GraphRAG vs Vector RAG Topology Simulator & Community Retrieval Workbench

Auth / Security (C1)

Core Payment (C2)

Data / Storage (C3)

Active Traversal

Global Synthesis Query Louvain Modular Clustering

Graph Critical Path

42 ms

Error Correlation

r = 0.84

Monotonicity λ

0.96

Retrieval Method Comparison Live Evaluation

Vector RAG (Top-k Cosine) Coverage: 40%

Retrieves isolated point-wise entities with high embedding similarity: Auth Agent, Payment API. Misses cross-community cascade dependencies (Audit Logger, Fraud Guard).

e1: Payment API (0.89) e2: Auth Agent (0.85)

GraphRAG (Hierarchical Synthesis) Coverage: 100%

Synthesizes C1 (Auth/Security) & C2 (Core Payment) community summaries. Identifies systemic cascade risk: Auth Agent queries User DB, verified by Fraud Guard, audited by Logger.

C1: Auth & Fraud C2: Payment Core C3: Persistence

Entity / Topology Inspector e1 (Payment API)

Entity Label

Entity Type

Community Cluster

Error Weight (0-1)

Topology Verification Proof

Total Entities: 5

Total Relations: 5

Graph Coverage Ratio: 100%

Vector Synthesis Loss: 60%

Last Action State: RUNNABLE_DEFAULT_LOADED