Updated Daily · All Results Verified

We open our books.

Every pick graded. Every loss counted. Every method documented. No cherry-picking, no hidden losses, no fake screenshots. This is what real AI-powered sports analysis looks like.

Win Rate
0%
Breakeven at -110: 52.4%
Props Analyzed
0
Across 5 sports
AI Agents
0
Independent analysis per prop
Sports Covered
0
NFL · NBA · NHL · MLB · Soccer

Performance by League

Each sport runs its own calibrated analysis pipeline with sport-specific agents. Win rates vary by sport, stat type, and market efficiency.

NFL 2025 Season
55.7%
Picks 5,349
ROI +6.3%
Agents 8
50% breakeven 55.7% actual
NBA 2025-26
56.1%
Picks 13,485
ROI +7.1%
Agents 8
50% breakeven 56.1% actual
NHL 2025-26
61.4%
Picks 11,911
ROI +19.0%
Agents 6
50% breakeven 61.4% actual
MLB 2025-26
67.2%
Picks 14,371
ROI +28.3%
Agents 8
50% breakeven 67.2% actual
Soccer EPL + MLS 25-26
84.0%
Picks 7,089
ROI +60.4%
Agents 7
50% breakeven 84.0% actual

When We Say 60%, We Mean It

A well-calibrated model's confidence scores match real-world outcomes. When our system says "65% confidence," that pick should win approximately 65% of the time.

50-54%
55-59%
60-64%
65%+
Expected (perfect calibration)
Actual outcome rate

Diagonal line = perfect calibration. The closer the pairs, the better our model is calibrated.

NFL Performance by Stat Type

Not all prop types are created equal. Some stats are more predictable than others. Here's where our model finds the most edge - sample sizes included.

Stat Type Win Rate Rate Sample
Rush + Rec Yards
62.3% n=245
Pass Completions
60.1% n=412
Pass Touchdowns
59.4% n=389
Rush Yards
58.7% n=678
Pass Yards
57.2% n=892
Receptions
56.8% n=756
Rush Attempts
54.3% n=234
Receiving Yards
53.9% n=743

The UNDER Edge

Our models consistently find more edge on UNDER picks. This is a structural advantage - the betting public tends to bet overs, creating mispricing on unders.

Over Picks
51.7%
NBA: 2,984 picks graded
Slightly above breakeven. The public bets overs heavily, making the market more efficient on this side.
Under Picks
61.4%
NBA: 2,773 picks graded
Significant edge. The under side is systematically undervalued by the market, creating persistent opportunity.
Why the asymmetry? Sports bettors have a well-documented "over bias" - they prefer betting on players to exceed their lines. This creates a structural inefficiency: sportsbooks shade their lines toward overs, leaving unders mispriced. Our model captures this edge without being deliberately tuned for it. The 9.7 percentage point gap between OVER and UNDER across 5,757 NBA picks is statistically significant (p<0.001).

How Every Pick Is Made

No black boxes. No secret sauce. Here's exactly how our multi-agent analysis pipeline works - from raw data to final confidence score.

01 - DATA COLLECTION
Real-Time Market Data
Player props, odds, and lines are collected from multiple sportsbooks via licensed APIs. Injury reports, lineup confirmations, and game context are gathered from official sources. All data is timestamped and versioned.
02 - AGENT ANALYSIS
8 Independent AI Agents
Each prop is analyzed by 6-8 specialized agents independently. Each agent examines one dimension: matchup quality, injury impact, market efficiency, usage trends, hit rates, and more. Agents score from 0-100 (above 50 = OVER lean).
03 - WEIGHTED ENSEMBLE
Calibrated Combination
Agent scores are combined via backtested weights - not equal weighting. Higher-performing agents get more influence. Neutral signals are dampened. The system applies sport-specific calibration offsets and regression detection to prevent overconfidence.
04 - CONFIDENCE SCORING
Final Output
The combined score maps to a confidence percentage and direction (OVER or UNDER). Higher confidence = stronger signal from more agents in agreement. Only picks above the minimum confidence threshold are surfaced.
Matchup
Defensive DVOA by position
DVOA
Team efficiency z-scores
Line Movement
Sharp money detection
Injury
Direct + cascade effects
No-Vig
True probability extraction
Trends
DB-powered hit rates
Advanced Stats
CPOE, separation, RYOE
Volume
Usage & target share

How We Grade Results

Grading Process

Every pick is graded against official box score results from verified data sources. A pick is graded WIN if the player's actual stat line meets the predicted direction relative to the line at the time of analysis. A pick is graded LOSS if it does not. Pushes (exact line hits) are excluded from win rate calculations.

All statistics on this page reflect out-of-sample performance - these are picks made before game time and graded after games conclude. No picks are retroactively added or removed. Sample sizes are displayed alongside every metric so you can assess statistical significance.

Win rates are calculated as: wins ÷ (wins + losses). ROI assumes flat $100 wagers at -110 standard juice: ((wins × $90.91) - (losses × $100)) ÷ total_wagers.

⚠ Important Disclaimer

Past performance does not guarantee future results. Sports betting involves risk. You should never bet more than you can afford to lose. Propeller provides sports analysis for informational and entertainment purposes only. We are not a sportsbook and do not accept wagers.

A 57.8% win rate, while statistically significant over 3,000+ picks, does not guarantee profitability for any individual bettor. Actual results will vary based on the lines available at your sportsbook, timing of your bets, and variance inherent to sports outcomes. Any form of gambling can be addictive. If you or someone you know has a gambling problem, call 1-800-GAMBLER.

Propeller's analysis represents our model's assessment of player prop outcomes. It is not financial advice, gambling advice, or a recommendation to place any specific wager. All betting decisions are yours alone.

See today's picks.

The numbers speak for themselves. Download Propeller and get AI-powered prop analysis for every game, every sport.

Download for iOS