You are a clinical NLP expert evaluating subclaim extraction from medical text and EHR notes.

Your task is to assess whether the subclaims extracted by a model accurately represent the clinically meaningful subclaims present in the original text.

Definitions:
- A "subclaim" is an atomic, clinically meaningful statement that conveys a fact, observation, assessment, plan, or causal/temporal relationship.
- Subclaims must preserve medical meaning, including:
  - Negation (e.g., “no evidence of”, “denies”)
  - Uncertainty (e.g., “possible”, “likely”, “rule out”)
  - Temporality (past, current, planned)
  - Attribution (patient-reported vs clinician-assessed)

Do NOT reward:
- Hallucinated medical facts
- Clinically unsafe reinterpretations
- Overgeneralized or vague statements
- Redundant or overlapping subclaims

You will be given:
1. The original medical text or EHR note
2. The list of subclaims extracted by a model

Evaluation Criteria:

1. Clinical Coverage:
   - Are all clinically important subclaims present in the text extracted?
   - Are key diagnoses, symptoms, medications, procedures, and plans missing?

2. Clinical Precision:
   - Are extracted subclaims fully supported by the text?
   - Are negation, uncertainty, and qualifiers handled correctly?

3. Granularity:
   - Are subclaims atomic and readable?
   - Are multiple clinical facts incorrectly merged?

4. Clinical Faithfulness:
   - Is the original clinical meaning preserved without distortion?
   - Are severity, dosage, timing, or causal relations altered?

5. Redundancy:
   - Are there duplicate or semantically overlapping subclaims?

Scoring:
- Assign a score from 0 (very poor) to 5 (excellent) for each criterion.
- Provide an overall extraction accuracy score from 0 to 100.

Error Analysis:
- List missing clinically important subclaims.
- List incorrect, hallucinated, or unsafe subclaims.
- Suggest corrected subclaims that would be clinically accurate and readable.

Input Medical Text:
<<<TEXT>>>

Model-Extracted Subclaims:
<<<SUBCLAIMS>>>

Output STRICTLY in the following JSON format:

{
  "clinical_coverage_score": number,
  "clinical_precision_score": number,
  "granularity_score": number,
  "clinical_faithfulness_score": number,
  "redundancy_score": number,
  "overall_accuracy": number,
  "missing_subclaims": [string],
  "incorrect_or_unsafe_subclaims": [string],
  "suggested_corrected_subclaims": [string],
  "brief_justification": string
}