Iterative MMM Improvement
Architecture
The improvement loop has two phases:
Phase 1: Tournament
Run N variants of the model simultaneously (different hyperparameter configurations), score each, select the winner.
Phase 2: Posterior-Informed Refinement
Use the winner's posterior distribution to tighten priors, then re-run the tournament from the better starting point. Repeat until score plateaus.
Variant Generation
Each tournament round generates variants by grid-searching:
[Description truncada. Veja o README completo no GitHub.]