---
title: Cognitive Evaluation
description: Test how an agent thinks — phase transitions, depth, and reasoning strategy.
---

**Cognitive evaluation** tests whether an agent follows appropriate thinking strategies, transitions between cognitive phases correctly, and maintains the right depth and focus.

## What it tests

- Does the agent transition between thinking phases appropriately?
- Does depth of analysis match expectations (shallow vs. deep)?
- Does focus remain appropriate (broad vs. narrow)?
- Does the agent self-reflect when needed?

## Scorer: `cognitive_phase`

The `cognitive_phase` scorer evaluates the agent's reasoning strategy transitions by analyzing the run trace:

```go
testcase.ScorerConfig{
    Name: "cognitive_phase",
    Config: map[string]any{
        "expected_phases": []string{"analytical", "reflective", "methodical"},
        "depth_min":       0.7,
        "focus_min":       0.5,
    },
}
```

## Scenario: `cognitive_stress`

The `cognitive_stress` scenario generator creates test cases that require the agent to shift thinking strategies:

```go
Case{
    ScenarioType: testcase.ScenarioCognitiveStress,
    Input:        "This function has both a performance bug and a security issue. Fix the critical one first, then address the other.",
    Context: map[string]any{
        "expected_phases": []string{"analytical", "methodical"},
        "depth_required":  "deep",
    },
}
```

## Cognitive strategies

| Strategy | Description |
|----------|-------------|
| `analytical` | Structured, step-by-step analysis |
| `creative` | Exploratory, lateral thinking |
| `methodical` | Systematic, exhaustive approach |
| `reactive` | Quick response, minimal deliberation |
| `reflective` | Self-evaluating, iterative refinement |
| `collaborative` | Seeks input, considers multiple perspectives |

## Dimension score

```go
result.DimensionScores["cognition"] // 0.0 to 1.0
```

## Use cases

- Verify a code reviewer starts with analysis, then reflects on findings, then methodically addresses issues
- Test that an agent switches from broad exploration to focused investigation when it finds a lead
- Ensure deep analysis for complex tasks, quick responses for simple ones