V9.4: How Our Latest Model Upgrade Is Delivering Back-to-Back Sweeps

Back-to-back sweep days. No flukes. No lucky bounces. Just a refined model doing exactly what it was built to do.

If you’ve been following Donnie Dimes, you already know — we’ve been on an absolute heater. But what’s behind the recent run isn’t luck. It’s the V9.4 prediction engine, and the data tells the story better than any hot streak ever could.

Every Version Made the Next One Smarter

Building a prediction model is an iterative process. The earlier versions of our engine — V7 through V9.3 — each brought something to the table. V6 introduced structural filters. V7 added projected margin scoring. V8 combined both with a gated approach. Each version taught us what worked, what didn’t, and where to push next.

The biggest breakthrough came when we identified a subtle data leakage issue in our backtesting methodology. The backtests looked strong, but the live results weren’t matching. Rather than ignore the gap, we treated it as a signal — and that’s what led to V9.4.

What Makes V9.4 Superior

V9.4 was born from a massive optimization sweep — 1.87 million parameter combinations tested to find the optimal configuration. No guesswork. No vibes. Pure data science.

The key upgrades over previous versions:

• Rolling efficiency ratings — the model now uses game-by-game rolling statistics with exponential decay, weighting recent performance more heavily while maintaining historical context
• Optimized parameters — calibration factor (k=0.10), 45-day rolling window, strength-of-schedule adjustment, and refined home court advantage modeling
• Bulletproof backtesting — every backtest is now fully rolling, meaning the model only sees data that was available at the time of each prediction. The result: 161-113 (58.8%) with +12.2% ROI — and those numbers are real.
• Intelligent filtering — this is where V9.4 truly separates itself

The Filter Is the Edge

Here’s something most bettors don’t understand: a good model with undisciplined pick selection will underperform. V9.4 generates predictions on every game, but the real edge comes from knowing which predictions are worth acting on.

Our analysis of the live results reveals a clear performance hierarchy:

The sweet spot is clear. When the model identifies a medium-spread edge — where there’s real information asymmetry to exploit — it hits at 70%. Previous versions would have played everything. V9.4 knows when to pull the trigger and when to sit on its hands.

We also found that the 65-78% model probability range is where the money lives. Picks in that zone hit at 68.2% — well above the ~52.4% breakeven needed against the juice. By combining spread size and probability filters, V9.4 isolates the highest-value plays.

From Volume to Precision

The biggest philosophical shift from earlier versions to V9.4 is the move from volume to precision. Previous models cast a wide net — more picks, more action. V9.4 takes the opposite approach: fewer picks, higher conviction, better results.

That’s why the Donnie Dimes channel picks are sweeping. They’re not random selections from a 15-game slate. They’re the filtered output — only the games that fall in the proven sweet spot make the cut.

The Numbers Tell the Story

• V9.4 honest backtest: 161-113 (58.8%), +12.2% ROI
• Filtered sweet spot: 15-7 (68.2%) since going live
• Back-to-back sweep days on the Donnie Dimes channel picks
• NCAAB + NBA coverage — the model works across both leagues

This is what happens when you combine real data science with disciplined pick selection. Every version of the model brought us closer. V9.4 is where it all comes together — the speed, the accuracy, and the discipline to only play the best spots.

We’re not done improving. But right now? The model is locked in. 🔥

Want access to the daily picks before they hit? The Lab is where the model outputs land first — every morning, graded every night.

Written by Donnie Dimes

AI-powered sports predictions. Every pick tracked. Every result graded. Learn more about Donnie →