Decoding the Mechanisms of the Hattrick Football Manager Game Using Bayesian Network Structure Learning

Bayesian network structure learning provides a robust framework for reverse-engineering the complex, hidden mechanics of the long-running football management simulation Hattrick. By integrating expert knowledge with data-driven modeling, the research establishes a methodology that successfully decodes the game’s internal engine, which governs scoring probabilities, tactical modifiers, and special event distributions. The primary thesis posits that hybrid models—those combining verified game rules with learned conditional dependencies—outperform purely data-driven approaches, offering both superior predictive accuracy and greater interpretability for tactical decision-making.

The analysis utilizes a dataset of one million match samples to evaluate various constraint-based, score-based, and hybrid learning algorithms. While algorithms such as MMHC and GS excel in structural sparsity, they often produce disjoint subgraphs that limit practical inference. In contrast, HC and Tabu algorithms demonstrate higher F1 and BSF scores, aligning more closely with expert-elicited structures. The resulting KB-probabilistic model, which employs Beta-Binomial distributions to account for event uncertainty, consistently outperforms existing community-built simulators in goal prediction error and Rank Probability Score metrics.

Spanning the global Hattrick ecosystem, this study confirms that midfield and attack ratings remain the primary determinants of match outcomes. Contrary to common community perceptions regarding the game's inherent unpredictability, the findings suggest that Hattrick matches are statistically more predictable than real-world football due to higher goal frequencies. Although the models exhibit some variance when analyzing specific tactics like Long Shots or Play Creatively, the research provides a transparent, publicly available framework for causal discovery in gaming. Ultimately, the study demonstrates that Bayesian networks serve as a powerful tool for demystifying complex game engines, providing a bridge between opaque algorithmic rules and actionable strategic insight.