mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-15 15:12:00 +00:00
ThearchyHelios
65d6a1e63c
第 2 轮的按值自查和面板那套宁可多遮的认法带进了误判，正式运行因此失败，代价远大于收益，这一轮回到规格原意： - 报告自查去掉按值在正文里找遮罩值（engine/evidence.py 的 masked_values、report.py 的收值）：两个字的姓名会撞上「销售 高峰」这类常用词，同名列的值会撞上门店名；照抄成正文的号码本来就会被「数字没有出处」拦下。时间线日志 report_masked_cell 不再带 written，decode.ts、terms.ts、证据审计的 CSV 回到「引用了遮罩列」一种说法，check-decode 同步 - 自查认引用时从 _mask_sets 的 sure 起，再按 SQL 血缘（sqlglot.lineage，按快照里冻结的表结构）核一遍值取自哪张表的哪一列 （api/evidence.py 的 _cited_masks / _owned_columns）：遮罩列不带表名，同名的列在几张表里都有、其中有的表另有别的遮罩列时 （customers 有 name 和 phone，stores 只有 name），只认那几张表，门店名不算客户姓名；值根本不取自遮罩列的（region AS phone）也不算。认不出出处的照算，没探查过表结构的照列名认。违规说明点名它取自的那一列遮罩列 - 交给裁判的摘录退回 M2 之前的做法，只按数据源设的列名遮（删掉 _judge_masks）：并进面板那套宁可多遮的列后，整份遮的快照、 只数个数的列从摘录里消失，被引用的金额、人数裁判看不到。按需裁判和节点里当场裁判都改回 - 证据面板、推演收「这次运行查到过的遮罩值」只收 sure 的列（_snapshot_secrets，同 masking 收值的口径）：SQL 认不准、整份 遮的快照只遮显示，里面的区域名、金额不再收成遮罩值 - 按快照工件 id 在进程里记下遮罩分析和收到的值（_SnapshotMemo，键里带数据源现在设的遮罩、在不在、类型，设置改了不拿旧的； 只记读得出来的快照，存取加锁，分析结果冻结、交出去的值另抄一份）：之后点开片段、推演都不再把全部快照读盘、复验、分析， 只读这一步要显示的那一份。30 份 500 行的快照，第一次约 0.7 秒，之后不到 1 毫秒 测试：新增 test_report_masked_owners.py（高峰、门店名写成文字照常出具，照抄号码由数字核对拦下，引用门店名不算、引用客户名算， 同一条查询里两张表的同名列、只设 name、* 展开、WITH、UNION、没有表结构、表值函数）、test_evidence_mask_memo.py（整份遮 的快照不收值、只读一次只分析一次、设置改了重新收、推演共用、再点开片段只读本步快照）；裁判摘录改测被引用的金额和计数 都在摘录里；按值自查的用例改成不按值认的回归
268 lines
14 KiB
Python
268 lines
14 KiB
Python
"""报告自查认得出引用了取自遮罩列的格子：像没有出处的数字一样记违规、交回写作者重写，仍不改的按「违规处置」办。
|
||
|
||
验收发现：写作要求点名引用 [[v:Q2.r0.tel]]（Q2 的 SQL 是 c.phone AS tel，phone 设了遮罩），报告自查判 ok、violations 为空，
|
||
正文写出号码，导出的可验证报告也把它带了出去，从发起到导出没有一步提示。写作目录不遮罩是已知设计（写作者要看得到数据），
|
||
所以拦在自查：
|
||
|
||
- 组装文档时（compose_doc 的 masked：{查询快照工件: 取自遮罩列的列}），引用了这些列的格子记一条 masked_cell，列名对上
|
||
的、按 SQL 认出改了名字的都算（报告节点按 api/evidence.py 的 _cited_masks 算，和证据面板同一个 _mask_sets 起步）
|
||
- 违规里不带那一格的值（不写 text、context）：违规会进 report.checked、重写的日志、出具缺口，这些地方不能再露一次号码
|
||
- 报告节点把它交回写作者重写（不是只标注的那几类）；重写后还在：「违规处置」为失败的判失败，为标注的照常产出，
|
||
时间线上另起一行说清（report_masked_cell），节点产出里写一条出具缺口，出口降档
|
||
- 没设遮罩、引用的是别的列：一切照旧
|
||
|
||
真跑的部分用虚构的 SQLite 商城库（号码是编的），写作者是 MockChatModel 照剧本写。
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import asyncio
|
||
import json
|
||
import sqlite3
|
||
|
||
import pytest
|
||
from langchain_core.messages import AIMessage
|
||
from sqlalchemy import select
|
||
|
||
from app.core import artifact_store
|
||
from app.data.engine import engines
|
||
from app.db.base import SessionLocal
|
||
from app.db.models import DataSource, Run, RunEvent
|
||
from app.engine import review as rv
|
||
from app.engine.evidence import compose_doc, describe_violations, iter_segments
|
||
from app.engine.runner import run_manager
|
||
|
||
# --------------------------------------------------------------------------
|
||
# 组装文档：compose_doc 的 masked
|
||
# --------------------------------------------------------------------------
|
