Methodology

We show our work because we believe in transparency. Here's exactly how every prediction is made.

Elo Rating System

Every team starts at 1505. After each game, ratings shift based on the result and margin of victory. Beating a strong team earns more points than beating a weak one.

new_elo = old_elo + K × MOV × (actual - expected)
  • •K-factor: 20 for football/basketball, 4 for baseball (more games = smaller swings)
  • •Home advantage: +65 Elo for NFL, +100 for NBA, +24 for MLB
  • •Margin of victory uses logarithmic diminishing returns — a 30-point win isn't 3x better than a 10-point win
  • •End-of-season regression toward the mean (67% carryover for NFL, 75% for NBA)

Applied to: NFL, NBA, MLB, NCAAF, NCAAB, NCAAW

Efficiency Ratings

Basketball-specific ratings inspired by Ken Pomeroy's methodology. Measures how many points a team scores and allows per possession, adjusted for opponent quality.

AdjEM = AdjO - AdjD (Adjusted Efficiency Margin)
  • •AdjO: Points scored per game, adjusted for opponent defensive strength
  • •AdjD: Points allowed per game, adjusted for opponent offensive strength
  • •Iterative adjustment: recalculated 10 times until values converge
  • •NCAAW uses the identical model as NCAAB — the basketball physics are the same

Applied to: NCAAB, NCAAW, NBA

Monte Carlo Simulation

We simulate the rest of the season 10,000 times using Elo-based win probabilities. This produces probability distributions for playoff odds, seeding, and championships.

P(home wins) = 1 / (1 + 10^(-(elo_diff + home_adv) / 400))
  • •Each simulation randomizes every remaining game using win probabilities
  • •Playoff odds = % of simulations where a team makes the playoffs
  • •Championship odds = % of simulations where a team wins it all
  • •March Madness brackets use the same engine with tournament structure

Applied to: All leagues

Fantasy Projections

Player projections combine recent performance with matchup and injury adjustments. We use exponentially weighted averages so recent games matter more.

projection = baseline × matchup_adj × injury_adj
  • •Baseline: Weighted average of last 10 games (recent games weighted 2x)
  • •Matchup adjustment: ±30% based on opponent defensive rank at the position
  • •Injury adjustment: 0% (out) to 100% (healthy) multiplier
  • •Floor/ceiling: 60% and 150% of central projection
  • •Boom probability: chance of scoring 2x projection

Applied to: NFL (more sports coming)

All predictions are for entertainment and fantasy sports use only.

We log every prediction and publish our accuracy. Check the Model Accuracy page for current results.