muse-agent-example/tests/端到端/test_改进验证与版本启用.py

"""旧 humanization 成功自动登记 win/lesson 的观察按 B07/W25 等价边界承接。

成功运行只形成有证据的运行观察；不因 win 标签或 promote 状态自动宣称改进有效、
方法或规则已启用。升格须真实反馈、精确目标与 owner 作者决定（S01/B10）。
"""

from dataclasses import replace
from uuid import uuid4

import pytest

from muse.任务运行.接口 import (
    任务服务,
    任务请求,
    步骤处理器,
    步骤结果,
    步骤计划,
)
from muse.作者经验.接口 import 作者经验服务, 改进定位, 运行观察来源, 运行观察请求
from muse.共享.调用身份 import 内容用途, 用途, 调用身份
from muse.共享.错误 import Muse错误
from muse.正式变更.接口 import 参与者目录, 正式变更服务
from muse.编排.接口 import 流程定义, 流程服务, 流程登记

pytestmark = pytest.mark.数据库


@pytest.fixture
def 运行观察环境(应用测试库):
    pool = 应用测试库[用途.生产]
    author = 调用身份("improvement-author", None, 用途.生产, 内容用途.生成)
    registry = 流程登记()

    def 处理(context):
        return 步骤结果({"recorded": True})

    registry.登记处理器(步骤处理器("observation.test", "1", 处理, "v1", "v1"))
    registry.登记类型("observation-test", 必需保护=())
    registry.登记恢复检查("observation-test", "1", lambda _: None)
    runtime = 任务服务(pool, registry)
    流程服务(runtime, registry).发布(
        流程定义("observation-test", "1", (步骤计划("测试步骤", "observation.test", "1"),))
    )
    formal = 正式变更服务(pool, 参与者目录())
    return {
        "pool": pool,
        "author": author,
        "run": runtime,
        "service": 作者经验服务(pool, formal),
    }


def _开始(env, *, command=None):
    actor = env["author"]
    request = 任务请求(
        "observation-test",
        command or str(uuid4()),
        actor.作者,
        用途.生产,
        actor.内容用途,
        {"private_input": "不应进入观察的正文"},
        "synthetic-policy-1",
        "synthetic-release-1",
        {
            "source_scope": {},
            "schema_versions": {},
            "authorization": "synthetic-grant",
            "budget": {},
            "stop_conditions": {},
        },
    )
    tid = env["run"].创建任务(request, "observation-test", "1")
    claim = env["run"].领取步骤("worker", ["observation.test"], 任务ID=tid, 作者=actor.作者)
    assert claim is not None
    return tid, claim


def _来源(env, tid, attempt_id):
    return env["service"].读取运行观察来源(env["author"], tid, attempt_id)


def _完成(env, **kwargs):
    tid, claim = _开始(env, **kwargs)
    env["run"].执行一步(claim)
    return tid, claim, _来源(env, tid, claim.尝试ID)


def _观察(source, *, phenomenon, quote, background, issue=""):
    return 运行观察请求(
        (运行观察来源(**source["locator"]),),
        phenomenon,
        quote,
        background,
        issue,
    )


def _无自动升格(env, saved):
    """成功运行只形成观察：不自动产生偏好、观察汇总或可启停的改进/规则目标。"""
    assert saved["state"] == "observation"
    assert saved["payload"]["causality"] == "not_established"
    assert saved["payload"]["measurement_source"] == "runtime_evidence"
    assert saved["payload"]["interpretation_source"] == "author_report"
    assert env["service"].列出偏好(env["author"]) == []
    assert env["service"].读取观察(env["author"])["observations"] == []
    with pytest.raises(Muse错误):
        env["service"].请求启停(
            env["author"],
            "no-auto-promotion",
            改进定位(str(saved["observation_id"]), 1, saved["content_hash"]),
            "enable",
        )


def test_接受正文候选只留决定反馈不自动登记胜例__9fc9e5(运行观察环境):
    """旧 accept 登记 win / discard 登记 lesson：新版只保留有证据观察与作者原话。

    旧观察同保留 candidate_sha256 与 creator 的输入在此转为实时反馈与观察，
    不自动产生任何 lessons/win 记录。
    """
    env = 运行观察环境
    tid, claim, source = _完成(env)
    candidate_sha256 = "abc"
    saved = env["service"].记录运行观察(
        env["author"],
        "decision-event",
        _观察(
            source,
            phenomenon="作者决定正文候选去留",
            quote="丢弃该候选，改写转折前的铺陈。",
            background=f"接受或丢弃候选决定（candidate_sha256={candidate_sha256}，creator=决定正文候选去留）。",
        ),
    )
    assert saved["payload"]["interpretation_source"] == "author_report"
    读回 = env["service"].读取运行观察(env["author"], saved["observation_id"])
    assert candidate_sha256 in 读回["payload"]["background"]
    assert "丢弃该候选" in 读回["payload"]["quote"]
    _无自动升格(env, saved)


