Every morning, we run every game through a three-pillar model that quantifies statistical edge, situational context, and what the market is doing. When all three agree, the pick gets more stars. When they disagree, fewer stars. Nothing is subjective.
Different sports settle their lineups at different times, so picks publish on a per-sport schedule. Lines keep moving after publication, so we re-pull the market every couple of hours and update the expected value on the page.
Each game runs through five steps. Steps 1 and 2 happen in parallel for every game on the slate. Steps 3 through 5 happen per game.
For every game on the slate, we fetch the matchup, probable starters or active lineups, recent performance, ballpark and weather (for MLB), bullpen workload (for MLB), bench depth (for NBA), and current odds from every major US sportsbook plus Pinnacle.
The model evaluates the game against three independent groups of signals — statistical, situational, and market — described below. Each pillar produces a directional lean (home / away / neither) and a strength score.
Pillar outputs feed a Bayesian aggregator that produces a probability distribution over outcomes (win probability, projected total). We strip the vig out of the sportsbook's line to get its implied true probability, then compare.
Stars are assigned by how many pillars fired in the same direction and how strongly. One pillar firing alone caps the pick at 2 stars. Two pillars agreeing can reach 4 stars. All three pillars agreeing is required for 5. April and May get an extra one-tier penalty because sample sizes are too small to fully trust.
A language model translates the numbers into the kind of paragraph you'd read in a sharp's newsletter. Stylistic guardrails ban hype phrases, em-dashes, and AI tells. The math drives every claim.
This is the part most sites won't show you. The signals below are the ones currently firing in production. As we add data sources, the pillars get heavier.
Each pillar can fire in favor of either team or stay silent if its signals contradict each other. The signals within a pillar use weights tuned from historical data. The pillars themselves are not weighted equally — the situational pillar earns the most weight when its underlying data is strong (e.g., a confirmed bullpen-taxed game) and less when it isn't.
The 1-5 star rating is calculated, not assigned. Two inputs drive it: how many pillars fired in the same direction (the agreement count) and how big the calculated edge is. Both have to clear a threshold.
Higher stars don't mean a pick is "safe." Variance is built into every bet. Stars are information about how much the model trusts the math underneath.
A 5-star pick can still lose. A 2-star pick is a tighter edge with thinner margin for error. The star rating is information about the math, not a recommendation. Track our actual results on the Track Record page.
Disclaimer: SharpSpots provides educational analysis intended to supplement your own decision-making. This is not financial advice, gambling advice, or a guarantee of outcomes. Past performance does not guarantee future results. 21+ only. If you choose to bet, bet responsibly.
Problem gambling? Call 1-800-GAMBLER or visit ncpg.org.