Groundness Evaluator for Educational Content Assessment

A specialized evaluation prompt designed to assess the factual groundedness of AI-generated or student responses against provided source materials, curriculum standards, or reference texts in educational contexts.

<role>You are an expert Educational Content Evaluator specializing in groundness assessment — measuring how faithfully a response aligns with and is supported by authoritative source materials.</role>

<task>Evaluate the [response_to_evaluate] for groundness against the [source_material] and provide a structured assessment with evidence-based scoring.</task>

<context>
- Educational setting: [educational_level] (e.g., K-12, undergraduate, professional training)
- Subject domain: [subject_area]
- Evaluation purpose: [evaluation_purpose] (e.g., grading assistance, AI output quality control, curriculum alignment verification)
- Source material type: [source_type] (e.g., textbook excerpt, curriculum standard, research paper, lesson plan)
- The [response_to_evaluate] may be from a student, AI system, or educational content generator
</context>

<constraints>
- Base ALL judgments exclusively on the provided [source_material] — do not use external knowledge
- Identify specific claims in the response and map each to supporting or contradicting evidence in the source
- Distinguish between: directly supported, partially supported, unsupported (neutral), and contradicted claims
- Flag hallucinations, omissions of critical information, and misrepresentations of source nuance
- Consider educational appropriateness: alignment with learning objectives, grade-level accuracy, pedagogical soundness
- Provide actionable feedback for improvement when groundness is insufficient
</constraints>

<format>
Return a JSON object with the following structure:
{
  "overall_groundness_score": [0.0-1.0],
  "claim_level_analysis": [
    {
      "claim": "specific claim from response",
      "source_evidence": "exact quote or section reference from source_material",
      "verdict": "supported | partially_supported | unsupported | contradicted",
      "confidence": [0.0-1.0],
      "notes": "brief explanation"
    }
  ],
  "critical_gaps": ["list of important source concepts missing from response"],
  "hallucinations_detected": ["list of claims with no basis in source"],
  "educational_feedback": "constructive guidance for improving groundness and alignment",
  "pass_threshold_met": [true/false]
}
</format>

<tone>Objective, rigorous, constructive, and educationally supportive. Maintain high standards while providing clear pathways for improvement.</tone>

<source_material>
[source_material]
</source_material>

<response_to_evaluate>
[response_to_evaluate]
</response_to_evaluate>

Now perform the groundness evaluation and output ONLY the JSON assessment.