def test_审查批次写回保留批次键与通过数__1f9573(运行观察环境):
    env = 运行观察环境
    _, _, source = _完成(env)
    saved = env["service"].记录运行观察(
        env["author"],
        "review-batch",
        _观察(
            source,
            phenomenon="知识卡三角色审核已写回",
            quote="保留两票通过的一卡。",
            background="批量键=review-3-cal-1-a；pass=2（旧 win 登记的批次 detail 转为观察背景）。",
        ),
    )
    读回 = env["service"].读取运行观察(env["author"], saved["observation_id"])
    assert (
        "review-3-cal-1-a" in 读回["payload"]["background"]
        and "pass=2" in 读回["payload"]["background"]
    )
    _无自动升格(env, saved)


def test_知识升格窗批完成保留已处理窗口数__b87d69(运行观察环境):
    env = 运行观察环境
    _, _, source = _完成(env)
    saved = env["service"].记录运行观察(
        env["author"],
        "upgrade-window",
        _观察(
            source,
            phenomenon="作品面知识升格窗批完成",
            quote="四个窗口全部处理完成。",
            background="windows_done=4；run_id=upgrade-5-done4。",
        ),
    )
    读回 = env["service"].读取运行观察(env["author"], saved["observation_id"])
    assert "windows_done=4" in 读回["payload"]["background"]
    assert "upgrade-5-done4" in 读回["payload"]["background"]
    _无自动升格(env, saved)


def test_章后知识抽取完成保留章序__679e5e(运行观察环境):
    env = 运行观察环境
    _, _, source = _完成(env)
    saved = env["service"].记录运行观察(
        env["author"],
        "chapter-extract",
        _观察(
            source,
            phenomenon="章后知识抽取落库成功",
            quote="第九章抽取结果为既有事实增量。",
            background="chapter_order=9；抽取不改写已确认事实。",
        ),
    )
    读回 = env["service"].读取运行观察(env["author"], saved["observation_id"])
    assert "chapter_order=9" in 读回["payload"]["background"]
    _无自动升格(env, saved)


def test_机器味案例批次落库保留64位运行身份__3b729b(运行观察环境):
    env = 运行观察环境
    _, _, source = _完成(env)
    run_id = "a" * 64
    saved = env["service"].记录运行观察(
        env["author"],
        "capture-cases",
        _观察(
            source,
            phenomenon="AI味案例卡批次已落库",
            quote="本批案例进入待复核，未自动启用规则。",
            background=f"run_id={run_id}；creator=记录机器味案例。",
        ),
    )
    读回 = env["service"].读取运行观察(env["author"], saved["observation_id"])
    assert run_id in 读回["payload"]["background"]
    assert "记录机器味案例" in 读回["payload"]["background"]
    _无自动升格(env, saved)


def test_预防材料装配完成只留观察不自动启用__f30e89(运行观察环境):
    env = 运行观察环境
    _, _, source = _完成(env)
    saved = env["service"].记录运行观察(
        env["author"],
        "prevention-assembled",
        _观察(
            source,
            phenomenon="生成前 AI 味预防合同已装配",
            quote="预防材料已就位，待作者启用。",
            background="creator=预防机器味；合同仅作候选，未消费于生成。",
        ),
    )
    读回 = env["service"].读取运行观察(env["author"], saved["observation_id"])
    assert "预防机器味" in 读回["payload"]["background"]
    _无自动升格(env, saved)


def test_稳定评测报告完成保留运行与创建者__5b8778(运行观察环境):
    env = 运行观察环境
    _, _, source = _完成(env)
    saved = env["service"].记录运行观察(
        env["author"],
        "stable-score",
        _观察(
            source,
            phenomenon="正文盲评稳定报告",
            quote="本次盲评结论稳定，原文不变。",
            background="run_id=score-run-1；creator=评估内容质量。",
        ),
    )
    读回 = env["service"].读取运行观察(env["author"], saved["observation_id"])
    assert "score-run-1" in 读回["payload"]["background"]
    assert "评估内容质量" in 读回["payload"]["background"]
    _无自动升格(env, saved)


def test_规则激活需评审凭据且候选保持__922777(章后环境):
    """旧 candidate→holdout→approver→active：导出凭据前 candidate 不能激活。"""
    from muse.审校修订.接口 import 规则状态命令

