RCS top-1 absolute consistency review

You are reviewing RCS top-1 candidate consistency.

Important:
- There is NO established ground-truth dataset in this project.
- Judge only whether `top1_name` is consistent with the original `query`.
- Do not compare against any previous LLM label.
- If the query is a compound structure and top1 covers only one part, label partial_or_narrower.

Labels:
- aligned: top1 directly names the same anatomical structure, accepted synonym,
  conventional spelling/word-order variant, or same named entity.
- broader_parent: top1 is broader than the query but anatomically the right
  parent/container. This is not exact, but it is not wrong.
- partial_or_narrower: top1 is only one component/subdivision of a broader or
  compound query, or is too narrow for the query.
- wrong: top1 is anatomically different/off-structure/off-lobe.
- ambiguous: multiple plausible interpretations; cannot decide confidently.
- source_or_ontology_issue: query appears misspelled/source-specific, or likely
  missing at HOMBA granularity, so candidate consistency cannot be strictly
  judged from the pair alone.

Return STRICT JSON only:
{
  "results": [
    {
      "record_id": <int>,
      "label": "<one label>",
      "certainty": "high|medium|low",
      "confidence": <float 0..1>,
      "reason": "<max 220 chars>"
    }
  ]
}

Bias: be strict. If a candidate is only a loose anatomical neighbor, label wrong.

You are reviewing RCS top-1 candidate consistency.

Important:
- There is NO established ground-truth dataset in this project.
- Judge only whether `top1_name` is consistent with the original `query`.
- Do not compare against any previous LLM label.
- If the query is a compound structure and top1 covers only one part, label partial_or_narrower.

Labels:
- aligned: top1 directly names the same anatomical structure, accepted synonym,
  conventional spelling/word-order variant, or same named entity.
- broader_parent: top1 is broader than the query but anatomically the right
  parent/container. This is not exact, but it is not wrong.
- partial_or_narrower: top1 is only one component/subdivision of a broader or
  compound query, or is too narrow for the query.
- wrong: top1 is anatomically different/off-structure/off-lobe.
- ambiguous: multiple plausible interpretations; cannot decide confidently.
- source_or_ontology_issue: query appears misspelled/source-specific, or likely
  missing at HOMBA granularity, so candidate consistency cannot be strictly
  judged from the pair alone.

Return STRICT JSON only:
{
  "results": [
    {
      "record_id": <int>,
      "label": "<one label>",
      "certainty": "high|medium|low",
      "confidence": <float 0..1>,
      "reason": "<max 220 chars>"
    }
  ]
}

Bias: distinguish exact/synonym from parent fallback. Use broader_parent when
the candidate is anatomically a valid parent but not the exact query.

You are reviewing RCS top-1 candidate consistency.

Important:
- There is NO established ground-truth dataset in this project.
- Judge only whether `top1_name` is consistent with the original `query`.
- Do not compare against any previous LLM label.
- If the query is a compound structure and top1 covers only one part, label partial_or_narrower.

Labels:
- aligned: top1 directly names the same anatomical structure, accepted synonym,
  conventional spelling/word-order variant, or same named entity.
- broader_parent: top1 is broader than the query but anatomically the right
  parent/container. This is not exact, but it is not wrong.
- partial_or_narrower: top1 is only one component/subdivision of a broader or
  compound query, or is too narrow for the query.
- wrong: top1 is anatomically different/off-structure/off-lobe.
- ambiguous: multiple plausible interpretations; cannot decide confidently.
- source_or_ontology_issue: query appears misspelled/source-specific, or likely
  missing at HOMBA granularity, so candidate consistency cannot be strictly
  judged from the pair alone.

Return STRICT JSON only:
{
  "results": [
    {
      "record_id": <int>,
      "label": "<one label>",
      "certainty": "high|medium|low",
      "confidence": <float 0..1>,
      "reason": "<max 220 chars>"
    }
  ]
}

Bias: expert conservative adjudication. Use ambiguous only when a human
neuroanatomy review is genuinely needed. Return pure JSON, no markdown.