How Predictions Work
Every prediction combines multiple data signals, including team performance, goaltending, roster health, betting market sentiment, schedule context, and player-level momentum, into a composite view of each team's strength on game day.
Each signal is scored and combined into a composite score from 0 to 100 for each team. The team with the higher composite is our pick, and the gap between the two scores determines confidence.
How It All Comes Together
Data Collection
NHL API, odds, injuries, goalies, lines
Factor Scoring
Multi-signal analysis
Composite Score
Combined team strength
Confidence
50 to 95% rating
Final Pick
Higher composite wins
What the Model Considers
Team Performance
Goal differential is the single strongest predictor of team quality in hockey. Teams that consistently outscore opponents tend to keep doing it. Beyond the net scoring margin, we evaluate shot generation, shot suppression, and special teams efficiency across both power play and penalty kill. Penalty kill is one of the most consistent team-level skills from game to game, while power play carries higher variance but separates elite offenses from average ones.
Goaltending
Goaltending has the largest impact on any single game's outcome. A hot goalie can steal a game any night. That's why we pull the confirmed starting goalie from Daily Faceoff rather than defaulting to a team's listed number one. The announced starter matters, and we weight the signal by how confident the confirmation is. Season save percentage for whoever actually takes the ice, not the roster's primary goalie.
Roster Health and Player Momentum
Injuries are not created equal. Losing a first-line center is far more damaging than losing a fourth-line winger. We use projected line combinations from Daily Faceoff to weight injury impact by lineup position. On the momentum side, we track individual player production against their own season averages at multiple time horizons, weighted by line position. A first line running hot matters more than a fourth line. At the team level, we also track broader form over a longer window to capture coaching adjustments and system changes that individual stats miss. Together, these two layers give a complete picture of recent performance.
Market Signals
Stanley Cup futures odds represent the betting market's aggregate wisdom on overall team quality. Contenders carry a different baseline than rebuilding teams, and futures prices capture information that box scores miss: front office moves, prospect development, coaching changes. We source odds from DraftKings via The Odds API and treat the market as a wisdom-of-crowds input alongside statistical signals.
Schedule and Situational Context
Teams playing on back-to-back nights historically win about 3 to 5 percent less often. Rest advantages, travel fatigue, and the game's place in a road trip all factor into performance. Home ice advantage is real and persistent, with home teams winning roughly 54% of games across the modern NHL. Last change, crowd energy, and familiar ice all contribute.
Home Ice Advantage
Home teams win about 54% of NHL games historically. Last change, crowd energy, and familiar ice all contribute. Our model accounts for this built-in advantage when calculating composite scores.
Confidence Scores
Every prediction includes a confidence percentage from 50% to 95%, based on the gap between the two teams' composite scores.
50 to 59%
Toss-up
60 to 74%
Moderate Edge
75 to 95%
Strong Edge
Over/Under Predictions
Every game includes an over/under prediction on total goals scored.
The Line
The over/under line (e.g. 5.5, 6.0) comes from consensus Vegas odds, averaged across multiple bookmakers via The Odds API. If odds are unavailable, we default to 6.0.
Our Projected Total
We project an expected game total by combining each team's offensive production with the scoring they typically allow. This gives us a data-driven estimate of how many goals the game should produce.
The Prediction
If our projected total exceeds the Vegas line, we predict OVER. If it falls below, UNDER. Confidence scales with how far our projection diverges from the line.
Important distinction: The line comes from Vegas. The over/under call is ours. We use the betting market as a benchmark, then compare it against our own scoring projections from team statistics.
Player Prop Picks
For each game, we surface the single best-value player prop.
Player props are sourced from The Odds API and filtered for quality: minimum 20 games played, 0.3 or more points per game, goal lines capped at 0.5, and maximum odds of +350.
Each prop is scored by expected value, comparing the player's recent statistical average against the bookmaker's line and odds. Established players with strong track records receive a bonus, while lesser-known players are penalized slightly.
Every pick includes a risk level (low, medium, high) and a justification explaining why the prop has value based on recent performance.
Thoughts on how to improve our methodology?