> ## Documentation Index
> Fetch the complete documentation index at: https://docs.enfors.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Evidence & the Logic Trace

> How Enfors reconstructs why an agent decided what it did — backed by evidence.

When a decision is challenged — because an agent drifted, or the spend looked too high —
you need to reconstruct and explain it. Enfors does this with two linked structures: the
**logic trace** and the **evidence vault**, both modeled as a graph.

## The logic trace

Open any decision and you can walk its **logic trace**: the outcome and the action the
agent decided to execute, plus the **reasoning step by step** that got it there.

<Frame caption="The logic trace — outcome, action, and step-by-step reasoning">
  <img src="https://mintcdn.com/enfors/EStKoMAmv9k4MqOd/images/logic-trace.svg?fit=max&auto=format&n=EStKoMAmv9k4MqOd&q=85&s=54ca7e8f07b5dc747a9de5c6801aca16" alt="Decision logic trace" width="1200" height="750" data-path="images/logic-trace.svg" />
</Frame>

## The evidence vault

Every step in the trace is **backed by evidence**. The evidence vault holds each piece —
a system prompt, a git check-in, a prompt, a document, or a tool call — so a decision can
be reconstructed from its logic *and* the evidence that supported it.

<Frame caption="The evidence vault for a decision">
  <img src="https://mintcdn.com/enfors/EStKoMAmv9k4MqOd/images/evidence-vault.svg?fit=max&auto=format&n=EStKoMAmv9k4MqOd&q=85&s=a4306b35f91cab1fd94cfeb290cc8784" alt="Evidence vault" width="1200" height="750" data-path="images/evidence-vault.svg" />
</Frame>

## The graph model

Internally, each decision's structure is written to a graph database so it can be traversed
and explored. Every node is scoped to its tenant, workspace, and user.

```mermaid theme={null}
flowchart LR
  Agent -->|MADE| Decision
  Decision -->|ABOUT| Subject
  Decision -->|USED| Evidence
  Evidence -->|DEPENDS_ON| Step
  Step -->|STEP_NEXT| Step
  Step -->|STEP_TO_FIND| Finding
  Finding -->|SUPPORTS| Decision
  Finding -->|REFUTES| Decision
  Step -->|MADE_ARTIFACT| Artifact
```

| Node         | What it represents                                       |
| ------------ | -------------------------------------------------------- |
| **Agent**    | The AI worker that made the decision.                    |
| **Decision** | The decision itself, with outcome and confidence.        |
| **Subject**  | What the decision was about.                             |
| **Evidence** | A supporting artifact (prompt, doc, tool call, commit…). |
| **Step**     | A reasoning step in the logic trace.                     |
| **Finding**  | A conclusion that supports or refutes the decision.      |
| **Artifact** | Something the agent produced.                            |

This is the structure you navigate on the [Explore canvas](/using/builder-lens), and
that you can query directly with the
[`query_graph`](/mcp/tool-reference#query_graph) MCP tool.

<Tip>
  The graph is also what makes the **speculation ratio** possible: a step with no backing
  evidence stands out structurally. See [Insights & Drift Detection](/concepts/insights-and-drift).
</Tip>