agrag.ingestion.resolve.zone_classifier.select_llm_pairs

select_llm_pairs(candidates:list[tuple[int, int, float]], *, max_pairs:int = MAX_LLM_PAIRS) -> list[tuple[int, int]]

Rank ambiguous candidates for LLM review and keep the first max_pairs.

The order is similarity, highest first. Pairs with equal similarity keep the lower (left_index, right_index) first, so the same input always yields the same pairs. The pairs this drops are the least similar, and the caller must report them as undecided.

Parameters:

• candidates (list[tuple[int, int, float]]) – (left_index, right_index, similarity) triples.
• max_pairs (int) – Maximum pairs to return.

Returns:

• list[tuple[int, int]] – Index pairs in the order above, at most max_pairs of them.