Skip to content

v8 Architecture

v9 note: This page documents the preserved v8 language/query architecture. V9 adds representation-independent logical state below/around these read capabilities. See Architecture-v9.

Layer model

                 canonical / authoritative
┌─────────────────────────────────────────────────────┐
│ v6 portable state + v7 semantic registries / Mesh  │
└─────────────────────────────────────────────────────┘
               │                         │
               │                         │
       lexical companions          derived read models
               │                         │
        Language Resolver          KQP Query Engine
               │                         │
               └──────────┬──────────────┘
                          │
                   Context Compiler
                          │
                 human / AI projections

v8 uses a CQRS-like separation: canonical state is optimized for integrity and semantic fidelity; derived read models are optimized for lookup, query and presentation. This does not require asynchronous infrastructure or a particular database.

Anti-corruption boundaries

Adapters isolate v6, v7 and external KOS formats from v8 query internals. External identifiers remain mappings; legacy fields are not silently reinterpreted to simplify implementation.

Federation and sharding

Large semantic or lexical universes may be sharded by subject, jurisdiction, organization or operational locality. Sharding is an operational partition: semantic identity remains stable across shards.

Derived material

Unless an inherited contract says otherwise, these are read-side material:

  • query/search/vector indexes;
  • lexical-resolution caches;
  • semantic slices;
  • AI context bundles;
  • compact symbol tables;
  • query plans;
  • rendered human views.

Derived material must remain traceable to source state/snapshot.

Graceful degradation

  • missing lexicon → preserve semantic ID;
  • unavailable index → slower exact query or explicit unsupported operation;
  • unavailable shard → partial result with missing-shard disclosure;
  • exhausted budget → truncated result with continuation/omission metadata;
  • unavailable tokenizer → bytes/items remain authoritative and token estimate is omitted or marked heuristic.

No degradation path may fabricate semantic content.