I need an independent ML engineer to audit and possibly fix an existing football/soccer match prediction system. The system is already built in Python/FastAPI and uses API-Football, The Odds API, scikit-learn, XGBoost/LightGBM/CatBoost/PyTorch models, and saved trained model files. The app runs locally and produces predictions, but the outputs look suspicious and I need an expert review before deciding whether to repair it or rebuild the model layer. Current issues found: - Many predictions are repeated with exact 95% confidence. - Batch test showed almost all matches predicted Under 2.5 / Under 3.5 / BTTS No. - Some models fail because live prediction has 103 features while trained models expect 122 features. - Some CatBoost models report missing features. - Some labels fail with previously unseen label errors. - Many fixtures fall back to “Default odds (API unavailable)” instead of real bookmaker odds. - Betting analytics/EV/Kelly staking cannot be trusted until real odds and calibrated probabilities are confirmed. - Need to check for data leakage and whether training/testing was done correctly. What I need: 1. Audit the codebase and trained model artifacts. 2. Identify exactly what is broken. 3. Check if the system is salvageable or if the model/training layer should be rebuilt. 4. Fix feature alignment if possible. 5. Explain and fix the repeated 95% confidence issue. 6. Verify The Odds API integration. 7. Create a simple chronological backtest using unseen matches and real odds where available. 8. Provide a short written report with findings, fixes, and recommendations. Important: I am not looking for someone to promise a magic betting model. I need a realistic ML engineer who understands football prediction, model validation, probability calibration, data leakage, feature pipelines, and backtesting. Please only apply if you can review an existing Python ML project and give honest technical feedback.