The open foundations, exposed directly: the 24-state run-expectancy matrix, the
empirical linear weights derived from it, park factors with their leakage rule attached, and the
win-probability calibration. Everything here is computed from the admitted Retrosheet capture
and hash-bound to a model release.
Run expectancy — the 24 base-out states
Bases
0 out
1 out
2 out
Unavailable.
Empirical linear weights — run value by event kind
Each kind belongs to exactly one
responsibility ledger, which is what stops a run of value from being counted twice.
Event
Mean run value
±SE
Events
Ledger
Unavailable.
Park factors — and the rule that governs their use