"""合成场景在隔离PG上跑真实公开用例；不声称自然语言路由或真实模型效果。"""

from pathlib import Path

import pytest

from muse.作品规划.接口 import 档案保存
from muse.效果评测.接口 import 执行行为场景, 服务观察适配, 载入行为场景
from muse.正文写作.接口 import 文本节点, 正文草稿, 段落

pytestmark = pytest.mark.数据库
场景集 = 载入行为场景((Path(__file__).parents[1] / "夹具/行为评测/诊断机器味场景.json").read_text())


@pytest.mark.case_id(
    "NC-w24-241005",
    environment="隔离PG真服务",
    given="固定六场景及逐例原始观察",
    when="显式脚本或真实隔离服务执行，注入缺失、禁止动作及伪报告",
    then=["场景逐个独立裁决，报告重算和读回、来源前后哈希、失败与模式均可追溯"],
    contract="docs/系统架构/新版设计/模块设计/B10-效果评测.md",
)
@pytest.mark.parametrize("scene", 场景集, ids=lambda s: s.scenario_id)
def test_真实服务观察报告读回及正文不变__241005(章后环境, scene):
    app, author = 章后环境["装配"], 章后环境["作者"]
    works, body, cases = app.要求作品(), app.要求正文(), app.要求审校()
    schema = works.新档案结构(author, "synthetic:demo", "work_core", 1)
    works.保存档案(
        author, "create-eval-work", 档案保存("synthetic:demo", 0, {"名称": "合成诊断场景"}, schema)
    )
    works.添加章节(
        author, "create-eval-ch", "synthetic:demo", "synthetic-ch", "场景", 预期目录版本=0
    )
    body.保存人工(
        author,
        "seed-eval-text",
        "synthetic-ch",
        0,
        正文草稿((段落("p1", (文本节点(scene.text),)),)),
    )
    target = {"work_id": "synthetic:demo", "chapter_id": "synthetic-ch", "branch_id": "main"}
    before = body.读取正文(author, "synthetic-ch")
    with app.要求数据库().连接(只读=True) as conn:
        commands = conn.execute("SELECT count(*) FROM muse_change_command").fetchone()[0]
    result = 执行行为场景(scene, 服务观察适配(body, cases, author, target))
    assert result["passed"], result
    assert result["mode"] == "service" and not result["model_verified"]
    assert body.读取正文(author, "synthetic-ch") == before
    with app.要求数据库().连接(只读=True) as conn:
        assert conn.execute("SELECT count(*) FROM muse_change_command").fetchone()[0] == commands
        assert (
            conn.execute(
                "SELECT count(*) FROM muse_attempt WHERE call_reference IS NOT NULL"
            ).fetchone()[0]
            == 0
        )
        assert conn.execute("SELECT count(*) FROM muse_diagnosis").fetchone()[0] == (
            1 if scene.expected == "report" else 0
        )
    if scene.expected == "report":
        observation = result["observation"]
        assert observation["report"] == observation["report_readback"]
        # 覆盖区间只统计真正执行的确定性规则；本场景没有激活规则，故为空。
        coverage = observation["report"]["coverage"]
        assert coverage["text_length"] == len(scene.text)
        assert coverage["rules_executed"] == 0 and observation["report"]["empty_library"] is True
        assert coverage["deterministic"] == (
            [[0, len(scene.text)]] if coverage["rules_executed"] else []
        )