Case Study: Knowledge Graph Validation Workflows for Pre/Post-LLM

This case study is the mdBook version of the website post:

• https://terraphim.ai/posts/kg-validation-workflows-case-study/

It includes the same narrative plus additional technical appendices.

Why This Existed

Terraphim already had powerful primitives (connectivity checks, fuzzy matching, role-aware graphs). The missing piece was an opinionated workflow and a stable interface (CLI + hooks) that made those primitives usable.

The Approach

• Pre-LLM: validate semantic coherence before spending tokens
• Post-LLM: validate outputs against domain checklists
• Developer UX: unify everything behind the terraphim-agent CLI and Claude Code hooks

The Results

• MCP connectivity tool wired to real RoleGraph logic
• New CLI commands (validate, suggest, unified hook handler)
• Skills + hook scripts that standardize the workflow

Appendix A: Workflow Overview (Mermaid)

flowchart TD
    A[Developer input] --> B[Pre-LLM validate]
    B -->|connected| C[Call LLM]
    B -->|not connected| D[Fuzzy suggest + rewrite]

    C --> E[Post-LLM checklist validation]
    E --> F[Warnings + missing terms]
    E --> G[Accept output]

    H[Claude Code hooks] --> B
    H --> E
    I[CLI: terraphim-agent] --> B
    I --> D
    I --> E

Appendix B: Primary Sources

• docs/sessions/research-underutilized-features.md
• docs/sessions/design-underutilized-features.md
• docs/sessions/implementation-summary.md
• docs/sessions/session-20251228-201509.md