||
|
||
ROWS = [["客户乙", "199-0000-0102", 380.0], ["客户甲", "199-0000-0101", 220.0]]
|
||
SECRET = [r[1] for r in ROWS]
|
||
|
||
|
||
async def catalog_of() -> tuple[dict, str]:
|
||
art = await artifact_store.put_json({"columns": ["customer", "tel", "spend"], "rows": ROWS, "source": "mall",
|
||
"sql": "SELECT c.name AS customer, c.phone AS tel, 1 AS spend FROM customers c",
|
||
"mask_columns": ["phone"]}, kind="query")
|
||
entry = {"alias": "Q1", "kind": "query", "artifact": art, "columns": ["customer", "tel", "spend"], "rows": 2,
|
||
"node_id": "top", "source": "mall"}
|
||
return {"Q1": entry}, art
|
||
|
||
|
||
def masked_of(doc: dict) -> list[dict]:
|
||
return [v for v in doc["violations"] if v["code"] == "masked_cell"]
|
||
|
||
|
||
async def test_a_cited_masked_cell_is_a_violation_without_its_value():
|
||
catalog, art = await catalog_of()
|
||
doc = compose_doc("消费最高的是 [[v:Q1.r0.customer]]，电话 [[v:Q1.r0.tel]]，花了 [[v:Q1.r0.spend]]。", catalog,
|
||
node_id="write", run_id="r1", masked={art: {"phone", "tel"}})
|
||
[v] = masked_of(doc)
|
||
seg = next(s for s in iter_segments(doc) if s.get("ref") == "v:Q1.r0.tel")
|
||
assert (v["ref"], v["column"], v["segment"]) == ("v:Q1.r0.tel", "tel", seg["id"]) and v["unit"] and v["span"]
|
||
assert "遮罩" in v["message"] and "「tel」" in v["message"] and "删掉" in v["for_model"]
|
||
# 违规本身不再写一次号码：它会进事件、日志和出具缺口
|
||
assert not [p for p in SECRET if p in json.dumps(v, ensure_ascii=False)]
|
||
assert "text" not in v and "context" not in v
|
||
assert not [p for p in SECRET if p in describe_violations(masked_of(doc))]
|
||
|
||
|
||
async def test_without_masks_or_for_other_columns_nothing_is_flagged():
|
||
catalog, art = await catalog_of()
|
||
text = "电话 [[v:Q1.r0.tel]]，花了 [[v:Q1.r0.spend]]。"
|
||
assert masked_of(compose_doc(text, catalog, node_id="write", run_id="r1")) == []
|
||
assert masked_of(compose_doc(text, catalog, node_id="write", run_id="r1", masked={})) == []
|
||
assert masked_of(compose_doc("花了 [[v:Q1.r0.spend]]。", catalog, node_id="write", run_id="r1",
|
||
masked={art: {"phone", "tel"}})) == []
|
||
|
||
|
||
async def test_a_whole_table_reports_every_masked_cell():
|
||
"""整表引用展开成每格一个 [[v:]]：遮罩列的每一格各记一条，写作者看得到要去掉哪一列。"""
|
||
catalog, art = await catalog_of()
|
||
doc = compose_doc("前两名：\n\n[[table:Q1]]\n", catalog, node_id="write", run_id="r1", masked={art: {"tel"}})
|
||
assert [v["ref"] for v in masked_of(doc)] == ["v:Q1.r0.tel", "v:Q1.r1.tel"]
|
||
assert all("cols=" in v["for_model"] for v in masked_of(doc))
|
||
|
||
|
||
def test_the_masked_cell_code_is_a_degraded_signal_of_its_own_kind():
|
||
"""复核层认得这条日志：报告照常产出、有缺口，按它自己的 kind 记，不落成笼统的 warn。"""
|
||
assert rv.WARN_CODES["report_masked_cell"] == ("report_masked_cell", rv.DEGRADED)
|
||
warn = {"type": "log", "node_id": "write",
|
||
