA Elo MethodologySourcesFiveThirtyEight's detailed methodology for their NBA Elo ratings; documents K-factor, home advantage, margin of victory adjustments, and CARMELO extension
Overfitting in Sports ModelsSourcesOverfitting occurs when a sports prediction model learns noise in training data rather than true patterns; prevention requires walk-forward validation, regularization, and small parameter counts
Bayesian Inference in Sports PredictionSourcesMethod of statistical inference updating probabilities as new evidence is acquired; foundational to modern sports prediction models combining prior beliefs with observed match data
Football-Data.co.ukSourcesFree historical football results, match statistics, and closing betting odds for 22+ European leagues spanning 25+ seasons; the gold standard for free backtesting data
Why Most AI Betting Models Fail in FootballSourcesAnalysis of why most AI betting models fail in football; focuses on overfitting to historical patterns that don't persist in live markets
Pinnacle APISourcesPinnacle Sports API provides sharp bookmaker odds for major sports; free API requires funded account and approval, historically the benchmark for market efficiency
Kelly CriterionSourcesFormula for sizing bets to maximize long-term geometric growth rate of bankroll; Kelly bet = (bp - q) / b where b=odds-1, p=win probability, q=loss probability
Poisson-ELO EnsembleSourcesEnsemble approach combining Poisson goal-scoring models with ELO rating systems via weighted average; captures both team quality and historical performance
Expected Value (EV)SourcesExpected value in sports betting is the average amount won/lost per bet if the same wager were placed repeatedly; positive EV (+EV) bets are the foundation of profitable betting strategy
Backtesting FrameworkSourcesBacktesting framework for sports betting models including walk-forward validation, CLV tracking, ROI metrics, and bet-level simulation across historical World Cups
Dixon–Coles CorrectionSourcesStatistical model extending independent Poisson distribution to better capture correlations in low-scoring football matches; applies correction to 0-0, 1-0, 0-1, 1-1 scorelines
Unabated CLV Calculator& EducationSourcesUnabated's CLV calculator and educational content; explains CLV as the expected value yardstick and why positive CLV is the fastest way to validate a betting model
scipy.stats.poisson — Official DocumentationSourcesOfficial SciPy documentation for scipy.stats.poisson — the Python PMF/CDF functions used in all Poisson-based sports models
Pinnacle CLV — Official Educational ArticleSourcesPinnacle's official educational article on closing line value (CLV); explains why the closing line is the market's most efficient price and how CLV measures predictive edge
Massey RatingsSourcesMethod for measuring past performance of sports teams using a system of linear equations; designed for order rather than prediction; basis for many computer rankings including BCS
Market Efficiency (Sharp vs. Soft Bookmakers)SourcesMarket efficiency in sports betting refers to how accurately odds reflect true probabilities; sharp vs. soft bookmakers, favorite-longshot bias, and market efficiency implications for modeling
De-vigging (Removing the Vigorish)SourcesDe-vigging is the process of removing the bookmaker's overround (vigorish) from betting odds to get fair implied probabilities; essential for accurate EV calculations
Maher (1982) — Modelling Association Football ScoresSourcesOriginal 1982 paper introducing the independent Poisson model for football goal prediction; foundational academic reference for the BP (Birnbaum–Poisson) model
Log-Loss (Cross-Entropy)SourcesLog-loss (cross-entropy) is a scoring rule that heavily penalizes overconfident wrong predictions; used for optimizing probabilistic classifiers
The Odds APISourcesAggregated sports betting odds API covering 80+ bookmakers, multiple sports, free tier with 500 credits/month, paid plans from $99/mo
Bradley–Terry ModelSourcesProbability model for pairwise comparisons between items/teams; foundational to Elo rating system; used to estimate team strength from head-to-head results
Forecasting: Principles and Practice — Time Series Cross-ValidationSourcesAuthoritative textbook chapter on time series cross-validation from Forecasting: Principles and Practice (3rd ed); the definitive reference on walk-forward vs. k-fold for temporal data
