Explanation

AMMM V2 is built around a principled, stage-gated Bayesian workflow: each stage writes artefacts to a numbered directory, diagnostics are assessed before downstream consumption, and uncertainty is preserved through interpretation and optimisation decisions.

This section explains why the library is designed this way and how to reason about modelling trade-offs.

• Methodology: Core Bayesian MMM design choices, gates, and interpretation limits.
• Workflow Stages: Stage 0–8 workflow logic, artefact dependencies, and gate policy.
• Baseline Trade-off: Assisted (Prophet) versus exogenous baseline specification and leakage trade-offs.
• Prior Predictive Checking: Why prior predictive checks are mandatory before model interpretation.
• Calibration Diagnostics: PIT, LOO-PIT, coverage, and what calibration diagnoses mean.
• Model Components: Baseline, media, controls, reconstruction identity, and uncertainty propagation.
• LOO-CV & Model Checking: ELPD, p_loo, Pareto k reliability, and model comparison logic.
• Optimisation Concepts: Response-curve optimisation, constraints, and uncertainty handling in decisions.
• Model Comparison: How to evaluate AMMM against alternative MMM tooling.