How We Track Accuracy (And Why Most Services Don't)

The Problem with Sports Betting Analytics

Most prop prediction services share a familiar playbook: show you the wins, hide the losses, and never let you independently verify anything.

We do the opposite.

How the Grading System Works

Our AI model assigns every prop bet a confidence grade from A+ to C based on three independent signals agreeing. The more signals that align, the higher the grade.

The Three Signals

Each pick is evaluated against three independent data sources:

1. Model Projection vs. Line Our LightGBM model analyzes 187+ features per prediction -- season averages, EWMA rolling trends, defense-vs-position matchups, home/away splits, and more. When the model projects a stat significantly above or below the line, that's signal one.

2. L10 Hit Rate How often has this exact prop hit in the player's last 10 games? A player who has gone over 22.5 points in 8 of their last 10 games carries more weight than one who has hit 5 of 10.

3. Line vs. Season Average Is this line set unusually high or low compared to the player's season average? A line set 15% below a player's average creates natural value.

Grade Definitions

A+ (Lock) -- 2+ confluence signals strongly agree. Model edge, hit rate history, and line value all point the same direction. These are the picks shown on our accuracy page and what we tell you to bet. Historical hit rate: 65%+.

A (Strong) -- 1 strong signal or 2 moderate signals agree. High confidence. Available in the full Pro dashboard. Historical hit rate: 55-65%.

B (Lean) -- Model shows moderate edge, limited supporting signals. Worth monitoring but not our core recommendation. Historical hit rate: 50-55%.

C (Speculative) -- Weak or conflicting signals. Minimal model edge. We track these but don't recommend betting them. Historical hit rate: under 50%.

How We Measure Performance

We track four core metrics for every grade tier:

Hit Rate

The percentage of picks that win. We calculate this after every game night and publish it on the accuracy page. At -110 odds, you need 52.38% just to break even.

ROI (Return on Investment)

Profit as a percentage of total wagered. We report two versions:

• Flat ROI: Assumes standard -110 odds on every bet. This is our baseline and makes comparison easy.
• Actual-Odds ROI: Uses the real sportsbook odds from each pick. When we recommend "PTS OVER 25.5" and the best available line is -113, we track profit at -113 -- not an assumed -110. This gives you a more accurate picture of real-world returns.

Standard vs. Alt Lines

We now flag every pick as either a standard line (the primary line most sportsbooks offer) or an alt line (an alternate line that differs from standard). This matters because alt lines often have higher hit rates but lower payouts.

On the accuracy dashboard, you'll see the split displayed as: Std: 731-344 | Alt: 488-107. This lets you evaluate our model's performance on standard lines independently from alt lines.

Units Profit

Total profit in betting units (1 unit = $10 in our tracking). This shows the absolute dollar impact of following our picks.

What Makes This Different

We grade BEFORE the game

Every pick receives its grade when the analysis runs, typically the morning of the game. The grade is locked in before tip-off. No retroactive upgrades.

We track ALL picks, not just winners

Our accuracy page shows every A+ and A pick we made, whether it won or lost. You can see the full record for any time period -- last night, last week, past month, or the full season.

We show actual odds

Most services report results at a flat -110. We go further: every pick in our drill-down shows the actual sportsbook odds it was graded at (e.g., "-113" or "+105"), along with whether it was a standard or alternate line. You see exactly what the real-world bet looked like.

We show the math

The accuracy page breaks down performance by prop type, grade, and time period. You can filter by sport (NBA, NCAAB, NHL), bet side (over/under), and view daily results with individual pick cards.

Anyone can verify

Our picks are timestamped and stored before games start. The results come from official box scores. If we say 8-3 last night, you can expand that day's results and cross-reference every single pick against the box score.

The Honest Truth About Edges

No model wins every bet. Here's what realistic edges look like:

See It Live

Don't take our word for it. Check the numbers yourself:

Updated after every game night. Every pick. Every outcome. Actual odds. No hiding.