Improving the +EV Engine

The +EV engine is the heart of our betting analytics — and this week we've been making it smarter.

For those new to the concept: positive expected value (+EV) betting is the foundation of how professional bettors make money over time. When a sportsbook prices a prop at 50% implied probability, but our model calculates the true odds at 55%, that 5-point edge compounds into real profit across hundreds of bets.

What's new this week

We've shipped several improvements to make the +EV board more useful and the signals clearer:

• 1.
  Stat glossary tooltips — Hover over terms like ROI, CLV, and de-vig to see plain-language explanations. No more Googling betting jargon.
• 2.
  Live track record display — We now show both the "stable" record (what we actually surfaced to users) and the backtest reconstruction (current model re-judging historical plays). This transparency lets you see exactly how the model performs.
• 3.
  Signal breakdown — Plays are now tagged by signal type: "Model" (our projection beats the book) vs "Line shop" (consensus/sharp money moving the line). You can see ROI by signal to know which edges are hitting.
• 4.
  Odds bucketing — Plays are grouped by price range (-150 & shorter, pick'em, +120 to +300, longshots) so you can quickly scan for the risk profile you want.

Model calibration

The engine now displays its calibration metrics directly on the page. We show:

• Average predicted probability vs actual hit rate
• Whether the model correction is active or still learning
• Sample size so you know how much data backs the numbers

This matters because a model that predicts 55% but actually hits 48% is leaking edge. Our calibration system adjusts for systematic bias so the probabilities you see reflect reality, not just raw projections.

Beat the close: why it matters

We've added "beat the close" tracking — the percentage of plays where we got a better price than what the line closed at. This is the sharpest signal that an edge is real:

A high beat-the-close rate suggests the model is identifying real inefficiencies, not just noise. It's one of the key metrics professional bettors use to validate their edge.

What's next

We're actively working on:

• Player prop expansion — Adding pitcher props, then NBA player props
• Kelly staking display — Show optimal bet sizing based on edge magnitude
• Historical play lookup — Search past +EV plays and see how they resolved
• Push notifications — Get alerted when high-confidence plays surface

— The Neural Network Team