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

> ## Agent Instructions
> These docs moved from docs.flintai.dev to docs.switchagents.ai. Use docs.switchagents.ai for every link and request.
> To search these docs from an AI tool, connect the MCP server at https://docs.switchagents.ai/mcp. The page index is at https://docs.switchagents.ai/llms.txt.

# How model evaluation works

> What gets tested on a model, how it's scored, and how a score becomes a finding

**Evaluation attacks a model on purpose and scores how well it holds up.** This page explains what runs, what the number means, and why a weak score also shows up as an issue.

## What gets evaluated

Model evaluation tests the **model**, on its own behavior, independently of the agents that call it. That's why one weak model can put several agents at risk at once. Model evaluation isn't something you run. A model either arrives with scores attached or it doesn't.

<Note>
  **Want to evaluate an agent, not a model?** Agent evaluation attacks your running agent end to end, with its instructions, tools, and guardrails in the loop, and you run it yourself. Refer to [Evaluate your agents](/switch-trust/evaluation/evaluate-agents). You can run the same kind of test from your terminal with [Switch Trust Eval](/switch-trust-cli/eval/getting-started) in the CLI.
</Note>

## How a model is tested

<Steps>
  <Step title="Probes send hostile prompts">
    Each probe is a named attack technique — a jailbreak pattern, a toxicity elicitation, a prompt injection — with a fixed idea of what it is trying to make the model do.
  </Step>

  <Step title="The model responds">
    The probe runs against the model directly, without your agent's instructions or guardrails in the way. The result describes the model itself, not your configuration of it.
  </Step>

  <Step title="Each probe is scored">
    Each response is scored by a detector, and the probe's score is the aggregate across all its prompts, expressed from 0 to 100. Higher is better.
  </Step>

  <Step title="Scores roll up">
    Probe scores aggregate into category scores, and those into the overall **Health score**.
  </Step>
</Steps>

## What each category tests

Every probe belongs to one category, and every category score runs the same direction: a higher score means the model performed better.

| Category | What it tests |
| - | - |
| **Jailbreaks** | Breaking out of system instructions |
| **Misuse** | Being turned to purposes it wasn't built for |
| **Toxicity** | Producing harmful, offensive, or misleading output |
| **Security** | Producing output that's dangerous to whatever consumes it |
| **Robustness** | Producing unreliable or fabricated output |

## How a score becomes a finding

Evaluation produces a number. Rules turn that number into a finding you can triage.

A set of model evaluation rules watches these scores and fires when one falls too low. Unlike most rules, which declare a single fixed severity, these **band by score** — the same rule produces a different severity depending on how far the model fell. **Model toxicity risk**, for example:

| Toxicity score | Severity |
| - | - |
| Below 0.2 | Critical |
| 0.2 to 0.35 | High |
| 0.35 to 0.5 | Medium |
| 0.5 to 0.75 | Low |

At 0.75 and above the rule doesn't fire.

This is why one rule can occupy more than one row in the **Issues** table. A banded rule rates each model separately, so a badly failing model and a merely mediocre one land on different rows — and the bad one stays visible instead of being averaged in with the rest.

<Note>
  **Two scales, one measurement.** Rule thresholds are written on a 0 to 1 scale while the health score is shown from 0 to 100. A toxicity score of 0.2 is a displayed score of 20. Both run the same direction: higher is better.
</Note>

For how severity behaves once a finding exists, and how it differs from CVSS, refer to [How discovery works](/switch-trust/discovery/how-discovery-works).

## Next steps

<CardGroup cols={2}>
  <Card title="Evaluation results" icon="gauge-high" href="/switch-trust/evaluation/results">
    Read the model health score and decide whether it's trustworthy
  </Card>

  <Card title="Evaluate your agents" icon="shield-halved" href="/switch-trust/evaluation/evaluate-agents">
    Attack your running agent end to end and score how it holds up
  </Card>
</CardGroup>