How Our +EV Model Finds Value the Market Misses

Every picks service on the internet tells you they have an edge. Most won't show you what that edge actually is.

We will. Here's exactly how our model works — every step, every filter, every reason a pick either makes the cut or gets thrown out.

The Problem With Most Picks Services

The standard approach in the industry is to post picks with no methodology attached. You get a player name, a line, and a thumbs up. No model probability. No market comparison. No explanation of why this pick has value vs. the next one.

That's not an edge — that's noise dressed up as analysis.

Our approach is different. Every single pick we generate comes with a model probability, a market no-vig comparison, a lambda value, and a DraftKings confirmation check. If you want to know why we liked Brunson over 24.5, you can see exactly why down to the decimal.

Step 1: Rolling Average With DNP Filter

The foundation of our player prop model is a rolling average over the last 10 active games for each player.

The key word is active. Most models pull season averages or raw game logs and call it a day. The problem is that box scores include DNP (Did Not Play) rows — games where a player was inactive, sat out with rest, or played garbage time minutes.

We strip every row where a player logged fewer than 1 minute. What's left is a clean 10-game window of real production.

The lambda (λ) you see on every pick card is this filtered rolling average. It's the single number our Poisson model runs on.

Step 2: Opponent DRTG Adjustment

Raw averages don't account for who a player is facing tonight. Scoring 28 against the Pistons is not the same as scoring 28 against the Celtics.

We pull each opponent's defensive rating (DRTG) from Balldontlie's advanced stats endpoint and compute an adjustment multiplier:

adjustment = LEAGUE_AVG_DRTG / opponent_DRTG

This adjustment is clamped at ±20% so one extreme defensive team doesn't blow up the model in either direction. A player facing a bottom-5 defense gets a lambda bump. A player facing a top-5 defense gets a lambda cut.

Step 3: Poisson Distribution

With a cleaned, adjusted lambda in hand, we run a Poisson distribution to convert it into a probability.

Poisson is the right distribution for counting stats — points, rebounds, assists. It models the probability of a discrete number of events occurring over a fixed interval, which is exactly what an NBA game is.

P(Over line) = 1 - Poisson_CDF(floor(line), lambda)

We skip any prop where the line is below our minimum thresholds (8.5 pts, 4.5 reb, 2.5 ast). Below those levels Poisson variance is too wide to trust.

Step 4: 50/50 Market Blend

Here's where most stat-based models go wrong: they trust their own numbers too much.

We blend our model probability 50/50 with FanDuel's no-vig probability:

blended_prob = 0.50 * model_prob + 0.50 * market_no_vig

This pulls our raw model toward market consensus. If our model says 62% but the market says 53%, the blend lands at 57.5%. That humility is deliberate — it filters out false edges where our model is just rediscovering what the market already priced in.

Step 5: Sharp Book Confirmation

The blend gets us to a probability. The confirmation step is what separates real edges from statistical noise.

We cross-reference DraftKings no-vig on every pick. If DraftKings disagrees with our blended probability by more than 5 percentage points, we pass on the pick entirely.

DraftKings attracts the sharpest bettors in the market. When their line diverges significantly from ours, that's a signal we're missing something, not that we found something.

Step 6: +6% EV Threshold

A pick clears all five steps above and still has to clear one more: it needs at least +6% expected value.

EV = (blended_prob × decimal_odds) - 1

Below +6% and the edge isn't wide enough to survive variance over a reasonable sample. Above +6% with sharp confirmation is where genuine long-term profit lives.

Why Transparency Is the Edge

We're not hiding this methodology because there's nothing to hide. The edge isn't in keeping the formula secret — it's in executing it correctly every single night with live data, live market prices, and a consistent set of filters.

Every other picks service charging $30-$100 a month shows you picks without showing you why. We show you the lambda, the DRTG adjustment, the FanDuel no-vig, the DraftKings confirmation, and the exact EV percentage.

If a pick doesn't clear every filter, it doesn't get posted. That's the whole model.

Want to know how to size your bets once you have the picks? Read the Kelly Criterion guide here. Comparing us against other services? See the full breakdown here.