mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-22 20:01:59 +00:00
证据接入按 W1-a、W1-c 的实际接口对齐:复算就是真的 app.analysis.recompute——load 交查询快照工件的原内容、load_result 交上一层结论(只认封存范围内交回过的查询、台账记过的结论,取不回照实说取不回哪一份),方法升没升版按结论里的 method 和现在的 contract.METHODS 比(升了版、复算成立的写结果同不同,现在的方法复算不了的写明原因),去掉为算子没落地准备的那一套;关键数的代入式按真的具名的数写(乘积拆解补上,波动基线按变化率比的按百分比写,负数加括号)。口径卡指标上的可比性结论(不进台账、只从卡上引用)点得开:证据面板的指标步骤带它整条出处链(判定 → 方法 → 计算 → 输入 → 复算 → 封存,封存按卡上记的 sha256 认),报告导出收在一个可展开的块里,可验证报告随附带的卡一起附上、④ 按已核对的卡认它、数字详情里展示各项检查和服务端复算;裁判的指标摘录、写作目录照卡上记的判定语写。现查结果分清按运行当天现算的界、一列一个值都没有、表的范围;可验证报告里分析工具的 tool.end、分析卡的 analysis.end 和结论同一个认法才带内容,输入的角色写叫法。测试改用真算的结论(run 交的输入照分析入口的写法,复算交工件原内容),下钻、总数闭合、方法升版、可比性端到端都过真的复算
This commit is contained in:
@@ -1,13 +1,14 @@
|
||||
"""分析结论进证据链(第 1 波 W1-b):证据接口、审计、报告导出。
|
||||
|
||||
运行是直接落库的虚构封存运行(虚构零售库 shop):一次查询(Q1)、一份按地区拆的分析结论(A1,分析卡或 Agent 做的),
|
||||
报告引用判定语、具名标量、结果表的格。算子第 1 波还没落地,复算按 monkeypatch 换。
|
||||
报告引用判定语、具名标量、结果表的格。结论用 app.analysis 真算(tests.test_analysis_evidence.analyzed),出处链里的复算是
|
||||
真的 recompute:按 sha256 取回封存的输入快照、下钻的上一层结论,逐字段比。只有要它出错、升版的几处用 monkeypatch。
|
||||
|
||||
- 证据图:A1 条目带算子、判定;「派生自」的边 A1 → Q1
|
||||
- 片段接口:判定 → 方法 → 计算 → 输入(接查询步骤)→ 复算 → 封存;Agent 写的假设标出来;复算不了照实写,不让接口出错
|
||||
- 审计:分析判定、分析结论的叫法;分析说法核对的违规单列一行
|
||||
- 报告导出:数据块里有分析结论那几步;文档带「分析范围」的,正文后面渲染、编号点得开
|
||||
- 指标步骤带可比性
|
||||
- 指标步骤带可比性:口径卡指标上的可比性结论(不进台账)照卡上记的 sha256 取回,点得开(判定 → … → 复算 → 封存)
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
@@ -16,21 +17,22 @@ import io
|
||||
import json
|
||||
import re
|
||||
import uuid
|
||||
from collections.abc import Sequence
|
||||
from typing import Any
|
||||
|
||||
import pytest
|
||||
from httpx import ASGITransport, AsyncClient
|
||||
|
||||
from app import analysis
|
||||
from app.analysis import contract
|
||||
from app.analysis.contract import ARTIFACT_KIND, ledger_entry
|
||||
from app.analysis.testing import fake_result
|
||||
from app.core import artifact_store
|
||||
from app.core.artifact_store import manifest_hash
|
||||
from app.db.base import SessionLocal
|
||||
from app.db.models import Run, RunEvent
|
||||
from app.engine.evidence import build_catalog, compose_doc, iter_segments
|
||||
from app.main import app
|
||||
from tests.test_analysis_evidence import SNAPSHOT
|
||||
from tests.test_analysis_evidence import CHANNELS, SNAPSHOT, analyzed
|
||||
|
||||
GRAPH = {"nodes": [{"id": n, "type": t, "data": {"label": label}} for n, t, label in [
|
||||
("fetch", "tool", "按地区取数"), ("an", "agent", "分析"), ("card", "metrics", "周报口径"),
|
||||
@@ -43,28 +45,38 @@ async def client():
|
||||
yield c
|
||||
|
||||
|
||||
def _query_entry(content: dict[str, Any], artifact: str) -> dict[str, Any]:
|
||||
return {"kind": "query", "node_id": "fetch", "exec": 1, "artifact": artifact, "tool": "db_query__shop",
|
||||
"source": "shop", "columns": content["columns"], "rows": len(content["rows"]), "truncated": False}
|
||||
|
||||
|
||||
async def sealed_run(text: str, *, item: str | None = "drop", call_id: str | None = None,
|
||||
result: dict[str, Any] | None = None, doc_extra: dict[str, Any] | None = None,
|
||||
card: dict[str, Any] | None = None) -> dict[str, Any]:
|
||||
"""落一次封存好的运行:查询快照、分析结论、(可选)口径卡、报告文档都按真实的工件写,事件按 runner 的算法封存。"""
|
||||
card: dict[str, Any] | None = None, queries: Sequence[dict[str, Any]] = (),
