Bankable numbers. A public scorecard.
Every profit figure on Grand Flip Out is realizable — net of the 2% GE tax and capped to what you can actually buy and sell, never the margin×limit fantasy other tools quote. And the advice itself is scored: a daily gp scorecard of our served top-5 against a random pick and a naive spread-flipper, updated hourly — flattering or not.
What we measure — and how
We don't headline a "model accuracy" number, because directional accuracy isn't what makes flipping money — the realizable spread is. What we do publish, labeled by exactly how it was measured: how many items we calibrate live, the backtested slot-strategy win rate, and the conformal band coverage (target ~90%) once enough live intervals resolve — until then it reads building, never a placeholder.
Calibration gauge
ValidatingReliability diagram
Recovery calibrationRecovery-probability calibration: predicted P(recover) on the x-axis vs observed recovery frequency on the y-axis, from out-of-fold (item-grouped) predictions. Dots are sized by bin count. A perfectly calibrated model sits on the dashed diagonal.
Published track record
Real, completed flips — synthetic seed data excluded. We publish full stats at 100 verified outcomes. Until the first real flip resolves, this stays an honest "building" state — no zeros dressed up as a track record.
Advice scorecard — realized gp, daily
Warming| Day (UTC) | Served top-5 | Random-5 | Naive spread | Verdict |
|---|---|---|---|---|
| Warming — every served list is scored ~4 hours later against real market prints; the first full day appears here automatically. | ||||
Track record by grade
| Grade | Resolved | Wins | Win rate |
|---|---|---|---|
| Loading… | |||
Recent verified outcomes
| Item | Grade | Result | Profit |
|---|---|---|---|
| Loading… | |||