ROI: Simulated return on stakes for model-graded bets by entry time, at the 3rd-best price available at entry.
CLV: simulated entry price vs the de-vigged closing line — positive means the entry beat the market's final word.
Basis: Predictions are the model's walk-forward (no look-ahead) probabilities, and every game is staked and graded with the Kelly parameters that were actually in force on its own date — never today's parameters applied to the past. The window starts where the dated parameter archive starts.
Note: These figures are a simulated backtest under the honest basis described above, not delivered-bet results — honest numbers run lower than the re-scored figures previously shown here, by design. Why do lower grades look the best at some times? The datasets here are still relatively small and sometimes a few good results can skew their results. The takeaway should not be the absolute number at a specific time point, but the direction and consistency of the pattern. Why do we recommend Cs? Cs are matches that are identified as value but are either longer shots or otherwise less likely. They are technically value but worth more of your own research before betting.