A transparent look at how Seamheads calculates win probability and identifies edges across all 308 D1 college baseball programs. No black boxes.
Z-Score Engine
4 Weighted Components
308 D1 Teams
01
What the Model Does
Purpose & Scope
The Seamheads model takes publicly available batting, pitching, and team-level stats for all 308 Division I college baseball programs and converts them into a single win probability for every scheduled matchup.
When sportsbook odds are available, the model compares its probability to the market's implied probability to surface potential edges — games where the data suggests more value than the line reflects. It does not predict scores, project player performance, or simulate game outcomes. It answers one question: based on what we can measure, which team has the statistical advantage?
02
How Win Probability Works
From Raw Stats to a Number
Every team's stats are converted to z-scores — how many standard deviations above or below the D1 average they fall in each category. This normalizes everything to a common scale so we can compare pitching talent, offensive production, and overall team quality in the same framework.
The model then evaluates each matchup across four weighted components. The composite matchup score is mapped to a probability using logistic scaling (K = 0.38), meaning a 1.0 standard deviation advantage translates to roughly a 59% win probability. A moderate home-field advantage is applied in log-odds space (~54% baseline). Teams with fewer than 15 games have their stats regressed toward the league mean to reduce small-sample noise. The final probability is always bounded between 12% and 88% — the model never assigns extreme certainty.
Team A
38%
Away
vs
Team B
62%
Home
35%
Team Anchor
Overall team strength measured by run differential per game and RPI value. The most stable predictor of team quality.
Sub-weights: Run Diff 60% · RPI 40%
30%
Pitching
Pitching skill measured by FIP (fielding-independent pitching), K-BB%, and WHIP. Captures true arm talent, not defense or luck.
Sub-weights: FIP 50% · K-BB% 30% · WHIP 20%
25%
Offense
Offensive production measured by OPS, OBP, ISO (isolated power), and plate discipline. Run creation from multiple angles.
Home-field advantage (~54% baseline home win rate) and recent win-loss form. Intentionally low weight to avoid recency bias.
Home field: ~54% baseline · Form: win% scaled
03
How Edge % Works
Model vs. Market
Edge % is the gap between the model's win probability and the market's implied probability (derived from the posted moneyline). If the model says a team wins 62% of the time but the odds imply only 52%, that's a +10% edge.
A higher edge means the model sees more value relative to what the market is pricing. Underdog value is flagged when a plus-money team has at least a 4% model edge — the market is pricing them as longshots, but the data suggests they're closer to a coin flip.
62%
Model Prob
−
52%
Market Implied
=
+10%
Edge
Premium Edge
9%+ edge
Strong Edge
6% – 8.9% edge
Slight Edge
3% – 5.9% edge
04
Key Inputs & Stats Used
What Goes Into Every Calculation
The model ingests season-level stats for all 308 D1 programs. Here's a breakdown of every stat used, organized by component.
FIP is calculated as: ((13 × HR) + (3 × BB) - (2 × SO)) / IP + 3.10
05
How to Read a Matchup Card
Anatomy of a Seamheads Game Card
Every matchup card on the site is built from the model's output. Here's what each element means.
1
Win Probability
The model's calculated chance each team wins. Always adds up to 100%. Bounded 12%–88%.
2
Edge Badge
Color-coded tier label (Premium, Strong, or Slight) showing how much value the model sees vs. the market line. Only appears when odds are available and edge is 3%+.
3
Moneyline Odds
The best available odds from tracked sportsbooks (via The Odds API). Negative = favorite, positive = underdog.
4
Component Breakdown
Bars or values showing each team's strength in the four model components: Team Quality, Pitching, Offense, and Context.
5
Edge Reason Tags
Short labels explaining why the model favors one side: "Pitching Edge," "Team Strength," "Underdog Value," "Home Field," or "Market Discount."
6
Key Stats
Quick-reference team stats like OPS, FIP, RPI, and run differential that inform the model's assessment.
06
Why the Model May Like One Team
Common Edge Drivers
When the model identifies an edge, it's driven by one or more of these factors. The reason tags on each card tell you which ones are in play.
Team Strength
One team's overall quality (run differential + RPI) is significantly higher. This is the most stable indicator and carries the most weight in the model.
