How EDGcast Works

Team Fingerprinting

Every team gets a 15-17 dimension fingerprint that captures how they play, not just how well. Pace, spacing, defensive pressure, transition tendency, shot selection, rebounding aggressiveness, rim protection. Every dimension is opponent-adjusted and z-scored against the league so you can compare any two teams on any axis.

We also track momentum with a 20-game exponential moving average layered on top of the season fingerprint. This separates what a team is from what a team has been doing lately. Both matter.

Player Fingerprinting

In basketball, every player gets offensive and defensive impact scores adjusted for opponent quality. The top five players on each roster feed into the rankings as a talent signal, weighted by usage.

In baseball, pitchers get a deeper treatment. Every qualified pitcher is rated 0-100 across four pillars: stuff (velocity, spin, whiff rate from Statcast), command (walk rate, chase rate, zone percentage), results (ERA, FIP, opponent-quality adjusted), and durability. Ratings use recency weighting so recent starts matter more than early-season ones.

Position players are profiled through platoon splits, rolling 30-game OPS windows broken out by pitcher handedness. This tells us how a lineup performs against the specific type of pitcher they're about to face.

Rankings

Team strength is a composite of five pillars: adjusted results, strength of schedule, predictive efficiency, roster talent (or rotation strength in baseball), and a consensus signal. The pillar weights shift as the season progresses. Early on, talent and preseason priors carry the load. By midseason, results and efficiency take over. Rankings update daily.

Style Archetypes

A clustering algorithm groups teams into natural play-style archetypes based on their fingerprints. These aren't hand-picked categories. They emerge from the data. The system tracks how archetype pairings have historically performed and uses that as an adjustment when two specific styles collide.

Predictions and Context

The model starts with a baseline margin from team strength, then adjusts for how two specific styles interact, recent momentum, rest advantages, schedule difficulty, and sport-specific context. In baseball that means the starting pitcher matchup, daily bullpen fatigue, platoon splits, and park factors.

Every prediction is walk-forward validated. The model only uses data available before the game. No hindsight. The same code that generates tonight's predictions generated last season's.

Accuracy Tracking

Every prediction is graded the moment a game goes final. We track accuracy across 90+ categories and review where the model has been strong and where it's been weak. The system updates every night: predictions graded, fingerprints recomputed, rankings refreshed, tomorrow's predictions generated from the new state.

See the unfiltered base-model track record — every lean broken out by favorite/underdog and lean size — on the public Model Performance page.