Model Validation · Updated Daily
Model Dashboard
Bankroll growth, closing-line value, edge-tier ROI, and backtested results — the full case for whether the edge is real. How it works →
LIVE POSTED PICKS
Tracked results from picks posted on the site before first pitch.
📊 Live results reflect posted picks only. Backtested validation is shown separately below.
Loading…
$5,000 Bankroll Growth — Flat $100/bet (2%) vs. Fractional Kelly · 2026 Season
Loading…
📈 Closing Line Value — Season Average
Last 30 picks shown · positive = market confirmed the pick
—
Average CLV · all 2026 picks
CLV measures whether the market moves toward our picks by close — the standard test of a real edge. It's currently near breakeven (~50% beat-close) on a small, partly-reconstructed sample; we're tracking it as live closing data accumulates rather than treating it as proven.
📅 Monthly ROI — 2026 Season
Flat $100/bet · consistency matters as much as overall ROI
Loading…
ROI by Model Edge — Does Higher Confidence Pay Off?
Edge Tier Performance — 2026 Season
If the model is well-calibrated, higher-edge picks should pay more over time. The live 2026 tiers aren't cleanly monotonic yet — watch whether higher edge keeps paying as the sample grows
| Tier | Bets | W–L | Win% | ROI | Units |
|---|---|---|---|---|---|
| Populates as picks settle. | |||||
Skill vs. luck — could a no-edge bettor have matched this? ▾
We re-ran our exact — settled bets 10,000 times, keeping every price the same and drawing each outcome at the market's implied win rate — i.e. a bettor with no edge taking these same lines. That no-edge bettor finishes around — on average (middle 90% land between —). We actually finished — — better than — of those no-edge seasons.
—
Our result (units)
—
No-edge median
—
Percentile vs luck
Monte Carlo over our exact bets and prices — only the outcomes are randomized, each at its line's market-implied probability. A genuine same-sample test of selection skill vs. luck; makes no assumptions about games we didn't bet.
BACKTEST
Historical model validation using prior seasons. Not live betting results.
Backtested Model Validation — Out-of-Sample 2022–2024 · Not Live Results
Improving Every Out-of-Sample Year
vs. −4.0% baseline (bet every game, full vig drain) · 7,359 games backtested
+10.4%
Sharpe Ratio
1.82
backtested · annualized
Brier Score
0.234
vs. 0.250 market
Max Drawdown
−12.4u
backtested
Edge Threshold
4pp
min to post
All figures above are out-of-sample backtest results (2022–2024). Each year trained on prior seasons only — no lookahead bias. Live 2026 results appear in the sections above as picks settle.