1. Historical evidence
The model evaluates matchup history and current context, including expected run environment, starting-pitcher information, bullpen availability, contact quality, lineup context, and workload.
2. Identity before prediction
Players, games, markets, and scheduled starts must resolve to stable identities. If a starter cannot be identified reliably, ClearInning does not invent one. The game is excluded unless a separately validated starter-agnostic model applies.
3. Price-aware value
A high win rate can still lose money when the prices are heavily juiced. Every result belongs beside the price that was actually published. Model probability, implied probability, and expected value are checked together before ranking.
4. AI checked
AI checked: data quality, identity consistency, and risk are reviewed. They do not manufacture missing starters or guarantee an outcome.
5. The primary risk
Every analysis states a meaningful way the expected game shape could fail. Risk is not hidden in fine print, and uncertainty is not replaced with guarantees.
6. Publication timing
The source receipt time and scheduled first pitch are preserved. A selection received before its scheduled start can remain eligible even if publication follows shortly after that scheduled time, but the timing is disclosed. A selection received at or after its scheduled start is excluded.
7. The ledger
Results reconcile against the original selection and published price. Wins, losses, pushes, voids, pending results, and published-price ROI remain part of the same accountable record. To inspect the fields in an actual edition, follow the pick-audit guide beside the result ledger.
8. Corrections
Material corrections identify what changed rather than silently replacing the historical record. Questions may be sent to [email protected].