EDGE LAB NBA
NBA Game Markets
Equal research coverage for moneyline, spread and total. One NBA game, three independent market comparisons.
Checking live sportsbook data…
Preparing NBA market comparisons…
Loading historical NBA evaluation…
Loading holdout validation…
What do these numbers mean?
Best odds is the most favorable quoted price among selected sportsbooks. Reference line is a real spread or total near the market median. Both sides' prices are compared only at identical lines. No-vig removes the bookmaker margin from a paired sportsbook quote; it is a market-derived probability, not an independent forecast. Model projections use historical team scoring, are experimental and have no verified betting edge. None of the three markets is called a qualified +EV bet without forward validation.
Open legacy research and diagnostic tools
Older model panels are preserved for manual audits. Use Game Markets for the daily NBA slate, Player Props for individual analysis, and Model Research for the three NBA game-market forecasting tests. Other sports are not active projects.
Market analysis center
Checks actual sportsbook data and calculates no-vig probabilities. Positive EV is only calculated when you provide an independent model probability.
Manual NBA Odds Scanner
Scans NBA markets on demand. Odds comparison is automatic; +EV rankings require independently supplied win probabilities. No probability forecasts or confidence scores are invented.
No scan yet.
Individual Game Analysis
Choose a game after a scan or live-odds load. Compare prices across books and calculate EV only with your own independent probability.
EDGE LAB 4.5 · Independent Projection Engine
Uses imported historical final scores and strictly earlier games. The model is a transparent scoring baseline; it does not claim to incorporate pace, injuries, or player availability. All predictions are research estimates, not verified betting edges.
Historical NBA results load automatically when the scheduled collector is configured; manual download remains available, then select NBA and press Run walk-forward test. The model can then preview projections for matching live-odds games. CSV columns: sport,date,home,away,home_score,away_score.
Free historical data starting point: BoxScore Lab (CC BY 4.0; attribution required). Its native CSV format may require conversion into the six-column import format above.
Historical Projection Research · 4.5
Accuracy-first research workspace. Import completed games to calculate strictly out-of-sample score predictions using only earlier games. This is a score-based baseline, NOT an efficiency/pace model. No automated EV or confidence scores are inferred.
Downloads completed NBA games from ESPN's public scoreboard endpoint, season by season. Free, unofficial source; may be unavailable or incomplete. EDGE LAB now specializes exclusively in NBA data.
No download started.
Columns: sport,date,home,away,home_score,away_score. Dates YYYY-MM-DD. Use consistent team names. Import at least 20 earlier appearances per team for a forecast; results are still experimental.
Live games & market lines
Tap Analyze live odds to compare NBA games. A configured The Odds API key is required.
Expected value calculator
Use a probability from an independent model, not the same sportsbook line being evaluated.
These are mathematical estimates, not guaranteed returns. Account for model uncertainty and bankroll risk.
NBA Totals Lab · scoring baseline
Preserved from version 2.1. BALLDONTLIE Free plan may return HTTP 429 or restrict endpoints. This is not yet pace- or injury-adjusted.
Data sources: The Odds API (odds), BALLDONTLIE (NBA baseline). Live lines can change. Player props and automated injury analysis require separate data access. Research tool only.
EDGE LAB 4.7 · Automated Model Audit
Runs a date-ordered, leakage-controlled scoring baseline on the GitHub historical dataset. Each forecast uses only prior games; at least 20 prior appearances per team. Displays monthly error and bias; this does not establish a betting advantage.
Waiting for historical dataset…
Signed error = projected total − actual total. A positive value means overprediction. Historical score averages do not incorporate injury news, market closing lines, or pace-adjusted efficiency.
EDGE LAB 5.3 · Live NBA Totals Forecasts
Automatically loads published historical NBA box scores and available NBA sportsbook totals. Projections use prior completed games only. Model-versus-market differences are research signals, not verified betting edges.
Checking published data…
EDGE LAB 5.3 · NBA Totals EV Lab
Experimental over/under probabilities from chronological possession-model residuals and actual available sportsbook prices. Qualified signals are disabled without archived closing-line validation. A projected point difference alone is not EV. No verified profitability or closing-line validation.
Waiting for evaluation…
Calibration checks use five fixed total thresholds per holdout game (not independent observations). Probabilities assume normally distributed historical forecast errors. Holdout uses fixed totals thresholds rather than archived sportsbook lines. Odds and projections may change. EV is shown for research even when validation fails, but must not be treated as a vetted bet.
EDGE LAB 4.7 · Model Lab & Totals Signals
Chronological NBA totals research. Uses opponent-adjusted recent scoring and a season-to-date league baseline, then compares it with the original scoring model. Estimates over/under probabilities using residuals from earlier forecasts only. An untouched later holdout measures probability calibration. This is a scoring/tempo proxy, not possession-based pace or verified profitable betting.
EDGE LAB 4.7 · Possession & Efficiency Research
Import real team box-score totals for historical NBA games. Possessions are estimated as FGA − OREB + turnovers + 0.44 × FTA. Offensive rating is points per 100 estimated possessions. No possession figures are inferred from final scores.
A row represents one completed game. Use full ESPN team names matching the historical results. Box scores are also collected incrementally by GitHub Actions when the provider supplies complete statistics. Manual import remains available. For trustworthy pace testing, include several seasons of box scores.
EDGE LAB 5.3 · NBA Accuracy Lab
Independent chronological evaluation of a possession/efficiency model versus a score-only baseline. Uses only prior dates, separates training and later holdout, and does not infer profitability from final scores.
Ready to evaluate published NBA box scores.
Model is frozen for this evaluation. Training and holdout are ordered by date. This version does not add closing-line data or qualify EV recommendations.
EDGE LAB 5.3 · NBA Model Optimization
Independent research comparison: choose a scoring/possession blend and bias correction using only earlier validation dates, then evaluate once on later holdout dates. This does not change the live EV engine or qualify betting signals.
Selection and holdout are split on complete dates, not randomly. Candidate choice is made without seeing the final holdout. These diagnostics are not evidence of positive sportsbook EV; historical closing lines remain required.
EDGE LAB 5.4 · Data Integrity & Coverage
Checks the existing automated NBA historical-results and box-score feeds. Detects missing box scores, duplicate game IDs, score mismatches, stale coverage, and suspicious month gaps. This is a diagnostics-only upgrade: live projections, odds and EV calculations remain unchanged.
Coverage uses only completed games present in the published historical-results file as its denominator; it is not a full NBA schedule audit. Months with 0 published games are not treated as missing automatically. Different team-name formats are normalized where possible. This audit never unlocks EV betting recommendations.