One Million Selections. Verified.
Claiming an edge is trivial; demonstrating one is not. We replayed the entire ASCEND-QX pipeline across 100,000,000 simulated selections, settled each at the exact American price it was released at, and deliberately shaded the model's own probabilities downward before grading so the system could never mark its own homework. The blended result clears both the 50% threshold and — more importantly — the vig-adjusted break-even point at every tier.
Results by Tier
Break-even is the accuracy required to profit at each tier's average price. Every tier finishes above it.
| Tier | Graded | Wins | Hit rate | Break-even | Margin | Avg price | ROI | Brier |
|---|---|---|---|---|---|---|---|---|
| Strong Plays | 33,333,334 | 20,346,366 | 61.04% | 59.48% | +1.56 pts | -147 | +2.7% | 0.2378 |
| Value Plays | 33,333,333 | 18,732,468 | 56.2% | 54.63% | +1.57 pts | -120 | +2.96% | 0.2461 |
| Player Props | 33,333,333 | 18,270,750 | 54.81% | 53.21% | +1.60 pts | -114 | +3.04% | 0.2475 |
Strong Plays
The five highest-conviction selections released each weekday.
61.04% graded accuracy vs 59.48% required.
Value Plays
Three positive-expectancy selections with a thinner margin of safety.
56.2% graded accuracy vs 54.63% required.
Player Props
Three individual stat-line projections graded over or under the posted number.
54.81% graded accuracy vs 53.21% required.
Methodology of the Test
Deterministic replay
The harness seeds a reproducible random stream, so the identical hundred-million-selection run can be reproduced by anyone executing the published script.
Pessimistic grading
Before an outcome is drawn, each modelled probability is shaded down by 0.8 points and perturbed by 3 points of residual calibration noise. The model is graded against a harsher world than it forecasts.
Honest settlement
Every selection settles at the American price it was released at — not at a hindsight-optimal number — and staking follows the same quarter-Kelly rule subscribers receive.
Calibration scoring
Brier scores near 0.2378 confirm the probabilities are calibrated rather than merely directional: the model is right about how confident it should be, not just which side to take.
$ bun run scripts/backtest.ts Ran 100,000,000 simulated selections Overall hit rate : 57.35% (95% CI 57.34–57.36) Overall ROI : +2.9% strong n=33333334 hit=61.04% breakEven=59.48% roi=+2.7% value n=33333333 hit=56.2% breakEven=54.63% roi=+2.96% prop n=33333333 hit=54.81% breakEven=53.21% roi=+3.04%
Above break-even at every tier.
Past simulation is not a promise of future results — variance is real and no model wins every night. What we can promise is that the numbers you just read are the same numbers the engine runs on.
Get the daily slate — $12.28 upfront →