Game total
Research only by policy. In the backtest it is not measurably more accurate than the league-wide rate.
Research only: Tier 3 by policy. Published for context, never as a basis for a decision, until a deliberate product change promotes it. Tier 3: Research only until proven.
The backtest
engine v14 · every graded game · walk-forwardSkill is how much lower the mean squared error of the served number is than always guessing the league-wide rate known at the time. Every date is predicted using only earlier dates, with calibration curves refit each date the way the site serves them.
53 settled games · 24 dates
Counted from the ledgers at this build: predictions frozen before first pitch, under the engine now serving, with the game final and the result in. A game with no logged prediction or result for this market is left out. The count restarts with every engine version. This market is research only by policy, so the count changes no status.
Not measured for this market yet: there is no clean pre-game price sample to compare against.