Skip to content

v0.2.0 — Phase 2: Knowledge Graph Engine

Latest
Latest

Choose a tag to compare

View all tags
@Soheil-jafari Soheil-jafari released this 18 May 18:41
v0.2.0
ed510b4

What's new

Phase 2 of Argus ships the knowledge graph engine — a pluggable, typed, production-grade graph layer with a NetworkX in-process backend and a Neo4j production backend, parity-tested byte-identical across both. The supply-chain domain pack now maps its four core entities (Order, Supplier, Shipment, EventSignal) into the graph and exposes flagship cascading-risk and subgraph-extraction queries.

Highlights

  • KGBackend Protocol with documented merge semantics (per-key property merge, source_refs union, type-mismatch raises)
  • NetworkXBackend for in-process use and unit tests
  • Neo4jBackend for production, with APOC requirement and clear startup error if missing
  • 7 NodeTypes (Supplier, Product, Region, Customer, Order, Shipment, EventSignal) and 10 EdgeTypes with stable-ID helpers
  • Flagship queries: cascading_risk() and subgraph() — parity-tested across both backends with hand-traced fixture scenarios
  • SupplyChainKGAdapter with four-rule entity resolution for EventSignal mentions plus an unresolved_mentions counter surfaced through IngestionReport.adapter_counters
  • docs/kg.md covering schema, ingestion, queries (Cypher vs NetworkX side-by-side), backend selection, APOC requirement
  • Coverage gate flipped from report-only to enforced at fail_under=83

Quality gates

  • 396 tests passing locally (45 new from flagship + end-to-end suite)
  • mypy strict clean across 64 source files
  • ruff + bandit clean
  • Integration tests green on real Neo4j 5.20-community + APOC via testcontainers in CI

Roadmap

Phases 3-5 (predictive head with evidential uncertainty, RAG with grounding rubric, human-in-the-loop dashboard) are documented in docs/PROJECT_CONTEXT.md as the next iteration. Phase 6 (multi-cloud Terraform) follows.

Assets 2
Loading