What this is

CanonicAI's interface is data, not a REST/MCP gateway. Consumers integrate by reading canonical outputs and the federation-spine contracts — this page is the equivalent of an "API reference" for a producer engine.

Canonical output contract (per book)

canonical_outputs/<book>/
  chapters/<chXX>/
    chunks.json         # segmented source chunks
    summary_core.json   # structured core summary (the machine contract)
    gaps.json           # detected coverage gaps
    summary.md          # human-readable summary
  book_level/
    book_level_sweep.json   # THE clean, current book-level store (read this)
    _sweep_ledger.csv       # coverage ledger

Read book_level_sweep.json, not the dead summary.json. Variable-relationship models in the sweep carry provenance and are the high-value extraction (T2 candidates downstream).

Federation spine contracts

Contract	Package	Consumer
Canonical measurement vocabulary	measurement-core (@people-analyst/measurement-core)	Principia + toolbox
Record publication	measurement-ingest	Principia canonical registry
Work identity (deriveLibraryId)	library-core	PA-site, vela
Variable normalization	variableizer	measurement-core → Principia
Reliability coefficients	reliability	Principia

Who consumes what

• Principia — instruments, constructs, citations, effect data, measurement records (the evidence spine).
• peopleanalyst.com — corpus-derived content.
• toolbox — measurement vocabulary + library identity. Consumers vendor the typed contracts; the extraction logic stays in CanonicAI.

Generated interface reference

docs/capability-manifest.json is program-oriented (discoverability, patent radar) rather than a service catalog, so the generated reference (reference-mcp.generated.md) is intentionally sparse — CanonicAI's surface is this data contract, not REST/MCP. Regenerate with node ~/devplane/scripts/gen-product-reference.mjs ..

See also

Architecture · Concepts · docs/PORTFOLIO-INTEGRATION.md