Architecture

The architecture pages document what each layer owns and how layers hand artifacts to each other. Read these when you want to understand the canonical contracts before authoring recipes, extending the runtime, or auditing manifests.

Looking for option definitions instead? Use the Encyclopedia — that’s where each axis has its own page listing every option (description, when to use, when NOT, references, related options). Architecture pages here are prose narrative explaining the why and how of the layer contracts; Encyclopedia pages are auto-generated lookup for the what of each option. Both are kept in sync against the same LayerImplementationSpec registry.

Canonical layer flow

L0 -> L1 -> L2 -> L3(DAG) -> L4(DAG) -> L5 -> L6 -> L7(DAG) -> L8
        |      |      |       |
       L1.5   L2.5   L3.5    L4.5 diagnostics

L0-L2 and L5-L8 are list-style layers.
L3, L4, and L7 are graph-style layers.
L6 and L7 are default off.
L8 is always the external export boundary.

Diagnostic hooks

L1.5 <- L1
L2.5 <- L1 + L2
L3.5 <- L1 + L2 + L3
L4.5 <- L4 + L3

Diagnostics are default off. With enabled: false, they create no DAG nodes and no sink. With enabled: true, they emit diagnostic artifacts that L8 can include through diagnostics_l1_5, diagnostics_l2_5, diagnostics_l3_5, diagnostics_l4_5, or diagnostics_all.

Foundation and philosophy

Foundation core — macroforecast.core Phase 0 contract that the layered runtime extends.
Philosophy — design intent that the layer contracts encode.
Layer boundary contract — what each layer may consume from upstream and what it must emit downstream.
Recipe layers — YAML key naming, DAG/list shape, diagnostic-layer semantics.
Artifacts and manifest — sink names, manifest fields, on-disk layout.
Reproducibility — bit-exact replicate contract.
Terminology — glossary.

Per-layer pages

Layers