Live model performance
Every pick we've published is tracked. No retroactive edits. No cherry-picking. If the models stop working, you'll see it here before we do.
Updated daily at 4:00 AM ET after overnight results settle • Mock data for layout preview
NBA • 2025-26 season to date
NBA performance
Moneyline BET picks ATS
55.8%
142W–112L on 254 BET-tier picks
LEAN+ picks ATS
53.1%
87W–77L on 164 LEAN+ picks
Closing Line Value (CLV)
+1.42%
avg edge vs close across all picks
Calibration
0.89
Brier-based score; 1.0 = perfect
Units (flat 1u/pick)
+18.6u
254 BET picks, 2025-26
ROI
+4.2%
on $100 flat bets, vigged
Explainers
What these numbers mean
ATS (Against the Spread) percentage
How often our BET-tier picks beat the sportsbook's line. For moneyline, this is win rate including vig. Anything over 52.4% is profitable at standard -110 vig.
CLV (Closing Line Value)
The edge our picks had vs. the line that closed. Positive CLV is the best predictor of long-term profitability — it means we were betting into lines that the market later corrected. Our +1.42% NBA CLV suggests the edge is real.
Calibration score
When we say a bet has a 58% chance to win, does it win 58% of the time across similar predictions? The Brier-based calibration score measures this. 1.0 is perfect; 0.5 is a coin flip dressed up. 0.85–0.95 is where credible models sit.
Flat 1u betting
All performance numbers assume flat 1-unit bets. No aggressive sizing, no pyramiding, no parlays. If you bet $100 on every BET pick, you'd be +$1,860 on NBA after vig, season-to-date.
Weekly recalibration log
Model changelog
Every Monday, we re-tune model weights against the prior week's closing lines. When we make a change, it gets logged here — publicly, with the change and its expected impact.
NBA2026-04-22
Shipped nba_v2. Fixed two bugs found in v1: (1) rest adjustment was chronically positive — 7 of 8 branches added points, only 1 subtracted, producing systematic OVER bias. Rebalanced to be symmetric around zero. (2) No playoff context adjustment; added -5pt correction for playoff games (lower pace + tighter half-court D). Also fixed canonical-key upsert to eliminate duplicate-game rows. v1 track record preserved unedited (32.79% BET ATS) for historical transparency; v2 forward-tracks from today.
NBA2026-04-07
Added playoff season-type flag; rotation data now pulls from playoff splits when active.
NBA2026-03-24
Reduced recency weight on back-to-back games (teams play differently, wider variance).
Source
Where these numbers come from
These numbers come directly from our backtest outputs, which live in our engineering repo. Picks are logged to a versioned table at the time of publication — we can't edit them retroactively. The full methodology explains how predictions become picks.