← All guides / Game Lab

Game Lab

What it is

Game Lab is the model-driven view of game-level outcomes: spreads, totals, and moneylines for MLB, NBA, NHL, and college basketball. Where All Edges is anchored to sharp books, Game Lab is anchored to our own predictive models. If both agree, you have convergent evidence, which is the strongest signal we publish.

When to use it

How to read it

FieldMeaning
MGSModel Game Score: our internal probabilistic forecast for the matchup. Outputs win prob, projected spread, and projected total.
KenPomFor NCAAB only. We blend KenPom-style efficiency adjustments. KenPom can veto a model lean in close conference games.
Blended ModelThe final projection, combining MGS with KenPom (where applicable) and recent-form weighting.
TierA: model and sharp consensus agree, big edge. B: model agrees, modest edge. C: model agrees but edge is small or noisy.
Projected LineWhat the model thinks the fair line should be.
Market LineWhat the books are actually offering.
EdgeProjected minus market. The size of the model disagreement.

Worked example

Example

NCAAB: Houston -8.5 vs SMU

MGS projection:     Houston -11.2
KenPom adjustment:  Houston -10.4 (defensive efficiency edge)
Blended model:      Houston -10.6
Market line:        Houston -8.5 -110
Edge:               2.1 points
Tier:               A (model strongly agrees, sharp anchor confirms)

Model says Houston should be a 10.6-point favorite. Market has them at 8.5. Two-point edge in college hoops is significant. Tier A because the edge is large, the sharp consensus aligns, and KenPom did not flag a veto.

Common mistakes

  1. Trusting the model when sharps disagree. If the v4 sharp consensus and the model disagree, you have a question, not an edge. Skip it.
  2. Ignoring KenPom vetoes in CBB. KenPom flags conference-dog scenarios where we historically lose. The veto exists for a reason.
  3. Betting before lineups are confirmed. MGS does not know your star was scratched 30 minutes ago. Always check inactives.
  4. Treating projected line as the fair price. The projected line is a midpoint. The book sets juice too. Compare via EV%, not raw points.
  5. Stacking same-game leans. If you bet the spread and the total based on Game Lab, you are double-counting the model's bias.

Related