"""两个半判据合成一个判定 —— 优先级不能反。

`combine_verdict` 是整套东西的裁决点,它的每一条优先级都对应一个具体的
坏结果:

* error 高于一切 —— 供应商抖一下不是「助手答得差」;
* 确定性阻断高于评委 —— 评委是模型读模型,不能给一个查得出来的编造引用翻案;
* 评委不可用 = error,不是 fail —— 否则评委自己的故障会长在助手的趋势线上;
* 硬维度地板由服务端兜 —— 「pass」不能等于「接地性只有 1 分但它说 pass」。
"""
from __future__ import annotations

from dataclasses import dataclass, field

from app.assistant.evaluation import checks
from app.assistant.evaluation.constants import (
    CHECK_TURN_FAILED,
    CHECK_UNGROUNDED_CITATION,
    DIM_ACCURATE,
    DIM_ACTIONABLE,
    DIM_COMPLETE,
    DIM_ESCALATION,
    DIM_GROUNDED,
    DIM_TONE,
    FAIL_JUDGE_ERROR,
    FAIL_KB_DOC,
    FAIL_TOOL_FIELD_OR_DATA,
    VERDICT_ERROR,
    VERDICT_FAIL,
    VERDICT_PASS,
)
from app.assistant.evaluation.runner import combine_verdict


@dataclass
class _Judge:
    available: bool = True
    scores: dict = field(default_factory=dict)
    verdict: str | None = VERDICT_PASS
    failure_type: str | None = None


def _good_scores(**overrides) -> dict:
    base = {
        DIM_GROUNDED: 5,
        DIM_ACCURATE: 5,
        DIM_COMPLETE: 4,
        DIM_ACTIONABLE: 4,
        DIM_ESCALATION: 5,
        DIM_TONE: 4,
    }
    base.update(overrides)
    return base


def _finding(code: str) -> checks.Finding:
    severity = (
        checks.SEVERITY_ERROR
        if code in {CHECK_TURN_FAILED}
        else checks.SEVERITY_BLOCKING
    )
    return checks.Finding(code=code, severity=severity, detail="")


def test_a_clean_attempt_with_a_happy_judge_passes():
    """🔴 锚。这条一旦不再通过,下面每一条阴性用例都失去意义。"""
    verdict, failure = combine_verdict(
        findings=[], judge=_Judge(scores=_good_scores()),
    )
    assert (verdict, failure) == (VERDICT_PASS, None)


def test_an_error_finding_beats_everything():
    """模型没答上话,这一轮什么质量都没观察到。"""
    verdict, failure = combine_verdict(
        findings=[_finding(CHECK_TURN_FAILED)],
        judge=_Judge(scores=_good_scores(), verdict=VERDICT_PASS),
    )
    assert verdict == VERDICT_ERROR
    assert failure is None


def test_a_blocking_finding_cannot_be_overruled_by_the_judge():
    """🔴 这是「确定性检查先跑」的全部意义。

    评委给了满分并判 pass —— 它当然会,一个编造出来的链接读起来和真的一样。
    但「这个链接在不在本轮任何工具结果里」是查得出来的,查出来的事实不接受
    投票翻案。
    """
    verdict, failure = combine_verdict(
        findings=[_finding(CHECK_UNGROUNDED_CITATION)],
        judge=_Judge(scores=_good_scores(), verdict=VERDICT_PASS),
    )
    assert verdict == VERDICT_FAIL
    assert failure == FAIL_TOOL_FIELD_OR_DATA


def test_the_judges_own_classification_wins_when_it_has_one():
    """确定性闸负责「拦下来」,分类仍优先采信评委 —— 它读过全文。"""
    verdict, failure = combine_verdict(
        findings=[_finding(CHECK_UNGROUNDED_CITATION)],
        judge=_Judge(
            scores=_good_scores(), verdict=VERDICT_FAIL, failure_type=FAIL_KB_DOC,
        ),
    )
    assert (verdict, failure) == (VERDICT_FAIL, FAIL_KB_DOC)


def test_an_unavailable_judge_is_an_error_not_a_failure():
    """🔴 评委自己的故障不能画进助手的趋势线。"""
    verdict, failure = combine_verdict(findings=[], judge=_Judge(available=False))
    assert (verdict, failure) == (VERDICT_ERROR, FAIL_JUDGE_ERROR)


def test_a_judge_that_returned_no_scores_is_an_error():
    verdict, failure = combine_verdict(
        findings=[], judge=_Judge(scores={}, verdict=VERDICT_PASS),
    )
    assert (verdict, failure) == (VERDICT_ERROR, FAIL_JUDGE_ERROR)


def test_a_missing_judge_object_is_an_error():
    verdict, failure = combine_verdict(findings=[], judge=None)
    assert (verdict, failure) == (VERDICT_ERROR, FAIL_JUDGE_ERROR)


def test_a_floored_hard_dimension_fails_even_when_the_judge_said_pass():
    """🔴 服务端兜底:「pass」不能等于「接地性 1 分但它说 pass」。"""
    verdict, _failure = combine_verdict(
        findings=[],
        judge=_Judge(
            scores=_good_scores(**{DIM_GROUNDED: 1}), verdict=VERDICT_PASS,
        ),
    )
    assert verdict == VERDICT_FAIL


def test_a_floored_soft_dimension_does_not_fail_on_its_own():
    """护栏:语气差不是不能给员工看。硬地板只有接地性和准确性两条。"""
    verdict, _failure = combine_verdict(
        findings=[],
        judge=_Judge(scores=_good_scores(**{DIM_TONE: 1}), verdict=VERDICT_PASS),
    )
    assert verdict == VERDICT_PASS


def test_a_partially_scored_reply_is_not_failed_for_the_dimensions_it_omitted():
    """评委只填了一部分维度时,没填的那些不该被当成 0 分。"""
    verdict, _failure = combine_verdict(
        findings=[],
        judge=_Judge(
            scores={DIM_GROUNDED: 5, DIM_ACCURATE: 4}, verdict=VERDICT_PASS,
        ),
    )
    assert verdict == VERDICT_PASS


def test_a_judge_with_scores_but_no_verdict_field_still_yields_one():
    """维度分齐了却漏了 verdict —— 这是一次可用的测量,不该丢掉。"""
    verdict, _failure = combine_verdict(
        findings=[], judge=_Judge(scores=_good_scores(), verdict=None),
    )
    assert verdict == VERDICT_PASS