|
||||
parents: Sequence[dict[str, Any]] = ()) -> dict[str, Any]:
|
||||
"""落一次封存好的运行:查询快照(SNAPSHOT,再加 queries)、分析结论(parents 在前,result 在后;result 缺省按 SNAPSHOT
|
||||
真算一份按地区拆的可加拆解)、(可选)口径卡、报告文档都按真实的工件写,事件按 runner 的算法封存。"""
|
||||
run_id = uuid.uuid4().hex
|
||||
q = await artifact_store.put_json(SNAPSHOT, kind="query_snapshot", run_id=run_id, node_id="fetch")
|
||||
res = result or fake_result(inputs=[{"ref": "Q2", "artifact": q, "role": "data",
|
||||
"columns": ["region", "week", "gmv"], "rows": "all"}])
|
||||
more = [await artifact_store.put_json(c, kind="query_snapshot", run_id=run_id, node_id="fetch") for c in queries]
|
||||
res = result or await analyzed({"Q2": (SNAPSHOT, q)})
|
||||
qentries = [_query_entry(SNAPSHOT, q), *(_query_entry(c, a) for c, a in zip(queries, more))]
|
||||
aentries = []
|
||||
for i, parent in enumerate(parents):
|
||||
found = await artifact_store.put_json(parent, kind=ARTIFACT_KIND, run_id=run_id, node_id="an")
|
||||
aentries.append(ledger_entry(parent, node_id="an", exec_no=1, artifact=found, item=f"up{i}"))
|
||||
a = await artifact_store.put_json(res, kind=ARTIFACT_KIND, run_id=run_id, node_id="an")
|
||||
qentry = {"kind": "query", "node_id": "fetch", "exec": 1, "artifact": q, "tool": "db_query__shop",
|
||||
"source": "shop", "columns": SNAPSHOT["columns"], "rows": len(SNAPSHOT["rows"]), "truncated": False}
|
||||
aentry = ledger_entry(res, node_id="an", exec_no=1, artifact=a, item=item, call_id=call_id)
|
||||
aentries.append(ledger_entry(res, node_id="an", exec_no=1, artifact=a, item=item, call_id=call_id))
|
||||
nodes: dict[str, Any] = {}
|
||||
events: list[tuple[int, str, str | None, dict[str, Any]]] = [
|
||||
(1, "run.started", None, {"input": {}}), (2, "node.finished", "fetch", {"evidence": [qentry]}),
|
||||
(3, "node.finished", "an", {"evidence": [aentry]})]
|
||||
(1, "run.started", None, {"input": {}}), (2, "node.finished", "fetch", {"evidence": qentries}),
|
||||
(3, "node.finished", "an", {"evidence": aentries})]
|
||||
if card is not None:
|
||||
card_id = await artifact_store.put_json(card, kind="metric_set", run_id=run_id, node_id="card")
|
||||
nodes["card"] = {**card, "artifact": card_id}
|
||||
events.append((4, "node.finished", "card", {"evidence": [{"kind": "metric_set", "node_id": "card",
|
||||
"exec": 1, "artifact": card_id}]}))
|
||||
catalog = build_catalog(nodes=nodes, ledger=[qentry, aentry])
|
||||
catalog = build_catalog(nodes=nodes, ledger=[*qentries, *aentries])
|
||||
doc = compose_doc(text, catalog, node_id="write", run_id=run_id)
|
||||
doc.update(doc_extra or {})
|
||||
doc_id = await artifact_store.put_json(doc, kind="report_doc", run_id=run_id, node_id="write")
|
||||
@@ -78,7 +90,7 @@ async def sealed_run(text: str, *, item: str | None = "drop", call_id: str | Non
|
||||
for s, etype, node, data in events:
|
||||
session.add(RunEvent(run_id=run_id, seq=s, type=etype, node_id=node, ts=float(s), data=data))
|
||||
await session.commit()
|
||||
return {"run": run_id, "doc": doc, "query": q, "analysis": a, "result": res}
|
||||
return {"run": run_id, "doc": doc, "query": q, "analysis": a, "result": res, "queries": more}
|
||||
|
||||
|
||||
def seg_id(doc: dict[str, Any], ref: str) -> str:
|
||||
