A story about an aggressive defense is not a forecast until it changes a number. Matchup modeling forces that step. The lesson breaks production into opportunities times output per opportunity, so 11 possessions at 2.3 points gives 25.3, while slower pace with better efficiency lands at 24.48, not an automatic over. You pick features that have a causal reason and a time lag, shrink early estimates toward a prior, split data by date and freeze settings before the final test. Using a closing line you could not have known that morning counts as leakage. For bettors who love digging into matchups, it shows which details earn a place in the model and which are just good stories.
Learning objectives
By the end, members should be able to implement and explain the method, reproduce the worked example, identify failure modes, and distinguish a supported research conclusion from an unsupported betting claim.
