Rules of Sport Database

A structured, searchable database of how sport rewrites its own rules, grown from a PhD dissertation into a 64,000-node, 12-sport research corpus.

Overview

Sport governing bodies rewrite their rule books constantly, but no structured dataset existed to study how, when, or why. Adam’s PhD dissertation applied computational text analysis to NFL rule books to answer that question for one league. The Rules of Sport Database generalizes that method into an ongoing, full-scale research database covering football, basketball, soccer, hockey, baseball, golf, bowling, lacrosse, pickleball, and more. In doing so, it proved the dissertation’s computational method holds up well beyond a single league, turning a one-league study into reusable, multi-sport research infrastructure.

What I Built

• Parsed and structured 312 rule book editions across 12 sports, 30 rulebook series, and 18 governing bodies into a normalized SQLite database: more than 64,000 individual rule provisions, roughly 9 million words.
• Built a full-text search layer (SQLite FTS5) so any rule, term, or phrase is searchable instantly across every sport and every year.
• Designed deterministic, sport-specific parsers that turn raw rule book text (PDF, OCR, scanned) into a consistent rule hierarchy (book, rule, section, clause), each checked against independent completeness audits.
• Run a nightly automated pipeline that parses new editions and runs an eleven-point verification battery (referential integrity, full-text-search sync, zero duplicate nodes, containment checks against near-duplicate editions), backing up the database before every change.
• The dissertation’s flagship study applied zero-inflated negative binomial and mixed-effects logistic regression to a corpus of 10,277 NFL rules (2001 to 2022), published as “From Play to Policy” in the Journal of Sport Management. This database extends that method and corpus into a multi-sport, continuously growing research asset.

Tech

Python, SQLite (FTS5), deterministic rule-hierarchy parsers, NLP and text analysis (word2vec, GloVe), R and Mathematica for statistical validation.

Status

Actively growing. Nine of twelve sports are fully parsed to rule level as of mid-2026, and a nightly automated pipeline continues to add editions and run data-quality audits.