How We Compare
This page compares AMMM’s design choices with common MMM tooling patterns. It focuses on architecture and workflow contracts rather than exhaustive feature matrices, which can change quickly across ecosystems.
Comparison Axes
Section titled “Comparison Axes”| Axis | AMMM V2 choice | Common alternative patterns |
|---|---|---|
| Workflow model | Stage-gated pipeline with numbered artefact folders | Flat output directories or notebook-only outputs |
| Diagnostics policy | Explicit gate policy (diagnostics_gating) | Diagnostics reported but not gate-enforced |
| Convergence/calibration artefacts | Structured machine-readable JSON + figures in 50_diagnostics/ | Plots/tables without stable machine-readable contract |
| Output schema | Canonical stage folders (00_... to 80_...) | Tool-specific ad hoc output structures |
| Optimisation integration | Native downstream stage (70_optimisation/) tied to fitted model outputs | Separate scripts/workflows with manual joins |
| Reporting integration | Optional interpretation stage (80_interpretation/) | External reporting pipelines only |
Practical Implications
Section titled “Practical Implications”- AMMM is optimised for reproducible, audit-friendly runs where downstream tooling can consume stable file contracts.
- The stage layout and gate fields reduce ambiguity in production workflows.
- The trade-off is a more opinionated pipeline structure compared with fully ad hoc notebook workflows.
Choosing a Tooling Style
Section titled “Choosing a Tooling Style”Choose a stage-gated approach (such as AMMM) when you need:
- repeatable artefact contracts
- explicit diagnostics checks before business decisions
- operational hand-off between modelling and downstream applications
Choose a looser workflow when you need:
- rapid exploratory modelling with minimal process constraints
- highly customised experimental model forms without fixed output contracts
Caveat
Section titled “Caveat”All MMM platforms rely on modelling assumptions. Diagnostic quality supports computational and predictive adequacy, but does not by itself establish causal validity.