PropEdge

A composite score from 0–100 derived from a weighted multi-factor model that evaluates each bet across seven dimensions: expected value, model-market agreement, sharp book alignment, projection stability, multi-book convergence, data quality, and stat-specific predictability. The weights are calibrated to historical ROI data. Grades range from A+ (strongest) to D (weakest). The "Best Bets" cards surface the day's highest-conviction plays ranked by this score.

The model's point estimate for the player's stat output tonight, generated by a multi-layer regression pipeline. Inputs include recency-weighted game logs (exponential decay), opponent defensive efficiency ratings, pace-of-play adjustments, rest and back-to-back factors, teammate injury reallocation (usage and minutes redistribution), and home/away splits. The projection represents the mean of the model's predicted distribution for that stat.

A Bayesian-adjusted mean that fuses the model's projection with the implied market line. The blending weight is stat-specific: low-variance stats like points anchor heavily to the model's regression output, while high-variance stats like blocks incorporate more market signal to reduce noise. This shrinkage estimator produces a more robust central tendency than either source alone, and is the value used downstream for edge and probability calculations.

The sportsbook's over/under threshold for the prop. For example, a line of 24.5 points means you're betting whether the player scores 25+ (over) or 24 or fewer (under). The value shown is the consensus (median) across all available books, which filters out outlier lines from any single book.

The opponent's defensive rating against this stat category, expressed as a letter grade. Derived from opponent DvP (Defense vs Position) data blended with Synergy play-type matchup analysis. A+ means the opponent is the worst defender against this stat (favorable matchup), while D means the opponent is elite at defending it (unfavorable). Incorporates both season-long and rolling 35-day defensive trends.

The recommended bet direction — OVER or UNDER. On the Sportsbook table, the side is determined by comparing the model's adjusted mean to the line and selecting whichever direction offers the highest EV. On the Longshots table, one-sided alt lines are inherently directional (e.g., over-only at +370).

The consensus (median) line across all books offering two-sided odds for this prop. Shown on the Longshots table to distinguish from the alt line being bet. For example, the standard line might be 24.5 while the longshot alt line is 29.5 — the gap between these indicates how far into the tail the alt line sits.

The delta between the model's adjusted mean and the sportsbook line, expressed in stat units. A positive edge on OVER indicates the model's distribution is shifted above the line. Larger edges correspond to greater detected mispricing.

The no-vig odds implied by the model's probability estimate, representing the theoretically fair price before any sportsbook margin ("juice"). If our fair price is -150 but the book offers -110, the difference is exploitable edge — you're paying less than the true cost of the risk. Negative odds (e.g. -150) mean risking $150 to win $100; positive odds (e.g. +130) mean a $100 bet returns $130 profit.

The expected profit per dollar wagered, expressed as a percentage — the single most important metric for long-term profitability. Calculated as EV% = (P(win) × payout) − (P(loss) × stake). A +5% EV means a $100 bet yields +$5 in expectation over a large sample (law of large numbers). By the Kelly criterion, consistently betting +EV opportunities produces geometric bankroll growth. We surface bets above +3% EV as actionable.

The expected value calculated using the sharp-book devigged probability instead of the model's probability. When Book EV is positive, the sharp market itself considers the book's odds mispriced — meaning even without trusting our model, the efficient market implies +EV. Bets where both the model EV% and Book EV are positive (marked with a ★ star on Best Bet cards) carry the strongest signal, as two independent estimators agree on mispricing.

The source used to derive the market's true probability, following a sharp-book hierarchy. Pinnacle is the primary source — the world's sharpest book with ~2% overround, widely used as the benchmark in quantitative sports modeling. BetOnline serves as the secondary sharp source. Consensus is a fallback that computes the median devigged probability across DraftKings, FanDuel, and Caesars, reducing single-book noise through cross-market aggregation.

The sportsbook offering the highest expected value for this prop. The system evaluates each book's line and odds independently against the model's distribution, then surfaces the one with the maximum EV%. Line shopping across books is one of the highest-ROI habits in sports betting — even a half-point line difference can shift EV by 3–5%.

The specific over/under threshold at the best book. This can differ from the consensus line — e.g., most books post 24.5 but the best book offers 23.5, shifting P(Over) by several percentage points. These cross-book line discrepancies are a primary source of exploitable edge in prop markets.

The American-format odds at the best book. -110 is the standard vig-loaded price (implied 52.4% probability). -105 carries less juice (implied 51.2%). +100 is even money (50%). +150 implies 40% probability — $100 returns $150 profit. The gap between the book's implied probability and our model's probability is what generates EV.

The recommended bet size in dollars, computed using a fractional Kelly criterion scaled to your bankroll (set via the Bankroll $ input). Full Kelly maximizes long-term growth but is aggressive; the system uses a fractional Kelly (typically 25–30%) to reduce volatility. Higher EV and higher confidence produce larger Kelly fractions. The dollar amount updates dynamically as you change your bankroll input.

The expected value component of the confidence score. Derived from the Kelly-implied edge — higher EV signals greater expected profit per unit wagered.

Measures convergence between the model's distribution and sharp-book implied probabilities. When two independent estimators agree on direction, the combined signal follows ensemble prediction principles — reducing variance and producing materially higher hit rates.

The magnitude and directional alignment of the model's edge relative to the devigged sharp-book baseline. Quantifies how much of the detected edge is corroborated by the most efficient market signal available.

A measure of projection reliability based on the recency-weighted coefficient of variation (CV) of the player's recent stat output and minutes. Lower CV (tighter distribution) produces higher stability. Also accounts for sample size — projections built on 15+ games carry more statistical power than those from 3–5.

Cross-book convergence — the fraction of sportsbooks independently showing +EV on the same side. If 5 of 6 books all price the over as +EV, it indicates systematic market mispricing rather than a single-book anomaly, significantly increasing the probability that the edge is real.

Reflects the inherent predictability of the stat category based on its Poisson-like variance profile. Points and rebounds exhibit lower coefficients of variation and approximate normal distributions at typical NBA volumes, making them more modelable. Blocks and steals follow higher-variance, lower-count distributions (closer to Poisson), producing wider confidence intervals and lower stat reliability scores.