@@ -118,11 +130,13 @@ async def test_the_phrase_opens_judgment_method_calculation_inputs_recompute_and
|
||||
assert [s["step"] for s in chain] == ["analysis", "analysis_method", "analysis_calc", "analysis_input", "query",
|
||||
"analysis_recompute", "analysis_seal"]
|
||||
judged, method, calc, given, query, again, seal = chain
|
||||
assert judged["phrase"] == made["result"]["phrase"] and judged["criterion"] == "一段占总变化一半以上算主要来源"
|
||||
assert judged["phrase"] == made["result"]["phrase"] == "销售额下降 194(-8.7%),主要来自华东(-120,占 61.9%)"
|
||||
assert judged["criterion"].startswith("一段占总变化的一半以上算主要来源")
|
||||
assert judged["verdict"] == {"code": "driver", "subjects": [["华东"]], "direction": "down"}
|
||||
assert judged["cited"] == {"what": "phrase"} and judged["render_ok"] is True and judged["eid_ok"] is True
|
||||
assert judged["origin"] == "card" and judged["item"] == "drop" and judged["label"] == "分析"
|
||||
assert judged["hypothesis"] == "地区拖累" and judged["hypothesis_note"].startswith("Agent 写的假设")
|
||||
assert judged["hypothesis"] == "华东拖累" and judged["hypothesis_note"].startswith("Agent 写的假设")
|
||||
assert {c["code"] for c in judged["caveats"]} == {"measure_unlabeled", "headline_basis"}
|
||||
assert method["method"] == "decompose.additive@1" and method["method_label"] == "变化拆解(可加)"
|
||||
assert {s["key"]: s["label"] for s in method["settings"]}["main_share"] == "「主要来源」的占比"
|
||||
assert {s["key"] for s in method["settings"]} >= {"main_share", "concentration", "table_top_k"}
|
||||
@@ -130,28 +144,30 @@ async def test_the_phrase_opens_judgment_method_calculation_inputs_recompute_and
|
||||
assert calc["table"]["columns"][0] == "region" and calc["table"]["labels"]["share"] == "占变化的比重"
|
||||
assert calc["table"]["total_rows"] == 4 and calc["table"]["rows"][0][:2] == ["华东", "820"]
|
||||
assert {f["label"]: f["text"] for f in calc["formulas"]}["占比"] == \
|
||||
"华东的占比 = 华东的变化 ÷ 总变化 = -120 ÷ -194 = 61.9%"
|
||||
"华东的占比 = 华东的变化 ÷ 总变化 = -120 ÷ (-194) = 61.9%"
|
||||
assert given == {"step": "analysis_input", "of": "A1", "ref": "Q2", "role": "data", "role_label": "数据",
|
||||
"artifact": made["query"], "columns": ["region", "week", "gmv"], "rows": "all", "query": "Q1",
|
||||
"node_id": "fetch", "sealed": True}
|
||||
assert query["alias"] == "Q1" and query["input_of"] == "A1" and query["total_rows"] == 8
|
||||
assert query["highlight"]["cols"] == ["region", "week", "gmv"]
|
||||
assert again["status"] == "failed" and again["text"] == "复算失败:系统里还没有这种分析的复算实现"
|
||||
# 真的复算:按 sha256 取回封存的输入快照,逐字段和封存的比
|
||||
assert again == {"step": "analysis_recompute", "alias": "A1", "status": "same", "method": "decompose.additive@1",
|
||||
"text": "复算一致:按封存的输入重新计算,结论逐项相同"}
|
||||
assert seal["sealed"] is True and seal["ledger"] is True and seal["hash_ok"] is True
|
||||
assert body["seal"]["covered"] is True
|
||||
assert body["note"].startswith("判定语由系统按平台的方法算出判定")
|
||||
assert body["segment"]["kind"] == "analysis"
|
||||
|
||||
|
||||
async def test_named_values_and_result_cells_point_at_what_they_cite(client, monkeypatch):
|
||||
monkeypatch.setattr(analysis, "recompute", lambda result, load, *, load_result=None: dict(result))
|
||||
async def test_named_values_and_result_cells_point_at_what_they_cite(client):
|
||||
