Module semantic_disk 

Semantic disk — taxonomy-embedded meaning space (E+D design).

Design: docs/SEMANTIC_DISK.md (ratified 2026-07-11). The domain’s concept taxonomy — derived from structure within the data (a category tree, a directory tree) — is Sarkar-embedded into its own Poincaré disk by inserting the concept paths into a private Store. Every data node’s position in that disk is a pure function of its affinity dimensions: the weighted Klein barycenter (Einstein midpoint) of the concept anchors. Nothing is stored; the position can never disagree with the dims; and because the mapping is deterministic, an epoch record of the dims replays the node’s path through meaning-space bit-identically.

Three query families fall out:

• SemanticDisk::concept_of — which concept does this node belong to right now (hyperbolic Voronoi cell location among the anchors; constant in data-node count). Compared with the node’s storage path, this is miscategorization detection as a primitive.
• SemanticDisk::nearest — k nearest data nodes in meaning-space (hyperbolic distance between derived positions; the semantic index’s metric tree with crate::metric_tree::HyperbolicMetric, epoch-cached).
• SemanticDisk::classify_trajectory — a temporal trajectory readout pushed through the anchor cells: a moving point becomes a sequence of discrete meaning-states across epochs.

The disk is a standalone object the application owns (anchors are few — dozens, not thousands — so building one is cheap). Store gains no new state. The (concept path ↔ affinity dim) mapping is calibration: fixed for a dataset’s life, like the dimensional schema itself.

Structs

SemanticDisk

The taxonomy-embedded meaning space (see module docs).