Challenge Optimizer
Should you burn one? Set the game state and how sure you are the umpire missed it — the optimizer prices the call, prices the challenge in your pocket, and gives a verdict.
Bottom of the 7th, 1 out, runner on second, down 1, 3-2 — strike called on my hitter, one challenge in hand.
Loading the policy table (about half a megabyte)…
q* = C / (G + C). Values from the reference perception model under optimal continuation; a sharper radar raises early-game challenge values by up to ~10% (q* by a point or two) and barely moves late-game ones; values are rounded to a thousandth of a point. Extras (X) use the “a team at zero regains one each extra inning” rule. Score is clipped at ±4.
How this works
A challenge succeeds with your confidence q and is retained; it fails with 1 − q and is lost. G is the win-probability swing from flipping this call (count-aware, including strike-three and ball-four flips). C is the shadow price of a challenge in hand — what having one more is worth over the rest of the game — solved by backward induction over the 2026 season’s wrong-call arrival rates under a perception model calibrated to how often teams actually catch wrong calls, and cross-checked against an independent replay of real game remainders. Challenge when q ≥ C / (G + C). Full write-up: Should You Burn One?