Our approach

Goodkat MMA is built around a simple question: what could a model reasonably have known before a fight?

Start with the timeline

Historical replay evaluates fights in order. A forecast must use information available before its prediction cutoff. Future results must not enter earlier predictions.

Keep the original estimate

Saved forecasts retain their creation time and model context. Reading an estimate on fight day does not mean it was regenerated after the result became known.

Look beyond the headline percentage

Evaluation should include the number of fights, the dates tested and how well confidence matched outcomes. A model that improves on one period may perform worse on another. Past results are not a guarantee of future performance.

Be explicit about gaps

Sources cover UFC and other promotions unevenly. Fighter identity conflicts, missing statistics and incomplete histories can affect estimates. Unsupported matchups may have no published forecast.