    app, author = 章后环境["装配"], 章后环境["作者"]
    review = app.要求审校()
    review.导入规则候选(author)
    库 = review.读取规则库(author, 预览候选=True)
    rule = next(iter(库["rules"].values()))
    with pytest.raises(Muse错误):
        review.打开规则状态审阅(
            author,
            规则状态命令(
                rule["id"],
                rule["version"],
                "enable",
                expected_activation_revision=0,
            ),
        )
    状态 = review.读取规则库(author, 预览候选=True)
    assert 状态["rules"][rule["id"]]["status"] == "candidate"
    # 候选未被激活时没有启用投影：读取规则状态只回未激活。
    assert review.读取规则状态(author, rule["id"])["current"] is None


def test_启用改进必须精确目标与作者审阅__0f3902(运行观察环境):
    """旧 promote 必须指明目标：新版改进启停锁定精确版本且拒绝非当前作者。"""

    env = 运行观察环境
    saved = env["service"].记录运行观察(
        env["author"],
        "target-required",
        _观察(
            _完成(env)[2],
            phenomenon="升格必须精确目标",
            quote="没有目标不启用。",
            background="旧目标类型校验的等价：改进必须精确版本目标。",
        ),
    )
    # 观察不是已批准的改进，任何启停都被拒绝；陌生作者同样不可读。
    wrong = 改进定位(str(saved["observation_id"]), 2, "0" * 64)
    with pytest.raises(Muse错误):
        env["service"].请求启停(env["author"], "bad-target", wrong, "enable")
    stranger = replace(env["author"], 作者="stranger")
    with pytest.raises(Muse错误):
        env["service"].读取运行观察(stranger, str(saved["observation_id"]))


def test_停用后旧凭据不能重新启用__c4f7f6(运行观察环境):
    """旧 reject 终态不再流转：新版停用后须重新验证，相同命令重放拒不覆盖。"""
    env = 运行观察环境
    saved = env["service"].记录运行观察(
        env["author"],
        "reject-once",
        _观察(
            _完成(env)[2],
            phenomenon="停用即终态",
            quote="停用后不再按旧意向恢复。",
            background="reject 路径：历史命令重放不改写内容。",
        ),
    )
    with pytest.raises(Muse错误, match="更改来源或内容"):
        env["service"].记录运行观察(
            env["author"],
            "reject-once",
            replace(
                _观察(
                    _完成(env)[2],
                    phenomenon="停用即终态",
                    quote="不同原话",
                    background="覆盖被拒绝",
                )
            ),
        )
    assert env["service"].读取运行观察(env["author"], saved["observation_id"]) == (
        env["service"].读取运行观察(env["author"], saved["observation_id"])
    )


def test_缺作者决定身份时启停被拒__b8d6b4(运行观察环境):
    """旧 started review 缺 decided_by 拒绝：新版作者身份与 S01 决定是启停前提。"""
    env = 运行观察环境
    saved = env["service"].记录运行观察(
        env["author"],
        "decided-by-required",
        _观察(
            _完成(env)[2],
            phenomenon="作者决定身份必填",
            quote="没有作者决定不启停。",
            background="decided_by 等价：作者必须本人操作。",
        ),
    )
    # 未经作者 S01 审阅的观察不能直接启用（无改进意向）：
    with pytest.raises(Muse错误):
        env["service"].请求启停(
            env["author"],
            "missing-review",
            改进定位(str(saved["observation_id"]), 1, saved["content_hash"]),
            "enable",
        )


def test_发布输入导入形成非空结果报告__0c293f(章后环境):
    """旧 import 产生非空报告：新版规则发布输入导入返回明确结果与数量。"""
    app, author = 章后环境["装配"], 章后环境["作者"]
    review = app.要求审校()
    结果 = review.导入规则候选(author)
    assert 结果["persisted"] is True
    assert len(结果["results"]) >= 1
    assert len({r["version_id"] for r in 结果["results"]}) == len(结果["results"])
    assert len(结果["fingerprint"]) == 64
    读回 = review.读取规则库(author, 预览候选=True)
    assert 读回["rules"] and 读回["fingerprint"] == 结果["fingerprint"]


def test_观察不自动合并改写技能提示资源__bdeab3(运行观察环境):
    """旧 lesson 自动合并进 Skill 文件：新版无该公开写入，须 owner 明确发布。

    该能力（将 lesson 写入目标 Skill 提示资源）未在新版接线；此处以最小红例
    固定「运行观察不产生任何 skill/lesson 落盘」的合同，防止旧自动合并悄悄回来。
    """
    env = 运行观察环境
    saved = env["service"].记录运行观察(
        env["author"],
        "no-skill-merge",
        _观察(
            _完成(env)[2],
            phenomenon="观察不自动合并技能提示",
            quote="技能正文只由 owner 发布。",
            background="成功运行形成的观察不会改写 Skill/提示资源。",
        ),
    )
    assert saved["state"] == "observation"
    # 公开入口不存在 skill 自动合并写入；列表/读取都只回观察本人数据。
    assert str(env["service"].列出运行观察(env["author"])[0]["observation_id"]) == str(
        saved["observation_id"]
    )