The project began as a modest experiment on FanDuel, where the author tested a basic baseball lineup and quickly became fascinated by the gaps in player selection that traditional projections missed.
Early on he relied on RotoGrinder projections and skewed normal distributions to model player scores, running 10,000 simulations per slate to uncover the most promising combinations.
A friend’s unexpected win of a $100,000 prize using the same Monte Carlo approach illustrated the method’s profit potential, though the market soon began to converge on similar tactics.
The Correlation Gap
The initial simulator lacked proper player correlation, producing lineups where a high‑scoring quarterback’s receivers underperformed, highlighting the difficulty of capturing inter‑player dynamics.
Back‑Testing and the First ROI Spike
A back‑testing suite compared simulated lineups against historical odds data, delivering a +15% return on investment before the team uncovered a critical source of data leakage.
The Leakage Crash
Correcting the leakage caused the ROI to plummet to -22%, forcing a two‑month pause in development while the model was rebuilt with stricter validation.
Relentless Ambition
Driven by the challenge, the author pushed toward a pitch‑by‑pitch simulator, embracing baseball’s rich data environment despite its computational complexity and the risk of overfitting.
He quit the project twice, each time returning because of the intellectual allure and the desire to achieve academic rigor and market predictability, ultimately shaping a more disciplined research agenda.