I have a set of back-testing results and the raw data that fed them: historical match records, detailed player statistics, and archived live-odds feeds. Everything is already cleaned into CSV files and ready for inspection. What I need now is for someone to peel back the curtain on the underlying model. Your first task is to study the data-to-output relationship and pin down the model structure that drives the historical predictions—architecture, workflows, any ensemble logic, and how the odds streams tie in. Feature engineering and hyper-parameter grids are less important; I specifically want the skeleton of the model. Once the structure is clear, you’ll rebuild it from scratch and point it at current-season data so it can generate forward-looking predictions for upcoming ATP & WTA, challenger, and ITF matches. Python with pandas, scikit-learn (or an equivalent ML framework) and Jupyter notebooks will keep things transparent and reproducible, but I’m open to other tools if they reach the same goal. Deliverables • A concise technical brief describing the inferred model structure (diagrams welcome). • Clean, well-commented code that recreates the algorithm and outputs win-probability predictions for the next week’s matches. • A short validation report comparing your model’s out-of-sample accuracy to the historical back-test figures. Acceptance Criteria • Recreated model matches accuracy of the historical back-test and is backtested against predictions made by the model. • If you enjoy cracking open black-box systems and turning them into living, breathing models, this should be a fun challenge.