A backtest replays the model against historical matches to measure how it would actually have performed, not on the data it learned from, but on matches it had never seen. It is the honest test of whether an edge is real or just curve-fitting.
Out-of-sample and walk-forward
The key discipline is out-of-sample testing: a model is only ever scored on matches dated after the data used to train it. Smashrs uses a walk-forward design. Train on everything up to a point in time, predict the next slice, roll forward, and repeat. This mimics real life, where you only ever have the past to predict the future, and it prevents the most common way backtests lie: leaking future information into past predictions.
What we report
A trustworthy track record reports more than a win rate. It includes accuracy, return on investment, and compounding outcomes (final bankroll, drawdown) across multiple seasons, because a flattering ROI can still hide brutal losing streaks.
See the live, continuously updated numbers on the predictions page, broken out by tour (ATP, WTA, and ITF). Each pick there carries its model edge and consensus.