Skip to content

ACTUAL_SCHEMA.md

Latest commit

 

History

History
123 lines (101 loc) · 4.12 KB

ACTUAL_SCHEMA.md

File metadata and controls

123 lines (101 loc) · 4.12 KB

Schema Reference

SQL Tables (Graph_KG)

nodes

CREATE TABLE Graph_KG.nodes (
    node_id VARCHAR(256) PRIMARY KEY,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
)

rdf_labels

CREATE TABLE Graph_KG.rdf_labels (
    s VARCHAR(256) NOT NULL,
    label VARCHAR(128) NOT NULL,
    CONSTRAINT pk_labels PRIMARY KEY (s, label)
)

rdf_props

CREATE TABLE Graph_KG.rdf_props (
    s VARCHAR(256) NOT NULL,
    "key" VARCHAR(128) NOT NULL,
    val VARCHAR(64000),
    CONSTRAINT pk_props PRIMARY KEY (s, "key")
)

rdf_edges

CREATE TABLE Graph_KG.rdf_edges (
    edge_id BIGINT IDENTITY PRIMARY KEY,
    s VARCHAR(256) NOT NULL,
    p VARCHAR(128) NOT NULL,
    o_id VARCHAR(256) NOT NULL,
    qualifiers VARCHAR(4000)
)

kg_NodeEmbeddings

CREATE TABLE Graph_KG.kg_NodeEmbeddings (
    id VARCHAR(256) PRIMARY KEY,
    emb VECTOR(DOUBLE, 768),
    metadata %Library.DynamicObject,
    CONSTRAINT fk_emb_node FOREIGN KEY (id) REFERENCES Graph_KG.nodes(node_id)
)
CREATE INDEX kg_emb_hnsw ON Graph_KG.kg_NodeEmbeddings(emb)
    AS HNSW(M=16, efConstruction=200, Distance='COSINE')

fhir_bridges

CREATE TABLE Graph_KG.fhir_bridges (
    fhir_code VARCHAR(64) %EXACT NOT NULL,
    kg_node_id VARCHAR(256) %EXACT NOT NULL,
    fhir_code_system VARCHAR(128) NOT NULL DEFAULT 'ICD10CM',
    bridge_type VARCHAR(64) NOT NULL DEFAULT 'icd10_to_mesh',
    confidence FLOAT DEFAULT 1.0,
    source_cui VARCHAR(16),
    CONSTRAINT pk_bridge PRIMARY KEY (fhir_code, kg_node_id)
)

Note: %EXACT preserves case (IRIS VARCHAR uppercases by default).

SQL Stored Procedures (Graph_KG)

Procedure Signature Description
kg_KNN_VEC (query_vector, k, label, property) HNSW vector search, returns JSON [{"id","score"},...]
kg_PPR (seeds_json, damping, max_iter, ...) Personalized PageRank, returns JSON [{"id","score"},...]
kg_NEIGHBORS (sources, predicate, direction) 1-hop neighborhood, returns JSON array
kg_BM25 (name, query, k) BM25 lexical search, returns JSON [{"id","score"},...]
kg_RRF_FUSE (k, k1, k2, c, vector, text) Reciprocal rank fusion (vector + text)

kg_BM25 is a thin wrapper over Graph.KG.BM25Index.Search() exposed as an SQL stored procedure for Cypher translator Stage CTEs.

Global Structures

^KG — Temporal + Structural Graph

^KG("out", 0, s, p, o) = weight              — structural outbound edges (shard 0)
^KG("in",  0, o, p, s) = weight              — structural inbound edges (shard 0)
^KG("tout", ts, s, p, o) = weight            — temporal outbound (time-ordered)
^KG("tin",  ts, o, p, s) = weight            — temporal inbound
^KG("bucket", bucket, s) = count             — 5-min pre-aggregated edge count
^KG("tagg", bucket, s, p, key) = value       — COUNT/SUM/AVG/MIN/MAX/HLL per bucket
^KG("edgeprop", ts, s, p, o, key) = value    — rich edge attributes
^KG("deg", s) = total_degree                 — pre-aggregated total degree
^KG("degp", s, p) = degree_for_predicate    — pre-aggregated per-predicate degree (for KHopCount O(1))

Note: The "out" and "in" subscripts use shard 0 (i.e., ^KG("out", 0, s, p, o)). The shard subscript reserves space for future horizontal partitioning. ^KG("degp", s, p) powers the KHopCount fast path — O(1) lookup instead of $Order scan.

^BM25Idx — BM25 Lexical Search

^BM25Idx(name, "cfg", "N")           — document count
^BM25Idx(name, "cfg", "avgdl")       — average document length
^BM25Idx(name, "cfg", "k1")          — BM25 k1 parameter
^BM25Idx(name, "cfg", "b")           — BM25 b parameter
^BM25Idx(name, "cfg", "vocab_size")  — distinct token count
^BM25Idx(name, "idf",  term)         — Robertson IDF value
^BM25Idx(name, "tf",   term, docId)  — term frequency (term-first subscript order)
^BM25Idx(name, "len",  docId)        — document token count

Term-first "tf" subscript enables O(postings) iteration: $Order(^BM25Idx(name,"tf",queryTerm,"")).

^VecIdx, ^PLAID, ^NKG

See ARCHITECTURE.md for full global documentation.