Live Validation Metrics
Research-Grade Transparency
This page documents our data sources, modelling methodology, backtesting approach, and real-time performance validation. We believe transparency builds trust.
Sample Size
0
Settled recommendations
Hit Rate
0.0%
Overall accuracy
High Conf
0.0%
80%+ confidence picks
Profitable
0.0%
ROI > 0 bets
Data Sources & Coverage
Our model is trained on 10+ years of historical racing data and validated against real-time market data. All sources are authoritative and regularly reconciled.
Official Racing Form
Primary
Historical race results, form data, going conditions, and official race classifications.
Coverage
10+ years of UK racing history
Update Frequency
Updated daily post-race
Trainer & Jockey Statistics
Primary
Win rates, place rates, and specialty performance (going, distance, track) sourced from Racing Form.
Coverage
5,000+ active participants
Update Frequency
Updated daily
Real-time Betting Markets
Primary
Live odds aggregation from major UK bookmakers for overlay detection and market sentiment.
Coverage
8 major sportsbooks
Update Frequency
Real-time (60s intervals)
Pedigree Database
Secondary
Sire/dam performance for genetic predisposition signals (distance, going affinity).
Coverage
100,000+ sires tracked
Update Frequency
Static, updated quarterly
International Racing Records
Secondary
International form for horses racing globally (IRE, FRA, AUS, HK, JPN).
Coverage
50+ venues tracked
Update Frequency
Updated daily
Historical Odds Data
Validation
Starting price (SP) and historic odds for backtesting accuracy validation.
Coverage
5+ years SP history
Update Frequency
Complete post-race
Data Quality Assurance
All data sources are validated through automated reconciliation pipelines. We maintain 99.7% data integrity across all racing records. Any anomalies trigger manual review before model ingestion.
Backtesting Methodology & Results
We backtest across 10+ years of racing data using walk-forward validation and blind testing to avoid overfitting. Results below reflect live model performance against unseen data.
Our Backtesting Process
1
1. Data Partition
Historical data split into training (70%), validation (15%), and test (15%) sets.
2
2. Feature Engineering
Extract 200+ features from form data, trainer/jockey stats, market conditions, and pedigree.
3
3. Model Training
Ensemble gradient boosting with cross-validation. No forward-looking bias.
4
4. Blind Testing
Test set predictions evaluated against actual results. No parameter tuning on test data.
5
5. Walk-Forward Validation
Quarterly model retraining. Historical performance tracked separately from live.
Historical Returns by Year
- Returns %
Hit Rate by Period (Sample Size)
- Hit Rate %
- Sample Size
Key Finding
Backtesting shows average hit rates of 54% across 6,000+ unseen races, with returns averaging +15.4% annually. Performance is stable across different market conditions, with higher accuracy on high-confidence selections (80%+ confidence: 62% hit rate).
Confidence Score Methodology
Confidence scores (0–100) reflect the model's conviction. Higher scores indicate stronger statistical evidence of edge. Each factor is weighted by its predictive power.
Form Consistency (30%)
Recent race performance trajectory and consistency patterns.
Example: 3-2-1 form: +8 points
Trainer Expertise (20%)
Trainer win rate for this race class + distance combination.
Example: Class 2 specialist: +6 points
Jockey Fit (15%)
Jockey win rate on this horse + track experience.
Example: Regular partnership: +5 points
Course Fitness (15%)
Historical performance at this specific venue and distance.
Example: Course specialist: +5 points
Market Overlay (10%)
Comparison of model probability vs betting odds. Positive edge increases confidence.
Example: 5%+ positive edge: +3 points
Conditions Match (10%)
Going suitability and distance preference alignment.
Example: Ideal conditions: +3 points
Score Interpretation
High conviction. Strong edge detected across multiple factors. Optimal bet size.
Moderate confidence. Reasonable edge but some uncertainty. Standard sizing.
Marginal edge. Proceed with caution. Reduced sizing or pass.
No meaningful edge detected. Typically not recommended.