Closing Line Value (CLV)SourcesClosing line value (CLV) measures the difference between the odds placed at and the closing odds before the event starts; key metric for validating predictive model quality
Dixon–Coles Python Walkthrough (dashee87)SourcesPython walkthrough implementing the Dixon–Coles model from scratch with time-weighting; blog post by dashee87 with complete code and data examples
Upwork Intake: Sports Prediction MVPSourcesProject intake for a World Cup sports prediction MVP combining Poisson expected-goals modeling with live odds feeds, Kelly staking, and Telegram delivery
Half-KellySourcesHalf-Kelly staking uses 50% of the full Kelly fraction to reduce volatility and account for model uncertainty while still capturing most of the growth benefit
Kelly (1956) — A New Interpretation of Information RateSourcesJohn Kelly's original 1956 Bell System Technical Journal paper introducing the Kelly criterion; connects information theory to gambling — 'A New Interpretation of Information Rate
Arpad Elo — The Rating of Chessplayers, Past& PresentSourcesArpad Elo's own book 'The Rating of Chessplayers, Past& Present' (1978); the original authoritative source on the Elo rating system with full mathematical derivation
Elo Rating SystemSourcesMethod for calculating relative skill levels of players/teams in zero-sum games; adapted for sports prediction including football, tennis, and American football
Line Movement & Steam MovesSourcesLine movement and steam moves track how betting odds change over time; steam moves indicate sharp bettor action and can signal market efficiency shifts
Kaggle Walk-Forward Validation NotebookSourcesKaggle notebook on walk-forward validation for time series; practical Python implementation with expanding/rolling windows and financial time series examples
Expected Goals (xG)SourcesStatistical metric assigning a probability to each shot in football/soccer that it results in a goal; sum of shot probabilities gives expected goals for a team/player
All PagesSources2026-06-1996 pages across 2 sections — machine-generated index
API-FootballSourcesComprehensive football REST API covering 1,200+ competitions, fixtures, match statistics, lineups, player data, and odds; plans from $19/mo
Multi-Sport Prediction StackSourcesUnified machine learning architecture for predicting matches across football, NFL, and NBA using shared infrastructure with sport-specific model heads
Glicko-2 Rating SystemSourcesImproved rating system over Elo that introduces rating deviation (RD) to measure uncertainty and rating volatility; used in chess, Dota 2, Counter-Strike, and other competitive games
Calibration PlotsSourcesCalibration plots (reliability diagrams) visually compare predicted probabilities vs. actual outcomes; a well-calibrated model shows points on the diagonal
AboutSources2026-01-01Compiled knowledge base for sports betting models — Poisson distributions, Elo ratings, Kelly criterion, Pinnacle closing lines, and more.
Kelly Criterion Blog (Stanford)SourcesStanford blog walking through Kelly criterion with illuminating examples; explains the relationship to information theory, Bernoulli, and why log-wealth maximization is optimal
Building Reproducible xG Models from StatsBomb Open DataSourcesAcademic paper using StatsBomb open data to build reproducible xG models; compares logistic regression and mixed-effects models, evaluates feature importance
PolymarketSourcesPrediction market for sports, politics, and world events built on Polygon blockchain; public API for market prices, volumes, and settlement
Brier (1950) — Original Paper ReferenceSourcesGlenn Brier's original 1950 paper introducing the Brier score as a proper scoring rule for probabilistic forecasts; published in Monthly Weather Review
Dixon& Coles (1997) — Original PaperSourcesOriginal 1997 Dixon–Coles paper introducing the DC correction for low-scoring football matches; published in Journal of the Royal Statistical Society Series C
Why CLV Is the Sharps' Preferred Metric (Pinnacle Odds Dropper)SourcesPinnacle Odds Dropper article on why CLV is the sharps' preferred metric; explains that the closing line is the sharpest pre-match line and how positive CLV validates a model's edge
FIFA Elo Ranking MethodologySourcesFIFA's official Elo-based ranking system (SUM method) for international football; adapted from standard Elo with weighted match importance, home advantage removed, and larger scale divisor