Edit Merge_pair_eval Template

Merge_pair_eval v1

Viewer

Test Case:

Template

Preview

History

System prompt (optional) User prompt (template)

Evaluate each candidate pair for potential duplication among entities extracted from a professional ethics case.

Entity type: {{ type_label }} ({{ storage_label }}){{ concept_line }}

For each pair, determine:
- "merge" if both refer to the SAME specific concept extracted from different document sections (Facts vs Discussion)
- "keep_separate" if they are DISTINCT concepts that should remain as separate entities

{{ pairs_text }}

STRICT RULES -- most pairs should be "keep_separate":
- merge: Same label or trivially rephrased label referring to the exact same specific concept
- keep_separate: Different roles of the same person (e.g., "Engineer A Design Role" vs "Engineer A Safety Role")
- keep_separate: Different BER case references (e.g., "BER 85-3 ..." vs "BER 98-8 ...")
- keep_separate: Related but distinct concepts (e.g., "Safety Obligation" vs "Competence Obligation")
- keep_separate: Different aspects or facets of a broader topic
- keep_separate: General vs specific versions (e.g., "Safety Constraint" vs "Structural Safety Constraint")
- When in doubt, use "keep_separate"

Return ONLY valid JSON:
{"evaluations": [{"pair": [id_a, id_b], "verdict": "merge"|"keep_separate", "reason": "brief explanation"}]}

Loading preview...

Version History

No version history yet. Save changes to create versions.

Saved

1229 chars

Reference

Template Variables

Storage type (class/individual)

Title-cased entity type (e.g. Roles)

Assembled candidate-pair descriptions

Per-category concept definition line, or empty

Output Schema

evaluations

pair* list[int]

reason str

verdict str

JSON Wrapper

Appended to the prompt by the extractor:

IMPORTANT: Wrap your response as a JSON object with exactly two keys:
{"new_merge_pair_eval_classes": [...], "merge_pair_eval_individuals": [...]}
If there are no individuals to report, use an empty array for "merge_pair_eval_individuals".