Bosh Sports Methodology

1. Market baseline

Where a defensible two-sided market is available, Bosh removes sportsbook hold to estimate a no-vig market probability. That baseline is the starting reference, not an opponent to be ignored.

2. Independent evidence

Research can include model projections, opponent-adjusted performance, role and usage, injuries, pace, weather, matchup metrics, line movement, price, and source-backed context. Correlated sources should not be counted as independent confirmation.

3. Research score vs probability

The research score ranks the strength and completeness of a case. It is not a hit probability. A displayed model probability requires a separate forecasting path and calibration evidence.

4. Player-prop probabilities

Bosh can convert a projected mean and variance into a raw distribution estimate. That raw estimate remains labeled unvalidated until the exact league/prop family has enough settled observations for out-of-sample calibration. The current publication gate requires at least 100 settled observations, validated profile status, and calibration error of 0.05 or better before the raw estimate can be displayed as a calibrated Bosh hit probability.

5. Promotion gates

Candidate research is filtered for price freshness, data quality, role/injury certainty, contradictory evidence, model agreement, uncertainty, and sport-specific feature coverage. A strong narrative alone does not make a candidate an official recommendation.

6. Market context

Opening/current movement, sportsbook price differences, and ticket-versus-money splits can explain market behavior. They are corroborating features, not standalone proof of positive expected value or so-called sharp action.

7. Quote verification

Research cards can expose a quote receipt showing the stored sportsbook, line, American price, capture time, market status, quote age, and opening/current movement when available. A stored quote is evidence of what Bosh observed; it is not a promise that the sportsbook still offers that price.

8. Frozen history and grading

Published research snapshots are written to an immutable decision ledger. Official recommendations are frozen separately and can be graded against outcomes and comparable closing prices. Losing or stale decisions are not supposed to be silently rewritten out of history.

9. Evaluation

Bosh prioritizes out-of-sample calibration, Brier score, log loss, closing-line value, sample size, and realistic tracked ROI. Short streaks are not treated as model validation.