A six-question checklist
Read the probability. Then read everything that tells you how much to trust it.
- What is being predicted? Winner, method, over/under, "at least one"? Different targets have very different reliability.
- What is the probability — and the fair price? Convert with the odds converter if needed.
- What is the edge, and against what? Model probability minus the no-vig market probability, not the raw price. See Vig & fair odds.
- How confident is the evidence? Confidence tier, data tier, sample-size flags.
- What is the publication status? Official, tracking, shadow, validation, or not published.
- What does the model card say? Out-of-sample Brier and log loss against baselines, calibration, and the market comparison.
Edge, done the way the network does it
−130 / +108 → raw implied 56.5% + 48.1% = 104.6% → vig removed → 54.0%. Model 64.0% − market 54.0% = +10.0 points. The market is de-vigged across the observed books (median across books, normalised so both sides sum to 100%), and market prices are compared only after the model has scored the fight — they are never a model input.
On the WNBA card the same figure is explicitly called disagreement with the market, not proven value, because the de-vigged market out-predicted the model on priced holdout games. An edge is a hypothesis to test, never a guarantee.
Publication status
| Status you may see | What it means |
|---|---|
| Official | Locked before the event, graded against the official result, part of the published record. |
| Tracking | Real pregame decisions, frozen and graded, but kept separate from the official record while a model earns promotion. |
| Shadow / validation | The model runs privately; its outputs are not shown as picks. |
| Not published | No PropBetEdge model is making calls for this sport or market right now. |
Status changes as models pass or fail their gates, so Learn does not list the current status of each product. Each product page shows its own. See PBE Picks & track records.
Learn explains how to read outputs; it does not reveal model features or weights, and nothing here is a recommendation to bet.