"data": {"level": "warn", "code": "report_masked_cell", "message": "报告引用了取自遮罩列的格子"}}
|
||
[signal] = rv.scan([warn], {"报告": "首位客户已联系"})
|
||
assert (signal.kind, signal.severity) == ("report_masked_cell", rv.DEGRADED)
|
||
|
||
|
||
# --------------------------------------------------------------------------
|
||
# 真跑：报告节点交回重写、按违规处置办、时间线和出具缺口说清
|
||
# --------------------------------------------------------------------------
|
||
|
||
CUSTOMERS = [(1, "客户甲", "199-0000-0101"), (2, "客户乙", "199-0000-0102"), (3, "客户丙", "199-0000-0103")]
|
||
ORDERS = [(1, 1, 120.5), (2, 2, 300.0), (3, 2, 80.0), (4, 3, 210.25), (5, 1, 99.5)]
|
||
PHONES = [c[2] for c in CUSTOMERS]
|
||
SQL_TOP = ("SELECT c.name AS customer, c.phone AS tel, ROUND(SUM(o.amount), 2) AS spend "
|
||
"FROM orders o JOIN customers c ON c.id = o.customer_id GROUP BY c.id, c.name, c.phone ORDER BY spend DESC")
|
||
CITES_TEL = "消费最高的客户是 [[v:Q1.r0.customer]]，联系电话 [[v:Q1.r0.tel]]，共消费 [[v:Q1.r0.spend]]。"
|
||
CLEAN = "消费最高的客户是 [[v:Q1.r0.customer]]，共消费 [[v:Q1.r0.spend]]。"
|
||
|
||
|
||
@pytest.fixture
|
||
async def engine_up():
|
||
await run_manager.setup()
|
||
yield
|
||
await run_manager.shutdown()
|
||
|
||
|
||
async def set_source(path, options=None):
|
||
async with SessionLocal() as session:
|
||
row = (await session.execute(select(DataSource).where(DataSource.name == "mall"))).scalar_one_or_none()
|
||
if row is None:
|
||
row = DataSource(name="mall", kind="sqlite", database=str(path), readonly=True, enabled=True)
|
||
session.add(row)
|
||
row.database, row.kind, row.enabled, row.readonly = str(path), "sqlite", True, True
|
||
row.options = dict(options or {})
|
||
await session.commit()
|
||
return row.id
|
||
|
||
|
||
@pytest.fixture
|
||
async def mall(tmp_path, engine_up):
|
||
path = tmp_path / "mall.db"
|
||
db = sqlite3.connect(path)
|
||
db.executescript("CREATE TABLE customers (id INTEGER PRIMARY KEY, name TEXT, phone TEXT);"
|
||
"CREATE TABLE orders (id INTEGER PRIMARY KEY, customer_id INTEGER, amount REAL);")
|
||
db.executemany("INSERT INTO customers VALUES (?, ?, ?)", CUSTOMERS)
|
||
db.executemany("INSERT INTO orders VALUES (?, ?, ?)", ORDERS)
|
||
db.commit()
|
||
db.close()
|
||
source_id = await set_source(path, {"mask_columns": ["phone"]})
|
||
await engines.invalidate(source_id)
|
||
yield path
|
||
await engines.invalidate(source_id)
|
||
|
||
|
||
def node(nid, ntype, **config):
|
||
return {"id": nid, "type": ntype, "position": {"x": 0, "y": 0}, "data": {"label": nid, "config": config}}
|
||
|
||
|
||
def graph(*, sql: str = SQL_TOP, contract: dict | None = None, **report) -> dict:
|
||
out = {"fields": [{"name": "报告", "value": "{{ nodes.write.text }}"}]}
|
||
if contract is not None:
|
||
out["contract"] = contract
|
||
nodes = [
|
||
node("start", "input"),
|
||
node("top", "tool", tool="db_query__mall", args={"sql": sql}),
