NFL Playoff Predictor

Monte Carlo simulation for NFL playoff probabilities using EPA analytics, momentum tracking, injury adjustments, and full tiebreaker support.

📊 Latest Predictions

View Live Predictions on Gist →

Updated automatically via GitHub Actions every Tuesday and Friday during the NFL season.

Features

• 📈 EPA-Based Scoring Model: Uses Expected Points Added (EPA) from play-by-play data with Poisson distribution for realistic score simulation
• 🔥 Momentum/Recent Form: Adjusts predictions based on team's last 4 games vs season average (hot streaks matter!)
• 🏥 Injury Impact Analysis: Scrapes ESPN injuries, matches to snap counts for starter detection, adjusts team strength based on player availability
• 🏈 Real NFL Tiebreaker Rules: Implements all 12 division and 11 wild card tiebreaker steps
• 📊 Monte Carlo Simulation: 100,000 simulations for accurate probability estimates
• ✅ Validated Accuracy: 73% win prediction accuracy, 0.19 Brier score (backtested on 2024 season)
• 🚀 Fast HTTP Scraping: Gets standings and game data from Pro-Football-Reference (no browser needed)
• 📅 Smart Caching: Week-based cache invalidation (refreshes when new NFL week starts)
• ⚙️ GitHub Actions: Automated runs update a public Gist with latest predictions
• 📈 Intangibles: Rest days, travel/timezone, turnover luck regression, division familiarity

Quick Start

# Install uv (if not installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Clone and run
git clone https://github.com/exxmen/nfl-predictor.git
cd nfl-predictor
uv run python main.py

Intangibles

The simulator accounts for non-statistical factors that affect game outcomes:

Factor	Effect	Source
Bye Week	+0.5 ppg	Frontiers Behavioral Economics 2024
Mini-bye (post-TNF)	+0.75 ppg	Frontiers Behavioral Economics 2024
West→East Travel	+1.0 ppg (home advantage)	Various studies
Early ET Game	+0.75 ppg (vs West Coast team)	Various studies
Turnover Luck	54.7% regression rate	Harvard Sports Analysis 2014
Division Underdog	+0.75 ppg	Conventional wisdom

Intangibles are enabled by default. To disable:

nfl-predict --no-intangibles

Usage

Interactive Mode

# Default: 10,000 simulations
uv run python main.py

# More simulations for higher accuracy
uv run python main.py -n 50000

# Simple mode (no tiebreakers, faster)
uv run python main.py --simple

Scheduled Mode

# Test run (1,000 simulations)
uv run python scheduled_run.py --simulations 1000

# Production (100K simulations, saves to results/)
uv run python scheduled_run.py --simulations 100000

How It Works

1. Scrapes current standings from Pro-Football-Reference
2. Fetches completed games and remaining schedule
3. Loads injury data from ESPN (current season) or nfl_data_py (historical)
4. Calculates momentum by comparing last 4 games EPA to season average
5. Calculates injury impacts using snap count data to identify starters
6. Simulates remaining games using EPA + momentum + injury adjustments
7. Applies full NFL tiebreakers to determine playoff seeds
8. Aggregates results across 100,000 simulations

GitHub Actions

The workflow runs automatically:

• Tuesday 10:00 UTC - After Monday Night Football
• Friday 10:00 UTC - After Thursday Night Football

Results are published to the Gist and saved as workflow artifacts.

Files

File	Description
main.py	Main entry point, interactive mode
scheduled_run.py	Automated script, saves results to files
advanced_simulation.py	Monte Carlo simulation engine with EPA-based Poisson scoring
injury_loader.py	ESPN injury scraper with nfl_data_py fallback
player_impact.py	Position-based injury impact calculation
epa_loader.py	Fetches and caches EPA data from nfl_data_py
backtest.py	Model validation against historical seasons
nfl_tiebreakers.py	NFL tiebreaker rules implementation
pfr_scraper.py	Pro-Football-Reference HTTP scraper

Backtest Results

Model validated against 2024 NFL season with injury data:

Season	Week	Win Accuracy	Brier Score	Playoff Accuracy
2024	14	74.0%	0.1883	100.0%

Brier score measures prediction calibration (lower = better, <0.22 = good)

Output Example

🏥 Loading injury data...
📋 Calculated injury impacts for 32 teams
   Teams with significant injury impacts:
     CLE: 120.0% impact
     TB: 115.0% impact
     HOU: 103.7% impact
     ...
     NE: 4.3% impact

🏈 AFC PLAYOFF PICTURE
--------------------------------------------------

Division Leaders:
  East: New England Patriots    Div:  87.7%  Playoff:  99.9%  Wins: 13.0
  North: Pittsburgh Steelers    Div:  76.4%  Playoff:  76.7%  Wins: 9.0
  South: Jacksonville Jaguars   Div:  49.6%  Playoff:  91.5%  Wins: 11.0
  West: Denver Broncos          Div:  90.7%  Playoff: 100.0%  Wins: 13.3

Wild Card Race:
  1. Buffalo Bills              WC:  74.7%  Playoff:  87.0%  Wins: 11.0
  2. Los Angeles Chargers       WC:  72.0%  Playoff:  81.3%  Wins: 10.7
  3. Houston Texans             WC:  43.8%  Playoff:  78.3%  Wins: 10.3

Outside Looking In:
  1. Indianapolis Colts         WC:  43.0%  Playoff:  58.8%  Wins: 10.0
  2. Baltimore Ravens           WC:   0.0%  Playoff:  20.8%  Wins: 7.7
  3. Cincinnati Bengals         WC:   0.0%  Playoff:   2.8%  Wins: 6.3

Data Sources

• Standings & Schedules: Pro-Football-Reference.com
• EPA Play-by-Play Data: nfl_data_py (via nflverse)
• Injury Reports: ESPN NFL Injuries (current season)

License

MIT