Explainability Guide

Description

Trace GraphRAG answers back to their sources using the Workbench

Difficulty

Beginner

Duration

15 min

You will need
Goal

Run an explainable GraphRAG query, review the reasoning trace, and trace selected edges back to source documents.

Understand how TrustGraph arrives at its answers

This guide walks through the explainability features using a GraphRAG query as the example. You’ll see how TrustGraph records each stage of the reasoning process and how you can trace facts back to the documents they came from.

New to explainability? Read the Explainability overview first to understand the concepts.

Prerequisites

This guide assumes you’ve already loaded and processed a document using Graph RAG. If you haven’t done that yet, follow the Graph RAG guide first.

What You’ll Learn

  1. Running a GraphRAG query with explainability enabled
  2. Reviewing the reasoning trace (question, grounding, exploration, focus, synthesis)
  3. Understanding why specific edges were selected
  4. Tracing edges back to source documents
  5. Browsing past explainability sessions

Step 1: Run an Explainable Query

In the Workbench, navigate to the Graph RAG query page.

Enter a question and enable the Explainability toggle before submitting the query.

The answer will stream as normal, but behind the scenes TrustGraph records the full reasoning trace.

Step 2: View the Reasoning Trace

After the query completes, navigate to the Explainability page in the Workbench.

You’ll see a list of all recorded sessions. Find your query — it will show the question text, type (GraphRAG), and timestamp.

Click on a session to view its full trace.

Step 3: Review the GraphRAG Trace

The trace view shows each stage of the reasoning pipeline:

Question

The original query and when it was submitted.

Grounding

The concepts extracted from your question that were used to seed the graph search. These are the terms TrustGraph used to find entry points into the knowledge graph.

Exploration

The graph traversal results: which entities were found as starting points, and how many edges were explored.

Focus

This is the most important stage. It shows which edges were selected from the explored subgraph, and why each edge was chosen.

The LLM provides reasoning for each selection, explaining why it considered that fact relevant to answering your question.

Synthesis

The final generated answer, along with a reference to the document context that was assembled for the LLM.

Step 4: Trace Back to Source Documents

From the Focus stage, you can trace any selected edge back to its source document. This uses the extraction provenance stored in the urn:graph:source named graph.

Select an edge and choose Show provenance. TrustGraph will show the chain:

Selected edge → Subgraph → Chunk → Page → Source document

This tells you exactly which text in which document produced the fact that TrustGraph used to answer your question.

Step 5: Browse Past Sessions

All explainability traces are persistent. You can return to the Explainability page at any time to review past queries — useful for auditing, debugging, or understanding how the system’s answers relate to the underlying knowledge.

Sessions are listed with their type (GraphRAG, DocRAG, or Agent), the question text, and timestamp.

Summary

In this guide you:

  • Ran a GraphRAG query with explainability enabled
  • Reviewed the 5-stage reasoning trace
  • Examined why specific edges were selected
  • Traced a fact back to its source document
  • Browsed past explainability sessions

Next Steps