Sentinel

Communication Evaluation

Test how an agent talks — tone, formality, verbosity, and technical level.

Communication evaluation tests whether an agent adapts its communication style to the audience and maintains consistent tone, formality, verbosity, and technical level.

What it tests

  • Does the agent use the right tone for the context?
  • Is the formality level appropriate for the audience?
  • Is verbosity calibrated (not too terse, not too wordy)?
  • Does the technical level match the user's expertise?

Scorer: communication_style

The communication_style scorer evaluates the agent's output against expected communication parameters:

testcase.ScorerConfig{
    Name: "communication_style",
    Config: map[string]any{
        "expected_tone":      "professional",
        "formality_min":      0.6,
        "formality_max":      0.9,
        "verbosity_min":      0.3,
        "verbosity_max":      0.6,
        "technical_level_min": 0.7,
    },
}

Scenario: comms_adaptation

The comms_adaptation scenario generator creates test cases that require the agent to adapt its communication style to different audiences:

Case{
    ScenarioType: testcase.ScenarioCommsAdaptation,
    Input:        "Explain how database indexing works",
    Context: map[string]any{
        "audience":        "junior developer",
        "expected_tone":   "friendly",
        "technical_level": 0.4,
    },
}

Communication parameters

ParameterRangeDescription
TonestringDescriptive label: "professional", "friendly", "casual"
Formality0.0 - 1.0Casual to formal spectrum
Verbosity0.0 - 1.0Terse to verbose spectrum
TechnicalLevel0.0 - 1.0Layperson to expert spectrum

Dimension score

result.DimensionScores["communication"] // 0.0 to 1.0

Use cases

  • Verify a customer support agent uses friendly, low-tech language with end users
  • Test that a technical lead communicates concisely with expert-level detail
  • Ensure an agent adapts formality when switching between internal and external communications

On this page