Sample · semantic proving ground

Supply Chain Semantic

A deliberately demanding example where application meaning is richer than CRUD and the caller is never treated as trusted.

What it demonstrates

Recursive traversalBounded multi-hop relationship paths.AuthorizationEntity, field, relationship and conditional policies.MutationsSemantic mutation graphs and generated-value dependencies.GroundingAmbiguous language becomes an explicit resolution decision.RetrievalProvider-backed candidate discovery with evidence.Untrusted claimsClient claims can narrow access but never widen it.

Authorization laboratory

The sample is intentionally designed so security rules are visible in the semantic model: sensitive fields, restricted relationships, tenant predicates, role-based writes and stronger named-operation requirements can all be tested independently.

For the complete matrix, read Semantic authorization cases.

Grounding and retrieval

PostgreSQL can supply fuzzy, full-text, optional BM25 and optional Apache AGE graph candidates. Those strategies provide evidence only. The semantic resolver decides the meaning, and authorization remains a separate step.

For the underlying implementation, see Lexical grounding and Grounding decisions.

Read the source walkthrough

Start at 00 — Overview and Setup, then follow 01–05 for claims/authorization, high-assurance scenarios, ambiguity, retrieval and adversarial testing.