prompt (do not assume the model still has earlier messages). Convert extracted fields to citation form:
{"quantity": {"value": 2, "word_id_groups": [7]}}→Quantity: 2[cite:7]- Never write
word_id_groups=[7]in the prompt.
Authentication and enablement
No integration. Uses the agent’s LLM. Default-on. Applied learning rules for this comparison are injected by the orchestrator.Inputs
prompt(required): the comparison, including both sides and citation markers.
Output
{response, model, tokens_used} — a text verdict. response is not a parsed object.
Limits and side effects
- LLM call; no external writes.
- Not a substitute for
structured_data_extraction(extract first, then compare).
Expected errors
- Empty
prompt. - Provider failure.