made = await sealed_run(TEXT)
|
||||
value = (await client.get(
|
||||
f"/api/runs/{made['run']}/evidence/segments/{seg_id(made['doc'], 'a:A1.top_share|pct')}")).json()
|
||||
assert value["chain"][0]["cited"] == {"what": "value", "key": "top_share", "raw": "0.6186", "conv": "pct"}
|
||||
assert value["chain"][0]["cited"] == {"what": "value", "key": "top_share", "raw": "0.618556701031", "conv": "pct"}
|
||||
assert value["segment"]["text"] == "61.9%" and value["chain"][0]["render_ok"] is True
|
||||
calc = value["chain"][2]
|
||||
assert next(v for v in calc["values"] if v["key"] == "top_share") == {
|
||||
"key": "top_share", "raw": "0.6186", "rendered": "0.6186", "cited": True}
|
||||
"key": "top_share", "raw": "0.618556701031", "rendered": "0.618556701", "cited": True}
|
||||
assert value["chain"][-2]["text"] == "复算一致:按封存的输入重新计算,结论逐项相同"
|
||||
assert value["note"] == "这个数取自分析结论的具名的数,由系统按固定规则显示,未经模型转写"
|
||||
|
||||
@@ -163,20 +179,41 @@ async def test_named_values_and_result_cells_point_at_what_they_cite(client, mon
|
||||
|
||||
|
||||
async def test_the_recompute_gets_the_sealed_inputs_by_hash(client, monkeypatch):
|
||||
"""复算的 load 交工件原内容(不加出处、不改列的类型),只认封存范围内交回过的查询;别的一律当取不回。"""
|
||||
seen: dict[str, Any] = {}
|
||||
real = analysis.recompute
|
||||
|
||||
def recompute(result, load, *, load_result=None):
|
||||
seen["input"] = load(result["inputs"][0]["artifact"])
|
||||
seen["missing"] = load("0" * 64)
|
||||
return {**result, "method": "decompose.additive@2"}
|
||||
return real(result, load, load_result=load_result)
|
||||
|
||||
monkeypatch.setattr(analysis, "recompute", recompute)
|
||||
made = await sealed_run(TEXT)
|
||||
body = (await client.get(f"/api/runs/{made['run']}/evidence/segments/{seg_id(made['doc'], 'a:A1')}")).json()
|
||||
again = next(s for s in body["chain"] if s["step"] == "analysis_recompute")
|
||||
assert again["text"] == "运行时用的是 decompose.additive@1,现在是 decompose.additive@2,复算结果不同"
|
||||
assert seen["input"]["rows"] == SNAPSHOT["rows"] and seen["input"]["node_id"] == "fetch"
|
||||
assert seen["missing"] is None
|
||||
assert again["status"] == "same" and seen["input"] == SNAPSHOT and seen["missing"] is None
|
||||
|
||||
|
||||
async def test_an_input_outside_the_sealed_scope_is_not_recomputed_from(client):
|
||||
"""结论记着的输入快照不是这次运行交回过的(台账里没有):复算照实说取不回,不去工件库里捞。"""
|
||||
stray = {**SNAPSHOT, "sql": SNAPSHOT["sql"] + " -- 另一次"}
|
||||
s_id = await artifact_store.put_json(stray, kind="query_snapshot")
|
||||
made = await sealed_run(TEXT, result=await analyzed({"Q2": (stray, s_id)}))
|
||||
body = (await client.get(f"/api/runs/{made['run']}/evidence/segments/{seg_id(made['doc'], 'a:A1')}")).json()
|
||||
again = next(s for s in body["chain"] if s["step"] == "analysis_recompute")
|
||||
assert again["status"] == "failed" and again["text"] == "复算失败:取不回 Q2 的查询快照"
|
||||
given = next(s for s in body["chain"] if s["step"] == "analysis_input")
|
||||
assert given["sealed"] is False and given["query"] is None
|
||||
|
||||
|
||||
async def test_an_upgraded_method_names_both_versions(client, monkeypatch):
|
||||
made = await sealed_run(TEXT)
|
||||
monkeypatch.setitem(contract.METHODS, ("decompose", "additive"), "decompose.additive@2")
|
||||
