Model results and validation
SignalXI records published analyses and outcomes to evaluate model behavior using accumulated evidence rather than selected winning examples.
Official track record started
SignalXI official public tracking begins on September 6, 2026 with V13.33.0 Conservative Goal Calibration.
Official statistics start from zero on that date. Earlier records belong to development and internal validation, remain archived, and are not mixed into official performance.
What is recorded
Each published analysis can include fixture, market, estimated probability, audited odds, classification, result and model version. Once settled, those fields allow overall and segment-level evaluation.
Metrics that matter
Hit Rate shows the share of settled selections that won. ROI relates profit or loss to units risked. Units provide a common performance reference, while average odds add context to win rate.
Calibration also matters: a 70% estimate should behave like roughly 70% over a sufficiently large sample if it is well calibrated.
Why sample size matters
Extreme results can appear by chance in small samples. SignalXI separates insufficient sample, observation phase and useful sample before treating a statistical signal as informative enough.
This helps avoid changing the model simply because one segment experienced a short positive or negative run.
Wins and losses are both recorded
Transparency requires both. Historical performance is not presented as a guarantee of future outcomes, and no metric removes the uncertainty inherent in sports events.
View the official track record
The application shows official performance filtered from September 6, 2026 for V13.33.0. Earlier versions and results remain available only as development history.
View official results →