How Parlay Buddy Analyzes Your Betting Patterns

We don't predict game outcomes. We analyze how you make picks across 11 iterations of the same parlay to surface decision patterns and biases you can't see yourself.

11 Iterations analyzed per parlay
100+ Behavioral features extracted
0 Team stats used
Model Architecture
Input 11 × n picks
Encoder Transformer
Output Confidence %
Self-attention allows every pick to be compared against every other pick simultaneously

Why Traditional Prediction Models Don't Work

Public data is already priced in

Team efficiency ratings, injury reports, historical matchups, weather conditions—bookmakers have better models for this data than you do. Any edge from public statistics is already reflected in the odds.

Example Your model says "Chiefs -7 wins 58% of the time." The line is -7 because Vegas already calculated that.

Upsets are behaviorally driven

The factors that cause upsets—team motivation, momentum shifts, coaching decisions under pressure—aren't captured in spreadsheets. They're qualitative, contextual, and often reflected in how bettors perceive matchups.

Example Data shows 78% favorite win rate. The 22% of upsets follow patterns in bettor sentiment, not statistics.

Our approach: Analyze the bettor, not the game

Instead of predicting who wins, we analyze how you make decisions. When you're confident, uncertain, risk-seeking, or conservative—these patterns have predictive value that bookmakers don't track.

Insight When you flip a pick between iterations 3 and 7, that uncertainty signal correlates with outcomes.

How the 11-Iteration System Works

One parlay tells us almost nothing. Eleven variations of the same parlay reveal how you think.

1

Select your matchups

Choose 3–8 games you want to include in your parlay. Any sport, any league. The model is sport-agnostic—it only sees your decision patterns.

2

Fill out the same parlay 11 times

For each of the same matchups, pick winners differently across iterations. Mix favorites and underdogs. The variation is the data—we're measuring which picks you're certain about and which you second-guess.

  • Iterations 1–3: Often reflect gut instinct
  • Iterations 4–7: Show reconsideration and hedging
  • Iterations 8–11: Reveal risk tolerance under repetition fatigue
3

Receive pattern analysis

The model outputs confidence scores per leg based on your behavioral patterns, plus insights like "You picked home underdogs 73% less often than away underdogs" or "High flip frequency on Leg 3 suggests low conviction."

Sample Iteration Matrix (3-leg parlay)
Game 1 Game 2 Game 3
Iter 1 F F U
Iter 2 F U F
Iter 3 U U F
... - - -
Iter 11 U F U
Game 1 Flipped 4 times High uncertainty
Game 2 Favorite in 9/11 Strong conviction
Game 3 Underdog trend late Risk escalation

Why We Use a Transformer Encoder

Traditional sequence models (RNNs, LSTMs) process inputs one at a time and lose context over long sequences. Transformers see all 11 iterations simultaneously.

Context Retention by Model Type
RNN
35%
LSTM
55%
Transformer
95%
Context retention measured as ability to relate first-iteration picks to last-iteration picks

Self-Attention Mechanism

The transformer's self-attention allows every pick in every iteration to be compared against every other pick. This captures relationships like:

  • "When you pick Team A as underdog in iteration 1, you flip to favorite by iteration 7 in 68% of cases"
  • "Your Game 2 pick correlates with your Game 5 pick across all iterations (potential anchoring bias)"
  • "Risk escalation pattern: underdog count increases monotonically from iteration 1 to 11"
Sample Self-Attention Weights
I1 I2 I3 I4 I5
Iter 1
Iter 2
Iter 3
Iter 4
Iter 5
High attention Medium Low
Higher attention between iterations 3–5 indicates decision clustering (common pattern)

Features Extracted From Your Picks

We extract 100+ behavioral features from your 11 iterations. Here are the key categories:

Flip Frequency

How often each leg changes between iterations. High flip rate = low conviction. Consistent picks across all 11 = strong belief.

Risk Gradient

Whether your picks get riskier or safer as iterations progress. Most bettors show escalation—more underdogs in later iterations.

Position Anchoring

Whether certain parlay positions (first leg, last leg) always get favorites or underdogs regardless of matchup.

Combination Locking

Picks that always appear together across iterations. These "locked" combinations reveal your core beliefs about correlated outcomes.

Upset Distribution

Where underdogs cluster in your iterations. Front-loaded upsets suggest different psychology than back-loaded ones.

Odds Sensitivity

Whether your picks correlate with line value. Some bettors chase plus-money, others avoid it regardless of edge.

Data We Don't Use

These traditional inputs are excluded because they provide no edge:

  • ×
    Team statistics Points per game, yards allowed, etc. Already priced into odds.
  • ×
    Player performance data Fantasy points, shooting percentages. Public information.
  • ×
    Historical head-to-head Past matchups don't predict future upsets—rosters change.
  • ×
    Injury reports Instantly reflected in line movements.
  • ×
    Weather and venue data Bookmakers adjust lines for these factors.
Traditional Models
Team Stats ML Model Same as Vegas
vs
Parlay Buddy
Your Patterns Transformer Unique Edge

What We're Building Next

Now

Group Competitions

Compare picks with friends on the same games. Private leaderboards, consensus tracking, streak stats.

Q3 2026

Reinforcement Learning Agent

A Q-learning agent that learns your risk/reward preferences and suggests optimal pick combinations.

Q4 2026

Live Pattern Matching

Real-time comparison of in-game instincts vs. pre-game predictions. Track how your live reads differ.

2027

Custom Model Training

Train specialized models for specific leagues, bet types (spreads vs. moneylines), or time periods.

How We Handle Your Data

Pick data only

We store your game selections and iteration patterns. No betting amounts, no sportsbook integrations, no financial data.

Analysis only

Parlay Buddy is a pattern analysis tool. You cannot place bets through our platform. We don't promote gambling.

Your results, no cherry-picking

We display your model's actual hit rate based on your history. Results vary by sport, season, and sample size.

See your betting patterns

Create your first parlay. Takes about 5 minutes.

Create Your First Parlay
Built with PyTorch No team stats required Free to start