harness_zbt/backend/tests/test_knowledge_chunking.py

from zbt.domains.knowledge.chunking import DocumentChunker


def test_chunker_preserves_short_paragraph_boundaries() -> None:
    chunker = DocumentChunker(max_chars=36, overlap_chars=8)

    chunks = chunker.split(
        "等待期为30天。\n\n意外伤害不受等待期限制。\n\n报销需提供发票和诊断证明。"
    )

    assert chunks == [
        "等待期为30天。",
        "意外伤害不受等待期限制。",
        "报销需提供发票和诊断证明。",
    ]
    assert all(len(chunk) <= 36 for chunk in chunks)


def test_chunker_splits_long_paragraph_with_context_overlap() -> None:
    chunker = DocumentChunker(max_chars=24, overlap_chars=6)
    content = "第一条说明等待期规则，疾病医疗等待期为三十天。第二条说明意外伤害不受等待期限制。"

    chunks = chunker.split(content)

    assert len(chunks) >= 2
    assert all(len(chunk) <= 24 for chunk in chunks)
    assert chunks[0][-6:] == chunks[1][:6]
    assert "".join([chunks[0], *[chunk[6:] for chunk in chunks[1:]]]) == content


def test_chunker_rejects_blank_document() -> None:
    chunker = DocumentChunker()

    assert chunker.split(" \n\n ") == []