docs: RDF / OWL ↔ OGAR alignment — Morris triad + brutal-upgrade vision

New doc `docs/RDF-OWL-ALIGNMENT.md` (820 lines). The OGAR-side pin for
how OGAR composes with the existing W3C / OMG / Foundry ontology stack
+ the architectural map for the next phase: OGAR as an ontology-aware
ingestion substrate (not "just Neo4j").

Six headline claims (§0 TL;DR):

  1. The Morris triad (syntax / semantics / pragmatics) cuts ACROSS
     the existing L1-L5 ontology stack. OGAR HHTL sits at the
     consumer end (inner = compile-time HHTL; outer = adapter
     pragmatic rebind; boundary = TTL serialization).
  2. OGAR is the runtime application-schema producer for L5 — the
     AR-pattern lift (Rails / Odoo / Django / Ecto / SurrealQL)
     into a unified Class/Attribute/Association/EnumDecl/ActionDef
     IR. vocab/ogar.ttl (168 OWL/RDFS declarations) makes OGAR a
     first-class OWL participant.
  3. THE BRUTAL UPGRADE (§4): OGAR grows an OWL/RDF/TTL DDL adapter
     (ogar-adapter-ttl) + per-element syntax/semantics/pragmatics
     auto-detection + pattern-AST recognition (FMA-Pattern-D,
     FIBO-FND-shape, schema.org-shape, SKR03/04-shape, PROV-O-audit
     -shape) + confidence scoring + actionable-item extraction
     (lifecycle → ActionDef + KausalSpec) + VART radix-trie cache +
     Claude Code session as the validation harness.
  4. The architectural litmus is FMA bones-rendering (§6) — ~75K
     static anatomy classes, ~2.1M relationships, perfect fit for
     compile-time HHTL. Bones → muscles → vasculature → innervation,
     each rebound at the per-deployment Adapter for 3D mesh / locale
     label / rendering material.
  5. Adjacent multi-hop alignment ontologies (§8) — odoo-to-fibo,
     GoBD ↔ PROV-O, FIBO-FND ↔ ogit:smb:Organization — are
     low-effort inputs (cross-vocab semantic work already done).
  6. Statistical confidence calibration (§4.10) — a one-time
     calibration against a 4096-dim Deep-NSM encoder
     (Wierzbicka's semantic primes as the substrate) upgrades the
     confidence surface from hand-coded heuristics to statistically
     grounded scores. Same encoder maps text + OWL into the same
     4096-dim space; calibrated heads emit per-axis confidence (SPO
     / TeKaMoLo / Morris-layer / DOLCE-slot / Action-likelihood).
     Trained once; const-table inference per subsequent import.

Cites and does not re-derive (§1):

  - Woa-rs/.claude/reference/erp_foundry_hhtl_ontology_distillation.md
    (686 lines, L1-L5 architecture)
  - Woa-rs/.claude/reference/four_way_alignment_seam.md (441 lines,
    odoo ↔ OWL/DOLCE/FIBO ↔ OGIT ↔ lance-graph)
  - Woa-rs/docs/LANCE-GRAPH-HYDRATORS-AVAILABLE.md (lance-graph #407,
    11 hydrators wired)
  - lance-graph/.grok/GLUE_LAYER_OGIT_TO_OWL_SPEC.md
  - lance-graph/.claude/knowledge/ogit-owl-dolce-ontology-compartments.md
  - Pillar-14 partial-order axioms (ndarray #188 + OntologySchema::
    is_ancestor #189)
  - bardioc PR #17 (Rubicon Phases 1-5), PR #18 (BindSpace
    dissolution handover), lance-graph PR #470

Direct OGAR ↔ OWL correspondences pinned (§5):

  Class.identity ↔ rdf:about URI
  Class.inheritance ↔ rdfs:subClassOf (Pillar-14 partial-order guard)
  Attribute ↔ owl:DatatypeProperty
  Association::BelongsTo ↔ owl:ObjectProperty + owl:FunctionalProperty
  EnumDecl::Static(items) ↔ owl:oneOf
  EnumDecl::Add { parent_selection } ↔ owl:unionOf
  EnumDecl::Computed ↔ runtime-computed; outside OWL DL
  KausalSpec::StateGuard ↔ SHACL sh:NodeShape (OGAR extension past OWL)
  KausalSpec::LifecycleTrigger ↔ pure OGAR extension
  TeKaMoLo ↔ DOLCE DnS role binding + Fillmore case grammar
  Adapter HHTL leaf-rename ↔ skos:prefLabel@locale rebind
  OntologyDto::project(…, locale) ↔ canonical Morris-pragmatic
     projection in code

Sequencing (§10): phase 1 = this doc; phases 2-9 = ogar-adapter-ttl
scaffold, vart-backend feature, ogar-pattern, ogar-actionable, Claude
Code validation harness, ImportSchemaSource trait, FMA hydrator,
Deep-NSM calibration. Each phase ships as a PR with green CI.

PII abort-guard scan: CLEAN (zero German PII tokens; FHIR resource
names used only as W3C / HL7 public-spec terms).

https://claude.ai/code/session_01PBTGaPCSnnt6u3pjXpbLwY

Author: claude