Pitching Edge
One team's pitching staff is measurably better by FIP, K-BB%, and WHIP. Pitching advantages tend to be more predictive than offensive advantages in college baseball.
Offense Edge
One team creates runs at a significantly higher rate (OPS, OBP, ISO, plate discipline). Offense is noisier game-to-game but still matters over a sample.
Home Field
The home team gets a small built-in advantage (~54% baseline). This alone won't create a large edge, but it can tip close matchups.
Underdog Value
A plus-money team (market underdog) has a 4%+ model edge. The market is pricing them too low relative to the data. These are often the most profitable spots.
Market Discount
A 9%+ gap between model probability and market implied probability. The market is significantly undervaluing one side, regardless of favorite/underdog status.
07
Important Limitations
What the Model Cannot Do
No model is complete. Understanding the limitations is just as important as understanding the methodology. Here's what this model does not account for.
×
Starting Pitchers
The model uses team-level pitching stats, not individual starting pitcher matchups. A team with a great staff may throw their worst arm on a given day.
×
Injuries & Lineup Changes
No injury data is integrated. If a key player is out, the model doesn't know. Always check lineups before making decisions.
×
Weather & Park Effects
Wind, altitude, temperature, and park dimensions can significantly affect run scoring. These are not factored in.
×
Motivation & Rest
Conference tournament implications, travel fatigue, mid-week vs. weekend rotations, and team morale are invisible to a stats-only model.
×
Early-Season Noise
With fewer than 15 games, stats are regressed toward the league mean. Early-season probabilities are less reliable. The model gets sharper as the sample grows.
×
Thin Odds Markets
College baseball odds are less efficient than NFL or NBA. Lines may be slow to update, only available for select games, or reflect limited action. Edge % is only as good as the odds it's compared against.
08
Frequently Asked Questions
Common Questions
Batting and pitching stats are sourced from NCAA/public college baseball stat feeds and updated regularly throughout the season. RPI values come from publicly available RPI rankings. Live odds are pulled from The Odds API (baseball_ncaa endpoint). Live scores come from ESPN's public API.
The underlying stat data is updated daily during the season. The model recalculates z-scores and probabilities fresh each time the page loads — there's no stale cache. Odds are fetched live from The Odds API each time you visit the Daily Edge or Matchups page.
Sportsbooks don't post lines for every D1 game. Many mid-major and non-conference matchups go unpriced. When odds aren't available, the model still calculates win probability, but edge % and edge tiers can't be shown since there's no market line to compare against.
A 62% win probability means: based on the season-level stats available, if these two teams played many times, the model expects one side to win roughly 62 out of 100 games. It does not mean they will definitely win this specific game. Any single game can go either way.
College baseball has high variance. Even the best team in the country loses to average opponents more often than you'd think. Capping at 88% (and flooring at 12%) reflects this reality and prevents the model from expressing false confidence. No team is ever a 95% lock.
Think of edge tiers as a signal-strength indicator, not a guarantee. Premium Edge (9%+) means the model sees a large gap between its assessment and the market. Strong Edge (6–8.9%) is meaningful but smaller. Slight Edge (3–5.9%) is the minimum threshold — the model sees value, but it's marginal. Higher edge does not always mean higher win probability; it means more perceived value relative to the price.
No. The model operates at the team level only. It doesn't know who's batting third or who's on the mound tonight. Team-level aggregates are more stable and available for all 308 programs, but this is a real limitation — a star pitcher sitting out won't be reflected.
No. Seamheads is an analytics tool that surfaces statistical information. It identifies where the data and the market disagree, but it does not tell you to bet. Any wagering decisions are entirely your own. Please gamble responsibly and within your means.
Important Disclaimer
This model is informational and intended for entertainment purposes only. It is not a guarantee of outcomes. Sports are inherently unpredictable — injuries, weather, lineup changes, and countless other factors can affect any game. No model can account for everything.
Past performance does not predict future results. The presence of an "edge" does not guarantee a win. Use this as one input in your decision-making process, not as the sole basis for any wager. Please gamble responsibly and within your means.
Model version: Skill Model v2 · Logistic K: 0.38 · Home advantage: 0.16 log-odds · Probability bounds: 12%–88% · Small-sample regression: <15 games