An honest account of the data, the model, and — importantly — what it does and doesn't beat.
Every FBS roster (136 programs, groups=80) and every skill-position player's per-game log is pulled from ESPN's public college-football feed. Season totals, per-game rates and national ranks are derived from a single canonical source, so the same number appears identically on the player page, the leaderboards and the cheat sheets.
For each player and market we take an exponentially-weighted average of prior-game production (recent games weighted highest), then regress heavily toward a positional baseline. College football has enormous roster churn and a full off-season of staleness between slates, so a single strong prior year is a weak prior — the model leans on the league mean until fresh in-season games accumulate. Yardage markets use a normal distribution with a volatility floor; counting markets (receptions, touchdowns, completions) use a Poisson. From that distribution we compute the probability of clearing each line.
Sports betting markets are efficient. Our projections are published as informational leans, not CLV-validated edges. PrizePicks posts a single line with no two-way price, so there is no de-vig-able market number to "beat" — we never claim a market-beating edge, and we never publish profit, ROI or units unless a track record is genuinely, verifiably profitable. The accuracy page shows calibration only: how well our probabilities match reality.
Once the season opens, every projection is graded against realized game logs on the correct league-local date. A missed game (DNP) is a void, never a loss; a result exactly on the line is a push. Only wins and losses feed the calibration curve.
For entertainment and informational purposes only. 21+. Please bet responsibly.