|
||
node("write", "report", instructions="写一句，联系电话写 [[v:Q1.r0.tel]]", **report),
|
||
node("out", "output", **out),
|
||
]
|
||
return {"nodes": nodes, "edges": [{"source": a["id"], "target": b["id"]} for a, b in zip(nodes, nodes[1:])]}
|
||
|
||
|
||
def script(monkeypatch, *replies) -> list[list]:
|
||
"""写作者依次回这几段，用完了一直回最后一段。返回每次调用收到的消息。"""
|
||
from app.providers import mock_model
|
||
|
||
seen: list[list] = []
|
||
|
||
def _decide(self, messages):
|
||
seen.append(list(messages))
|
||
return AIMessage(content=replies[min(len(seen), len(replies)) - 1])
|
||
|
||
monkeypatch.setattr(mock_model.MockChatModel, "_decide", _decide)
|
||
return seen
|
||
|
||
|
||
async def finish(spec: dict) -> Run:
|
||
run_id = (await run_manager.start(graph=spec, input_payload={})).id
|
||
for _ in range(400):
|
||
await asyncio.sleep(0.05)
|
||
async with SessionLocal() as session:
|
||
row = await session.get(Run, run_id)
|
||
if row.status in ("succeeded", "failed") and row.manifest_seq is not None:
|
||
return row
|
||
raise AssertionError(f"运行 {run_id} 没跑完：{row.status}")
|
||
|
||
|
||
async def events(run_id: str, etype: str | None = None) -> list[RunEvent]:
|
||
async with SessionLocal() as session:
|
||
q = select(RunEvent).where(RunEvent.run_id == run_id)
|
||
if etype:
|
||
q = q.where(RunEvent.type == etype)
|
||
return list((await session.execute(q.order_by(RunEvent.seq))).scalars())
|
||
|
||
|
||
def logs(rows: list[RunEvent], code: str) -> list[dict]:
|
||
return [e.data for e in rows if e.type == "log" and e.data.get("code") == code]
|
||
|
||
|
||
def leaks(value) -> list[str]:
|
||
raw = value if isinstance(value, str) else json.dumps(value, ensure_ascii=False, default=str)
|
||
return [p for p in PHONES if p in raw]
|
||
|
||
|
||
async def test_the_writer_is_asked_to_rewrite_and_the_rewrite_passes(monkeypatch, mall):
|
||
"""第一稿引用了 tel（phone 改的名字）：自查记违规、交回重写，重写稿不引用就通过。重写请求和日志里不再写号码。"""
|
||
seen = script(monkeypatch, CITES_TEL, CLEAN)
|
||
row = await finish(graph())
|
||
assert row.status == "succeeded", row.error
|
||
[checked] = await events(row.id, "report.checked")
|
||
assert checked.data["ok"] is True and checked.data["repairs"] == 1, checked.data
|
||
repair = str(seen[1][-1].content)
|
||
assert "tel" in repair and "遮罩" in repair and leaks(repair) == []
|
||
[log] = logs(await events(row.id), "report_repair")
|
||
assert "v:Q1.r0.tel" in log["message"] and leaks(log) == []
|
||
|
||
|
||
async def test_a_writer_that_insists_is_flagged_and_the_issuance_says_why(monkeypatch, mall):
|
||
"""重写后还引用：「违规处置」为标注时照常产出，自查结果、时间线、出具缺口都说清，这几处都不再写一次号码。"""
|
||
script(monkeypatch, CITES_TEL)
|
||
row = await finish(graph(on_violation="flag", contract={"report_from": "write"}))
|
||
assert row.status == "succeeded", row.error
|
||
rows = await events(row.id)
|
||
