What this is

A new ogar-fma-skeleton crate: the hand-curated FMA skeleton as the
clamped convergence anchor for the splat-native ultrasound arc
(docs/SPLAT-NATIVE-CUSTOMER.md §6 litmus). Operator-directed across the
session; the address model converged through several corrections recorded
below.

The constraint (one, not two): bones must be hardcoded and stable
because they are the cross-modal registration frame that ViT, X-ray, and
ultrasound × Doppler all project onto. Bones are the only rigid bodies in the
human, so the skeleton is not data being fit — it is the boundary
condition of the fit. Clamping the splat-fit to a hand-curated skeleton
makes the convergence well-posed and deterministic (the Class IIa SaMD
requirement).

The address — the [256:256] [container:member] GUID

Each bone resolves to a 16-byte Guid: a uniform stack of [256:256] tiers,
each the same scale-free relation [container : member] (Galaxy:planet,
country:city, bodypart:bone, cm²:mm², residue:atom). Bones are
Located-primary (they ARE the position anchor):

classid  = 0x0A03                  Anatomy : bone
HEEL/HIP = rest-pose 3-axis CRS    HEEL coronal x:y · HIP depth z   (ArcGIS / Cesium-addressable)
LEAF     = familyNode : identity   [bodypart : bone]

• HEEL/HIP make the body a first-class GIS layer (Morton/quadkey), so
  off-the-shelf spatial ops work and ViT / ultrasound × Doppler project onto
  the same coordinate. Two HHTL modes are supported (HhtlMode):
  Located (Cesium, position-preserving — the bone case) and Cascade
  (ontology, the self-speaking path — the soft-tissue case).
• LEAF family node IS the group: identity == 0 is the family node,
  identity ≥ 1 are its bones ("identity attached to family node").
• Each byte is a directly-addressable 256-codebook and a 4×4 Morton tile
  carrying the perturbation pyramid's (exponent, location) (OGAR owns those;
  ndarray owns phase, magnitude).

Consumes D-MORTON (DISCOVERY-MAP §2.1); recorded as D-FMA-SKELETON.

Codebook

New ConceptDomain::Anatomy (0x0A) in ogar-vocab: anatomical_structure
/ skeleton / bone / joint (0x0A01..04), in lockstep across CODEBOOK +
class_ids consts + ALL + builders. Anatomy is public reference, not
PHI — kept out of Health 0x09 so it is not dragged into medcare's
fail-closed Health RBAC set (preserves its Health == 7 invariant). The ~206
bones are cascade-path nodes under the bone concept, not codebook slots.

Relations

No 12+4 edge block — that is the pre-family-node taxonomy and was removed.
With family nodes, relations ARE the addressing: local = shared family
prefix (same_family), cross = a reference to another node's Guid.

Atlas coverage & honesty

~80 nodes: full vertebral column C1–coccyx, skull, sternum, 12 rib pairs,
major appendicular bones per side — structured to extend to the full ~206
without schema change. Rest poses are a proportional T-pose v0 (to be
refined against a real FMA-aligned mesh); FMA ids are cross-referenced only
where confidently known (the address identity is a stable atlas id, never a
fabricated FMA id). HIP is coronal-tile + depth-slice (not yet a single 3D
octree — its odd bits are reserved). The address structure is CODED;
the splat-fit convergence claim is CONJECTURE, gated on the splat-native
arc.

Tests / gates

cargo test -p ogar-fma-skeleton — 24 + 2 doctests green: Morton math,
classid tier, HEEL laterality, HIP depth, unique node keys + leaves,
family-node grouping with attached identity, parent-link partonomy, the
clamped-anchor invariant, unit-quaternion frames, classid stability.
ogar-vocab 94 tests green (drift gates updated for the anatomy mints).
The new crate is cargo clippy-clean; the 12 pre-existing ogar-vocab
clippy warnings are untouched (zero added).

Docs

docs/FMA-SKELETON-CONVERGENCE-ANCHOR.md (the unity, the GUID tier model,
located/cascade, the projection operators) + D-FMA-SKELETON in the
DISCOVERY-MAP.

🤖 Generated with Claude Code

Generated by Claude Code