tests/test_compatibility_regression.py

"""相容性确定性回归基准（compatibility-engine-upgrade 任务 8.1）。

金标准用例覆盖经典相互作用 + 纠偏 + 别名，断言风险**档位**与命中**反应集合**。
确定性、离线运行（直接喂官能团 id，不依赖网络/LLM/RDKit），守护规则/SMARTS 改动
不致非预期回归（需求 1.4 / 2.2 / 2.3）。
"""

from __future__ import annotations

import pytest

from skills.compatibility import rules


def _assess(group_ids, excipients):
    return rules.evaluate([{"id": g} for g in group_ids], excipients)


def _excipient_reactions(assessment, idx=0):
    er = assessment.excipient_results[idx]
    return er.risk, {r.reaction for r in er.reactions}


# ---------------------------------------------------------------------------
# 经典相互作用金标准
# ---------------------------------------------------------------------------

@pytest.mark.parametrize(
    "group_ids,excipient,expected_overall,expected_reactions",
    [
        # 美拉德：伯胺 + 还原糖（乳糖）→ 高风险
        (["primary_amine"], "乳糖", rules.RISK_HIGH, {"maillard"}),
        # 碱催化水解：酯 + 偏碱（硬脂酸镁）→ 高风险
        (["ester"], "硬脂酸镁", rules.RISK_HIGH, {"hydrolysis"}),
        # 内酯 + 碱 → 高风险水解
        (["lactone"], "硬脂酸镁", rules.RISK_HIGH, {"hydrolysis"}),
        # 氧化：酚 + 过氧化物（聚维酮）→ 中风险
        (["phenol"], "聚维酮", rules.RISK_MEDIUM, {"oxidation"}),
        # 氧化：巯基 + 过氧化物 → 高风险
        (["thiol"], "聚维酮", rules.RISK_HIGH, {"oxidation"}),
        # 酸碱：羧基 + 偏碱 → 中风险
        (["carboxylic_acid"], "硬脂酸镁", rules.RISK_MEDIUM, {"acid_base"}),
        # 光降解：酚 + 二氧化钛（光催化剂 + 金属）→ 中风险（氧化 + 光降解）
        (["phenol"], "二氧化钛", rules.RISK_MEDIUM, {"oxidation", "photodegradation"}),
        # Michael：Michael 受体 + 亚硫酸盐 → 中风险
        (["michael_acceptor"], "亚硫酸氢钠", rules.RISK_MEDIUM, {"michael"}),
    ],
)
def test_classic_interactions(group_ids, excipient, expected_overall, expected_reactions):
    a = _assess(group_ids, [excipient])
    assert a.overall_risk == expected_overall, (group_ids, excipient)
    risk, reactions = _excipient_reactions(a)
    assert expected_reactions.issubset(reactions), (reactions, expected_reactions)


# ---------------------------------------------------------------------------
# 纠偏：淀粉 / MCC 假阳性（需求 2.2）
# ---------------------------------------------------------------------------

def test_starch_amine_not_high_maillard():
    """淀粉 + 伯胺 → 中风险（经还原糖杂质途径），而非默认高风险。"""
    a = _assess(["primary_amine"], ["淀粉"])
    risk, reactions = _excipient_reactions(a)
    assert "maillard" in reactions
    assert risk == rules.RISK_MEDIUM
    assert a.overall_risk == rules.RISK_MEDIUM


def test_mcc_amine_no_maillard():
    """微晶纤维素 + 伯胺 → 不触发美拉德（仅可能吸附低风险）。"""
    a = _assess(["primary_amine"], ["微晶纤维素"])
    risk, reactions = _excipient_reactions(a)
    assert "maillard" not in reactions


def test_lactose_still_high():
    """乳糖（真还原糖）+ 伯胺仍为高风险（纠偏不误伤真阳性）。"""
    a = _assess(["primary_amine"], ["乳糖"])
    assert a.overall_risk == rules.RISK_HIGH


def test_aromatic_amine_maillard_differentiated_from_aliphatic():
    """芳香胺 + 还原糖 → 低风险美拉德，区别于脂肪伯/仲胺的高风险（需求 2.1.3）。"""
    aromatic = _assess(["aromatic_amine"], ["乳糖"])
    arom_reactions = {r.reaction: r.risk for r in aromatic.excipient_results[0].reactions}
    assert arom_reactions.get("maillard") == rules.RISK_LOW

    aliphatic = _assess(["primary_amine"], ["乳糖"])
    ali_reactions = {r.reaction: r.risk for r in aliphatic.excipient_results[0].reactions}
    assert ali_reactions.get("maillard") == rules.RISK_HIGH

    # 差异化：脂肪胺风险严格高于芳香胺。
    assert rules.RISK_ORDER[ali_reactions["maillard"]] > rules.RISK_ORDER[arom_reactions["maillard"]]


def test_aromatic_amine_oxidation_prone():
    """芳香胺 + 过氧化物辅料 → 中风险氧化（苯胺类氧化敏感）。"""
    a = _assess(["aromatic_amine"], ["聚维酮"])
    reactions = {r.reaction: r.risk for r in a.excipient_results[0].reactions}
    assert reactions.get("oxidation") == rules.RISK_MEDIUM


# ---------------------------------------------------------------------------
# 最长别名（需求 2.3）
# ---------------------------------------------------------------------------

@pytest.mark.parametrize(
    "name,wrong_parent_tag_absent",
    [
        ("预胶化淀粉", "reducing_sugar_impulse"),  # 不应带淀粉的还原糖杂质途径标签
        ("交联聚维酮", None),
    ],
)
def test_longest_alias_not_parent(name, wrong_parent_tag_absent):
    a = _assess(["primary_amine"], [name])
    er = a.excipient_results[0]
    assert er.known is True
    if name == "预胶化淀粉":
        # 预胶化淀粉档案 tags=[high_moisture]，不含 reducing_sugar_impurity → 不命中美拉德。
        assert "reducing_sugar_impurity" not in er.tags
        assert "maillard" not in {r.reaction for r in er.reactions}
    if name == "交联聚维酮":
        # 交联聚维酮 tags=[peroxide]（无 high_moisture），区别于母体聚维酮。
        assert "high_moisture" not in er.tags


# ---------------------------------------------------------------------------
# 不确定 + 多辅料聚合
# ---------------------------------------------------------------------------

def test_no_groups_uncertain():
    a = _assess([], ["乳糖", "硬脂酸镁"])
    assert a.overall_risk == rules.RISK_UNCERTAIN
    assert a.overall_confidence == rules.CONF_LOW


def test_multi_excipient_worst_aggregation():
    """多辅料取最坏：乳糖(高) + 甘露醇(无) → 整体高。"""
    a = _assess(["primary_amine"], ["乳糖", "甘露醇"])
    assert a.overall_risk == rules.RISK_HIGH


def test_determinism_repeated_runs():
    first = _assess(["primary_amine", "ester"], ["乳糖", "硬脂酸镁"]).summary()
    again = _assess(["primary_amine", "ester"], ["乳糖", "硬脂酸镁"]).summary()
    assert first == again