How the Lyzos Score works.
No black box. No mystique. Here's exactly what goes into every score, and why each factor matters.
The Lyzos Score is a 0–100 number that combines six factors into one signal. The higher the score, the more aligned the underlying data is with the prop pick. It is not a prediction. Past performance never guarantees future results.
The six factors
L5 / L10 trend
Recent games matter more than the season average. We weight the last 5 and last 10 game performances heavily — a player on a hot streak gets credit; a player in a slump gets penalized.
vs. this exact line
How often has the player gone over (or under) this specific number recently? A 7-of-10 hit rate over 24.5 points is a much stronger signal than a season-long average.
Head-to-head history
Some players just have a team's number — and some matchups are nightmares. We pull head-to-head splits going back multiple seasons to flag whether this opponent has historically helped or hurt the player's output.
Venue splits
Home/away differences are real for a lot of players — and they're often baked into how lines are set. We look at venue-specific output to see if the model agrees with the line or finds an edge.
Fatigue penalty
Back-to-backs, short rest, and travel days reduce expected output for most players. We track days of rest leading into the game and apply a fatigue adjustment.
Auto-flagged
Live injury data is pulled on every analysis. A questionable tag, a recent return from IR, or a teammate's absence (which changes usage rates) all get factored in automatically — no stale information.
How the score is built
Each factor produces a sub-score. The sub-scores are weighted (some factors matter more than others depending on sport and stat type) and combined into the final 0–100 number. The weighting is calibrated to reflect what actually predicts player output most strongly in each sport.
Score interpretation
78–100 · Strong Play
All six factors aligned. The data is unusually strong. Rare — about 1 in 20 picks.
64–77 · Good Lean
Most factors aligned. Real edge. Worth a measured play if it fits your bankroll.
50–63 · Lean
Slight edge. Small unit only — these are coin flips with a thin tilt.
0–49 · Skip
Data does not support the pick. The model is telling you to pass — listen to it.
What we don't include
Some "factors" sound smart but don't actually improve predictions. We deliberately exclude these:
- Season averages alone. Too noisy and too slow to react. Recent form does the same job, better.
- Public betting percentages. Useful for market analysis, not for predicting player output.
- Vegas line movement. Tells you what the market thinks, not what the player will do.
- "Hot/cold" narratives from media. If it's not in the box score, we don't trust it.
- Vibes. No vibes.
How we handle goalies and pitchers
Goalies and pitchers don't fit the same model — their usage and stat lines are fundamentally different from skaters and position players. We use modified versions of the model for these positions, with different weights and stat sources. For goalies specifically, we calculate a Fantasy Score using a simplified DraftKings-style formula (saves × 0.6 − goals against × 1.0).
Where the data comes from
All statistical data is sourced from established third-party sports data APIs — currently Tank01 via RapidAPI. We do not fabricate data, scrape sportsbook sites, or use private signals. Everything you see is verifiable from public box scores.
What the Lyzos Score isn't
- It isn't a prediction. Sports outcomes are inherently random in the short term. The model identifies edges, not certainties.
- It isn't financial advice. We don't tell you to bet anything. We show you what the data says.
- It isn't a guarantee. Even an 88-rated play loses sometimes. That's how probability works.
- It isn't a replacement for discipline. The best model in the world won't help you if you over-bet your bankroll or chase losses.
Updates and improvements
The model isn't static. We refine the weights and add new factors over time as we see what works. Major changes are documented on our changelog.
Have a methodology question we didn't cover? Email support@lyzos.net or drop us a line — we love this stuff.