[checked] = [e.data for e in rows if e.type == "report.checked"]
|
||
assert checked["ok"] is False and checked["repairs"] == 1 and not checked["failed"]
|
||
[v] = [v for v in checked["violations"] if v["code"] == "masked_cell"]
|
||
assert v["ref"] == "v:Q1.r0.tel" and v["column"] == "tel" and leaks(v) == []
|
||
[log] = logs(rows, "report_masked_cell")
|
||
assert log["count"] == 1 and log["refs"] == ["v:Q1.r0.tel"] and log["columns"] == ["tel"] and log["level"] == "warn"
|
||
assert "重写 1 次" in log["message"] and leaks(log) == []
|
||
finished = next(e.data for e in rows if e.type == "node.finished" and e.node_id == "write")
|
||
payload = artifact_store.load(finished["artifact"])
|
||
[gap] = payload["issuance_gaps"]
|
||
assert "遮罩" in gap and "v:Q1.r0.tel" in gap and leaks(gap) == []
|
||
issuance = row.output["_issuance"]
|
||
assert issuance["tier"] == "degraded" and any(g.startswith("报告撰写「write」：") and "遮罩" in g
|
||
for g in issuance["gaps"]), issuance["gaps"]
|
||
assert leaks(issuance["gaps"]) == []
|
||
|
||
|
||
async def test_a_writer_that_insists_fails_the_node_when_violations_fail(monkeypatch, mall):
|
||
"""「违规处置」为失败：重写后还引用就判失败，原因说是遮罩列，不写号码。"""
|
||
script(monkeypatch, CITES_TEL)
|
||
row = await finish(graph(on_violation="fail"))
|
||
assert row.status == "failed" and "遮罩" in (row.error or "") and "tel" in row.error, row.error
|
||
assert leaks(row.error) == []
|
||
[checked] = await events(row.id, "report.checked")
|
||
assert checked.data["failed"] is True
|
||
assert logs(await events(row.id), "report_masked_cell") == [], "节点判失败时不另起「照常产出」的那一行"
|
||
|
||
|
||
async def test_the_masked_column_itself_is_flagged_too(monkeypatch, mall):
|
||
"""列名直接就是遮罩列（没起别名）的也算。"""
|
||
script(monkeypatch, "首位客户 [[v:Q1.r0.name]]，电话 [[v:Q1.r0.phone]]。")
|
||
row = await finish(graph(sql="SELECT name, phone FROM customers ORDER BY id", on_violation="flag"))
|
||
[checked] = await events(row.id, "report.checked")
|
||
assert [v["ref"] for v in checked.data["violations"] if v["code"] == "masked_cell"] == ["v:Q1.r0.phone"]
|
||
|
||
|
||
async def test_without_masks_the_report_is_unchanged(monkeypatch, mall):
|
||
await set_source(mall, {})
|
||
script(monkeypatch, CITES_TEL)
|
||
row = await finish(graph(on_violation="flag", contract={"report_from": "write"}))
|
||
assert row.status == "succeeded", row.error
|
||
[checked] = await events(row.id, "report.checked")
|
||
assert checked.data["ok"] is True and checked.data["repairs"] == 0
|
||
assert logs(await events(row.id), "report_masked_cell") == []
|
||
finished = next(e.data for e in await events(row.id, "node.finished") if e.node_id == "write")
|
||
assert "issuance_gaps" not in artifact_store.load(finished["artifact"])
|
||
assert row.output["_issuance"]["tier"] == "formal" or not any("遮罩" in g for g in row.output["_issuance"]["gaps"])
|