Phase: P5.5 | Risk de-risked: R-03, R-09
Context
Inserted between P5 (real LLM) and P6 (entity wiring) to validate the Evidence→Normalize→Knowledge pipeline before wiring 24 entities. If this fails, fix architecture before scaling.
Scope
Validate Medallion-inspired flow:
Connector → Raw Evidence (Bronze) → Canonical Evidence (Silver) → Knowledge Graph (Gold) → Decision Context → Workspace Projection
Tests for: duplicate evidence, out-of-order events, stale updates, conflicting facts, missing foreign keys, partial connector failures, replay after crash, idempotent reprocessing, tenant isolation.
Key requirement
Unify Silver: standardize on knowledge/semantic_layer.py + connector_bridge.py (5-tier data-driven) and retire knowledge/entity_resolver.py hardcoded extraction (Sarah Chen / Mike Liu). Add the missing conflict-detection stage.
Acceptance
- Bronze/Silver/Gold all persist to real stores (no in-memory fallback).
- Duplicate/dedup, out-of-order, stale, conflict, missing-FK, partial-failure, replay, idempotency, tenant isolation all tested.
- Conflict detection is a real stage, not a stub.
Verification
Phase: P5.5 | Risk de-risked: R-03, R-09
Context
Inserted between P5 (real LLM) and P6 (entity wiring) to validate the Evidence→Normalize→Knowledge pipeline before wiring 24 entities. If this fails, fix architecture before scaling.
Scope
Validate Medallion-inspired flow:
Tests for: duplicate evidence, out-of-order events, stale updates, conflicting facts, missing foreign keys, partial connector failures, replay after crash, idempotent reprocessing, tenant isolation.
Key requirement
Unify Silver: standardize on
knowledge/semantic_layer.py+connector_bridge.py(5-tier data-driven) and retireknowledge/entity_resolver.pyhardcoded extraction (Sarah Chen / Mike Liu). Add the missing conflict-detection stage.Acceptance
Verification
uv run pytest tests/v6/