Layered Architecture L0–L6

ELI5

STRATT is built like a kitchen brigade: each station only talks to the one above and below it. The cleaver (identity) prepares ingredients, the line cook (schema) plates them, the expediter (graph) sequences orders — nobody reaches across the pass.

Technical Deep Dive

Layer Map

flowchart TB
    L6["L6 MERIDIAN<br/>apps/meridian — Astro/Starlight site"]
    L5["L5 CLI<br/>@stratt/cli — 9 Phase 1 commands"]
    L4["L4 Council Registry<br/>councils/ content sets"]
    L35["L3.5 n8n Exporter<br/>@stratt/n8n-exporter — chain compiler"]
    L3["L3 Graph<br/>@stratt/graph — DAG, blast radius, CI"]
    L2["L2 CRDT<br/>@stratt/crdt — Yjs documents"]
    L1["L1 Schema<br/>@stratt/schema — Zod validators, URI parser"]
    L0["L0 Identity<br/>@stratt/fingerprint — Blake3 + canonicalisation"]
    L6 --> L5 --> L4 --> L35 --> L3 --> L2 --> L1 --> L0

Layer Contracts

L4 is content-only (councils/ directory), not a code package. Every upward call must pass through one layer at a time — @stratt/cli does not import @stratt/fingerprint directly; it goes via schema.

Key Terms

• Layer → a package boundary with a single responsibility and a fixed downstream dependency.
• Council → a content set of related units, registered at L4.
• MERIDIAN → the L6 documentation app that renders units from R2.

Q&A

Q: Why is @stratt/n8n-exporter numbered L3.5 instead of L4? A: It depends on the L3 graph (chains are DAG nodes) but is optional and orthogonal to council content, so it sits between layers rather than above them.

Q: Can L1 schema code call L0 fingerprint code? A: Yes — adjacent layers may call downward. The rule forbids skipping or calling upward.

Examples

A stratt validate invocation: CLI (L5) → schema validators (L1) → URI parser (L1) → fingerprint verify (L0). The CRDT layer (L2) is bypassed because validation reads canonical YAML, not Yjs state.