We don't guess the numbers. We settle them.

Bonus lands win or lose. We hedge the small qualifier, then convert the bonus bet at long odds with a covering hedge — profit is locked the moment both legs settle.

The first bet is refunded as a bonus if it loses, so we force the safety-net branch and then convert the refunded bonus — a guaranteed lock across both outcomes.

Non-withdrawable credit cleared through a real settled rollover at the offer's true hold. Recorded as an estimate — the bled cost emerges from settlement, not a formula.

Each boost token is played at long odds and fully hedged, so the boosted payout is harvested no matter who wins. The engine derives the boosted return from raw odds + boost %, independently of our math.

The bonus only pays if the first bet wins, so we take the win side and hedge it; the recorded figure is the conservative guaranteed floor, and every run is asserted to land at or above it.

The simulated bettor imports the exact same math the live app uses (lib/calc.ts) to size every back and hedge. No special-cased shortcuts — if the app would tell you to bet it, the simulation bets it.

Each leg is actually placed and settled on SimBook — an independent bookkeeping engine that tracks cash and bonus balances in American odds. It imports none of our math, so settlement is a genuine second opinion, not our formula echoed back.

After both legs resolve, we compare what the app predicted against the cash that actually landed. A run only passes if they match to within a cent (and, for guaranteed structures, never drop below the floor).

Every offer is run 134× per bankroll scenario — randomized game outcomes across 3 bankroll sizes — so the lock holds no matter who wins. Same seed always reproduces the same ledger.