Learn / SHAP Explainability
Why did this AI pick fire? SHAP explainability for sports betting.
Every NuroPicks pick ships with a top-feature attribution. No black box. No mystery. The model shows its work, you decide whether to tail.
Plain-English answer
SHAP, short for SHapley Additive exPlanations, is the math that tells you exactly which features pushed an AI pick toward a side. For every NuroPicks pick we ship the top contributing features as JSON on the picks.shap_top column. A Lakers minus 4.5 spread call is not a black-box yes; it is a 0.18 pace mismatch plus a 0.11 rest edge plus a 0.07 injury delta, with each weight signed and ordered. Sportsbook touts post results without rationale and freely cherry-pick wins. NuroPicks instead publishes the model's attribution alongside the pick so you can audit, reverse-engineer, and disagree before tailing. See every recent attribution at nuropicks.com/record.
Worked example
Tuesday 9:30 PM ET. Lakers at Nuggets, NBA regular season. NuroPicks publishes a pick: Lakers minus 4.5 at minus 110.
The model output is +4.7% expected ROI. Instead of stopping at the call, we attach the SHAP attribution. The top three features that pushed this pick are:
- pace_mismatch: +0.18 Lakers run a 102 pace, Nuggets run 96; Vegas total of 218 misprices the higher-tempo side
- rest_edge: +0.11 Lakers on three days rest, Nuggets on second night of a back-to-back with travel
- injury_delta: +0.07 Nuggets center listed questionable; usage-rate redistribution favors Lakers paint
The 0.36 combined positive contribution is what pushed the model from a roughly neutral spread read to a Kelly-positive call. If you disagree with the rest-edge weight or know the center is now listed probable, you can fade the pick with the model's own arithmetic in your hand.
Every recent attribution lives on /record and on each per-pick page at /record/[id].
Why this is a competitor moat
- No tout posts feature weights. Capper Twitter, Discord picks groups, and the big paid services all post a side and a price. None of them publish what the model actually weighted, and none of them can because they do not run a model in the first place.
- Auditability beats hype. A 60% pace_mismatch hit rate across 200 picks is testable. A vibes call is not. SHAP turns the pick into a hypothesis you can sample-size your way past.
- Paired with immutability and CLV. Picks cannot be edited after publication (database trigger picks_immutable_guard). Closing line value is captured within two minutes of kickoff. SHAP is the third leg: every pick is fully reverse-engineerable, end to end.
- Disagreement is welcome. If the model says rest_edge but you watched the back-to-back team rest its starters in the prior game, the attribution gives you the point of disagreement. Tail or fade with full information.
See it live
Every NuroPicks pick has a public SHAP attribution on its receipt page. No paywall. No edits. Auditable from day one. Pair this with CLV for the full transparency stack.
21+ only. Not financial advice. 1-800-GAMBLER.
Must be 21+. Gambling problem? Call 1-800-GAMBLER. NJ/PA: 1-800-GAMBLER · MI: 1-800-270-7117 · VA: 1-888-532-3500 · CO: 1-800-522-4700 · NY: 877-846-7369 (HOPENY).
NuroPicks LLC · 30 N Gould St Ste R, Sheridan, WY 82801