Measuring the accuracy of live projections

A projection of 164.9 does not mean the game will finish at exactly 165. The error distribution and the margin against the market line matter. If the model's typical absolute error is 9 points, an edge of +2 should not be presented as strong.

Errors by period and market

MAE should be calculated separately for Q2, H2, Q4 and team totals. Errors depend on time remaining, competition, live-feed quality and the score margin. One overall number hides the most problematic conditions.

Calibrating the decision

A strength rating of 75/100 should correspond to a similar observed frequency in an independent sample. PLAY and LEAN thresholds are checked with walk-forward testing: each game is assessed only with information available before it.

What the model checks

• MAE by market and time
• Edge after allowing for error
• A walk-forward split
• No leakage of the final score

Analysis describes probability and risk; it does not guarantee outcomes. This material is intended for readers aged 21+.