CONFIDENCE FORMULA
Confidence = (Form × 0.30) + (Trainer × 0.20) + (Jockey × 0.15) + (Course × 0.15) + (Overlay × 0.10) + (Conditions × 0.10)
Each component normalized to 0–100 scale. Final score capped at 100. Penalty applied for conflicting signals.
Live Validation Metrics (2025 YTD)
Real-time performance tracking across all live recommendations. Updated daily post-race settlement.
Overall Accuracy
0.0%
Across 0 settled recommendations
High Confidence (80%+)
0.0%
High conviction selections significantly outperform
Accuracy Trend (Monthly)
Cumulative Returns (£ from £1 stakes)
Validation Standard
Metrics are calculated post-race against actual results. We use standard SP (starting price) odds and official outcomes. No cherry-picking or survivorship bias. All selections published pre-race with no retroactive filtering.
Model Limitations & Risks
No model is perfect. Below are the key limitations of our approach. We believe transparency about risks is essential for responsible use.
Sample Bias
Model trained primarily on UK racing (70% of data). International performance may be lower. Rare race types (handicaps on novel courses) have limited historical precedent.
Structural Changes
Racing rules, betting markets, and horse populations change. Model performance may degrade if significant regulatory changes occur.
Black Swan Events
Model cannot predict unprecedented events (veterinary incidents, jockey errors, mechanical issues). Confidence scores reflect historical patterns only.
Market Anomalies
Extreme market moves, insider betting, or coordinated betting syndicates can distort odds-based signals. Model does not account for intentional manipulation.
Lag in Feature Updates
Trainer/jockey stats updated daily, but major form changes within 24h may not be reflected. Recent injuries or equipment changes require manual oversight.
Overfitting Risk
Despite walk-forward validation, some parameters may be optimized to historical data. Live performance could differ from backtesting.
Our Mitigation Strategies
Ensemble Approach
Multiple independent models vote on each selection, reducing single-model bias.
Regular Retraining
Model retrained quarterly to capture recent market regime changes.
Manual Review
All high-confidence selections reviewed by specialists before publication.
Confidence Bands
Lower confidence thresholds for uncertain market conditions. No recommendations in extreme volatility.
Risk Disclosure & Legal
Please read carefully. Betting involves financial risk.
No Guaranteed Returns
RaceEdge Pro provides data-driven betting recommendations. Past performance does not guarantee future results. All betting carries risk of total loss. Our model produces probability estimates, not certainties.
Bankroll Management
We recommend never betting more than 2–5% of your total bankroll on a single selection. Kelly Criterion suggests conservative fractional kelly (0.25–0.5 kelly) for long-term viability.
Responsible Gambling
If you experience gambling-related difficulties, support is available:
- • UK: Gamblers Anonymous (0207 384 3040)
- • UK: National Problem Gambling Clinic (020 7054 2000)
- • Ireland: Problem Gambling Ireland (1800 936 722)
- • International: GamCare.org.uk
Disclaimer of Liability
RaceEdge Pro provides information and analysis for informational purposes. We do not provide financial advice. Users assume all responsibility for betting decisions. We are not liable for losses resulting from use of our platform or recommendations.
Odds and Markets
Odds fluctuate. Recommended odds are provided as guidance only. You are responsible for ensuring you receive acceptable odds before placing bets. Early prices shown may not be available at time of placement.
Data & System Limitations
Our data sources, while authoritative, may contain errors or delays. System downtime can occur. We do not guarantee availability or accuracy. Check official sources (Racing Post, Weatherbys) for final confirmation.
Responsible Betting Guidelines
✓ Do
- • Set a monthly betting budget
- • Only bet disposable income
- • Track all bets and results
- • Take breaks if losing
- • Use unit-based staking
✗ Don't
- • Chase losses
- • Bet with borrowed money
- • Increase stakes during downswings
- • Ignore losing streaks
- • Bet when emotionally compromised
You must be 18+ to use this service. RaceEdge Pro is licensed under UKGC regulations and operates responsibly.
This documentation reflects our methodology as of 9/14/2026. We update this section quarterly or when significant changes are made to our modelling approach.
For questions regarding our methodology or data sources, contact our research team. This is not investment advice. Past performance does not guarantee future results.