Files
AgentLab/backend/tests/test_metrics_issuance.py
T
ThearchyHelios 941723b17c 端到端验收修复与指标出具:
- 身份与工具:登录页建第一个管理员(认人接口的 401 说明能不能建);搭建者有只读的人和组名录,选人给候选并核对名字;授权三行带标题;端点请求头名校验(空名、非法名存不进、发不出时说清原因);写入能力在工具页试运行一律演练、不发送,HTTP 状态照实说成败;写记忆不算只读、查询能力只能选同一端点上的只读能力;审批与遮罩的说法和实际规则一致;开登录时不要署名、导入和发布记登录的人;使用者不显示推演等只给搭建者的操作,403 按是否开登录说话;登录页只认一次人
- 流程与实例:图里用到的岗位(派给谁、超时升级给谁、发送确认派给谁)全部派生、没绑人的发布时报错;写外部系统的参数按能力声明的类型收拢后再算哈希和发送,收不成就不发;发起人的实例页只看这一步的状态,处理人才给待办卡;会签进度带登录核实和时间,升级后照实写规则;实例状态写流程的结论(已退回、未确认、等待核实写入结果),核实生效后不再挂失败原因;流程应用首页不派生最近封存运行的结果块,看板没数据时是安静的空态;样例⑩的金额带类型和下限、待办标题带金额,表单上下限前后端一起生效
- 看板与证据:定时任务模板加 ISO 周、上一 ISO 周、本月、上月;出处链穿过原样解析查询结果的整形节点,追不到原始数据的照实说停在哪里,样例⑨改用 cell();应用首页和版本历史都能做成应用;指标块单位不重复;版本列表带退役说明;定时任务页头跟着刷新;删视图要确认、新视图能起名;编辑器属性用中文名、术语带解释
- 指标出具(用户 2026-10-15 拍板):契约只写「指标来自」、上游没有报告撰写节点、成果里没有模型写的话时按指标出具——必需指标齐、指标输入追到原始数据、成果字段里的数回指得上才完整出具;受管级要求每个指标静态地追到原始数据(governed.metric_untraced);出具横幅和应用印章写明按指标
- 受管工作流降为已发布只给管理员(Q8 顺延,服务端同样拦);工作流接口带当前发布版的级别,画布上改过之后也认得出受管工作流
2026-10-15 07:15:37 +08:00

171 lines
8.2 KiB
Python

"""指标出具(engine/issuance.metrics_only,用户 2026-10-15 拍板):纯指标看板——契约只写「指标来自」,没有「报告来自」
也没有「叙述」——出具的就是指标本身。三档的语义不变,核对的对象从叙述换成指标:
- 必需指标齐、每个指标的输入追到原始数据(查询快照的格、运行输入)、成果字段里给人看的数都回指得上:完整出具
- 指标的输入停在别的节点(表达式整形、带过滤器的解析……):降档,缺口写明哪个指标停在哪个节点
- 成果字段里有回指不上的数:降档(strict 下不予出具);必需指标缺了:不予出具
- 受管门禁:只写「指标来自」不再报「既没有报告来自、也没有叙述」;每个指标都要静态地追到原始数据,追不到的挡住发布
库和数字都是编的。
"""
from __future__ import annotations
import asyncio
import sqlite3
import uuid
import pytest
from app.data.engine import engines
from app.db.base import SessionLocal
from app.db.models import DataSource, Run
from app.engine.governance import lint_for_publish
from app.engine.issuance import metrics_only
from app.engine.runner import run_manager
from app.engine.schema import GraphSpec
SQL = "SELECT SUM(amount) AS gmv, COUNT(*) AS orders FROM orders"
def node(nid, ntype, label=None, **config):
return {"id": nid, "type": ntype, "position": {"x": 0, "y": 0}, "data": {"label": label or nid, "config": config}}
def chain(*nodes):
return {"nodes": list(nodes), "edges": [{"source": a["id"], "target": b["id"]} for a, b in zip(nodes, nodes[1:])]}
def board(source: str, *, gmv: str = "vars.q.rows[0][0]", fields=None, required=("gmv", "orders"), strict=False):
return chain(
node("start", "input", "开始", fields=[{"name": "week", "default": "2026-W40"}]),
node("fetch", "tool", "取数", tool=f"db_query__{source}", args={"sql": SQL}),
node("parse", "transform", "解析查询结果", mode="json", template="{{ nodes.fetch }}", assign_to="q"),
node("raw_n", "transform", "整理查询结果", mode="json", template="{{ nodes.fetch | trim }}", assign_to="raw"),
node("card", "metrics", "周报口径卡", caliber="周报口径", caliber_version="v1", metrics=[
{"id": "gmv", "name": "销售额", "unit": "元", "decimals": 2, "expression": gmv},
{"id": "orders", "name": "订单数", "expression": "cell(nodes.fetch, 0, 'orders')"},
]),
node("out", "output", "成果", fields=fields if fields is not None else [
{"name": "本期", "value": "{{ input.week }}"}, {"name": "指标", "value": "{{ nodes.card.text }}"}],
contract={"metrics_from": ["card"], "required": list(required), "strict": strict}),
)
@pytest.mark.parametrize("contract, want", [
({"metrics_from": ["card"], "required": ["gmv"]}, True),
({"metrics_from": "card"}, True),
({"metrics_from": ["card"], "narrative": "销售额 {{ nodes.card.text }}"}, False),
({"metrics_from": ["card"], "report_from": "write"}, False),
({"required": ["gmv"]}, False),
("card", False),
])
def test_metrics_only(contract, want):
assert metrics_only(contract) is want
# --------------------------------------------------------------------------
# 运行时
# --------------------------------------------------------------------------
@pytest.fixture(autouse=True)
async def engine_up():
await run_manager.setup()
yield
await run_manager.shutdown()
@pytest.fixture
async def shop(tmp_path):
path = tmp_path / "shop.db"
with sqlite3.connect(path) as db:
