Skip to content

Extra Credit – DFD Fidelity, Canonical Models & RAG Implications

Focus: Understanding and mitigating the fidelity issues that arise when converting visual Data Flow Diagrams into structured representations for RAG-based security and SDLC evaluation.

Learning Outcomes

By completing this extra-credit module you will be able to:

  • Distinguish visual, structural, semantic, and provenance fidelity—and know which losses actually matter for compliance evaluation.
  • Explain how incompleteness in a structured DFD propagates into retrieval, graph path queries, skills, and final evaluation metrics.
  • Design a canonical DFD JSON schema and a validation skill that surface missing security-relevant information rather than silently ignoring it.
  • Measure the impact of controlled fidelity loss on the answers produced by the existing hybrid + graph + skill pipeline.

Agenda

  1. Theory – Taxonomy of fidelity loss and its implications for RAG systems
  2. Lab EC.1 – Canonical DFD schema, validation skill, and controlled conversion patterns
  3. Lab EC.2 – Measuring fidelity impact on retrieval, path queries, and compliance scores

Prerequisites

  • Completed Weeks 1–3 (chunking, hybrid/graph storage, modular skills)
  • Familiarity with the sample DFD JSON and graph schema from Lab 2.2
  • Optional but recommended: Week 4 evaluation concepts