How research became a typed world-model

The schema was built in two passes. The first pass read the 23 accepted research briefs from Waves 0–5 — covering world-machine eras, Bryant’s onto-cartography, DeLanda’s assemblage theory, Pearl causal DAGs, West scaling laws, Hidalgo economic complexity, Sassen global cities, Castells network society, Dator future archetypes, and the civilizational atlas stubs — and synthesized them into an independent schema draft. The second pass reconciled that draft against the prescribed specification from the MVP Plan and research/SYNTHESIS.md §2.

The two independent passes converged on ~43 of ~45 mandatory fields — a strong signal that the research-to-schema pipeline is reproducible. Of five reconciliation points, three converged with prior leans and two diverged with defensible rationale.

Wave 6 added four significant extensions to the baseline: a five-value cross-era coupling typology (zombie_dependency, adapted_inheritance, hostile_inheritance, parasitic_extraction, substrate_provision — later extended to 10); a lineage and succession machinery for when a machine’s typology breaks across eras; phase-snapshot trajectories within cards; and a multi-embedding architecture (semantic + state + graph) for Phase 2 retrieval.

Key architectural decisions

Decision	Rationale	Source
is_place / is_path discriminators	Flat ontology: Place and Path are Machine subclasses, not separate top-level types. Reduces schema surface while preserving the distinction.	SYNTHESIS §2.1
gravitational_type derived via PageRank	Never set directly by card authors. Derived from the coupling graph once populated. Prevents fabrication of structural position.	SYNTHESIS §2.8.2
temporal_window allows null end	Ongoing relationships need [year, null]. Requiring a closed sentinel year forces authors to invent dates. Fixed in v0.1 iteration.	spec-iteration-notes S2
Three worked reference cards	JSC lineage (3 cards: Mercantile → Industrial → Platform), Venice 1500 Maritime Trade, Wikipedia. Every field on every card populated honestly before schema commit.	MVP_PLAN §2 Phase 0
Classes-first granularity	Cards are machine classes ("joint-stock company"), not instances ("VOC") — unless singularly load-bearing for an era’s identity (Wikipedia warrants its own card; VOC does not).	MVP_PLAN §2, RESEARCH_AUDIT_BRIEF §6.1
motor null for LM machines	LM machines must have motor: null (Wave 0 Q2 invariant). LM coordination is emergent rather than driven by pull or push.	Wave 0 Q2

Spec iteration surfaces

Four surfaces were caught during the three-card validation round-trip:

• S1: place_subtype=city_state absent from PlaceSubtype enum — worked around via place_subtype=polity + spatial_extent=city_state.
• S2: temporal_window second element must admit null — fixed to tuple[int, Optional[int]].
• S3: SourceCitation.notes inline annotation field missing — added notes: Optional[str].
• S5: YAML 1.1 scientific notation edge case — 1.0e9 parses as string; convention set to always use 1.0e+9 form.