Strategy Simulator
Replay the model's entire public ledger — every surfaced pick, settled against the real box score — under your rules. Flip between the raw model prob and the isotonic-calibrated one, drop the families the ledger proved are losers, size flat or Kelly, and watch what the equity curve would have done. No synthetic data anywhere on this page.
Read this first: this is an in-sample historical replay — the calibration layer was learned from these same outcomes, and picking filters after seeing results is hindsight. It demonstrates what the honesty layer learned; it is not a forward return promise. A win pays at the pick's recorded fair price (loss −1u), so "everything, flat" reproduces the public track record exactly.