db.executescript("CREATE TABLE orders (id INTEGER PRIMARY KEY, amount REAL);")
db.executemany("INSERT INTO orders VALUES (?, ?)", [(1, 100.5), (2, 200.0), (3, 300.0)])
name = f"mx_{uuid.uuid4().hex[:6]}"
async with SessionLocal() as session:
row = DataSource(name=name, kind="sqlite", database=str(path), readonly=True, enabled=True)
session.add(row)
await session.commit()
source_id = row.id
yield name
await engines.invalidate(source_id)
async def issuance_of(graph: dict) -> dict:
run = await run_manager.start(graph=graph, input_payload={"week": "2026-W40"})
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:
assert row.status == "succeeded", row.error
return (row.output or {})["_issuance"]
raise AssertionError(f"运行 {run.id} 没跑完:{row.status}")
async def test_a_board_whose_metrics_reach_the_data_is_formal(shop):
issuance = await issuance_of(board(shop))
assert (issuance["tier"], issuance["mode"]) == ("formal", "metrics")
# 不再记「出具契约未设置『叙述』」:没有叙述不是缺口,核的是指标本身
assert issuance["gaps"] == [] and issuance["unmatched_numbers"] == []
# 成果字段里的数(口径卡的文字)都回指到了指标
assert issuance["matched_numbers"] >= 2
async def test_an_input_that_stops_short_of_the_data_degrades(shop):
issuance = await issuance_of(board(shop, gmv="vars.raw.rows[0][0]"))
assert issuance["tier"] == "degraded"
[gap] = issuance["gaps"]
assert "销售额" in gap and "没追到原始数据" in gap and "整理查询结果" in gap
async def test_numbers_in_the_fields_must_trace_and_required_metrics_must_be_there(shop):
loose = await issuance_of(board(shop, fields=[{"name": "备注", "value": "本周目标 12345 元"}]))
assert loose["tier"] == "degraded" and [u["token"] for u in loose["unmatched_numbers"]] == ["12345"]
strict = await issuance_of(board(shop, fields=[{"name": "备注", "value": "本周目标 12345 元"}], strict=True))
assert strict["tier"] == "withheld"
missing = await issuance_of(board(shop, required=("gmv", "orders", "refund_rate")))
assert (missing["tier"], missing["missing_required"]) == ("withheld", ["refund_rate"])
# --------------------------------------------------------------------------
# 受管门禁
# --------------------------------------------------------------------------
def gate(graph: dict, level: str = "governed") -> dict[str, str]:
return {i.code: i.level for i in lint_for_publish(GraphSpec.model_validate(graph), level=level).issues if i.code}
def test_a_metrics_only_contract_is_not_missing_its_checks():
codes = gate(board("shop"))
assert "contract.report_from_missing" not in codes and "governed.report_from_required" not in codes
assert "governed.metric_untraced" not in codes
def test_governed_metrics_must_reach_the_data_statically():
codes = gate(board("shop", gmv="vars.raw.rows[0][0]"))
assert codes.get("governed.metric_untraced") == "error"
# 已发布级不按这一条挡:运行时照样降档、照实写缺口
assert "governed.metric_untraced" not in gate(board("shop", gmv="vars.raw.rows[0][0]"), level="published")
def test_a_contract_that_checks_nothing_is_still_refused():
graph = board("shop")
graph["nodes"][-1]["data"]["config"]["contract"] = {"required": ["gmv"]}
assert gate(graph).get("contract.report_from_missing") == "error"
def test_a_board_with_a_report_or_model_text_is_not_metrics_only():
"""上游有报告撰写节点、或成果字段里有模型写的话:那是叙述,不能因为契约少写了「报告来自」就改成只核指标。"""
with_report = board("shop")
with_report["nodes"].insert(-1, node("write", "report", "写周报", metrics_from=["card"]))
with_report["edges"] = [*with_report["edges"][:-1], {"source": "card", "target": "write"},
{"source": "write", "target": "out"}]
assert gate(with_report).get("contract.report_from_missing") == "error"
with_model = board("shop", fields=[{"name": "点评", "value": "{{ nodes.say.text }}"}])
with_model["nodes"].insert(-1, node("say", "llm", "点评", prompt="点评一下"))
with_model["edges"] = [*with_model["edges"][:-1], {"source": "card", "target": "say"},
{"source": "say", "target": "out"}]
assert gate(with_model, level="published").get("contract.report_from_missing") == "warning"