body = (await client.get(f"/api/runs/{made['run']}/evidence/segments/{seg_id(made['doc'], 'a:A1')}")).json()
|
||||
again = next(s for s in body["chain"] if s["step"] == "analysis_recompute")
|
||||
assert again["method"] == "decompose.additive@1" and again["method_now"] == "decompose.additive@2"
|
||||
assert again["text"] == "运行时用的是 decompose.additive@1,现在是 decompose.additive@2,复算结果相同"
|
||||
|
||||
|
||||
async def test_recompute_errors_do_not_break_the_panel(client, monkeypatch):
|
||||
@@ -194,7 +231,8 @@ async def test_recompute_errors_do_not_break_the_panel(client, monkeypatch):
|
||||
async def test_a_tampered_result_is_said_and_not_shown(client):
|
||||
# 工件按内容寻址:改的是这一次独有的那份结论(别的测试的结论内容相同、是同一个文件)
|
||||
q = await artifact_store.put_json(SNAPSHOT, kind="query_snapshot")
|
||||
own = fake_result(inputs=[{"ref": "Q2", "artifact": q, "role": "data"}], criterion=f"只给这一次:{uuid.uuid4().hex}")
|
||||
own = await analyzed({"Q2": (SNAPSHOT, q)}, hypothesis=f"只给这一次:{uuid.uuid4().hex}")
|
||||
assert own["hypothesis"].startswith("只给这一次:")
|
||||
made = await sealed_run(TEXT, result=own)
|
||||
path = artifact_store._path_of(made["analysis"])
|
||||
path.write_text(path.read_text("utf-8").replace("华东", "华南"), "utf-8")
|
||||
@@ -210,27 +248,97 @@ async def test_agent_analyses_are_marked_as_such(client):
|
||||
assert body["chain"][0]["origin"] == "agent" and body["chain"][0]["call_id"] == "call-9"
|
||||
|
||||
|
||||
async def test_headline_inputs_show_the_closure_and_drill_downs_the_parent(client):
|
||||
base = fake_result()
|
||||
made = await sealed_run(TEXT)
|
||||
q, a = made["query"], made["analysis"]
|
||||
child = fake_result(inputs=[{"ref": "Q2", "artifact": q, "role": "data", "columns": ["region", "gmv"],
|
||||
"rows": [0, 1]},
|
||||
{"ref": "Q2.r1.gmv", "artifact": q, "role": "headline_target", "row": 1,
|
||||
"column": "gmv", "value": "2026"},
|
||||
{"ref": "A3", "artifact": a, "role": "parent"}],
|
||||
values={**base["values"], "target_total": "2026"})
|
||||
drilled = await sealed_run("[[a:A1]]。", result=child)
|
||||
body = (await client.get(f"/api/runs/{drilled['run']}/evidence/segments/{seg_id(drilled['doc'], 'a:A1')}")).json()
|
||||
async def test_headline_inputs_show_the_closure(client):
|
||||
"""给了总数(headline)的拆解:两格各是一份输入,和结论里的总数闭合;复算照样按 sha256 取回两份快照。"""
|
||||
totals = {"columns": ["gmv_base", "gmv"], "column_types": {"gmv_base": "number", "gmv": "number"},
|
||||
"rows": [[2220, 2026]], "sql": "SELECT 2220 AS gmv_base, 2026 AS gmv", "source": "shop",
|
||||
"truncated": False}
|
||||
t = artifact_store.content_hash(artifact_store.canonical_json(totals))
|
||||
q = artifact_store.content_hash(artifact_store.canonical_json(SNAPSHOT))
|
||||
result = await analyzed({"Q2": (SNAPSHOT, q), "Q5": (totals, t)}, headline=["Q5.r0.gmv_base", "Q5.r0.gmv"])
|
||||
made = await sealed_run("[[a:A1]]。", result=result, queries=[totals])
|
||||
body = (await client.get(f"/api/runs/{made['run']}/evidence/segments/{seg_id(made['doc'], 'a:A1')}")).json()
|
||||
inputs = [s for s in body["chain"] if s["step"] == "analysis_input"]
|
||||
assert [s["role"] for s in inputs] == ["data", "headline_target", "parent"]
|
||||
assert [s["role"] for s in inputs] == ["headline_base", "headline_target", "data"]
|
||||
assert inputs[0]["closure"] == {"total_key": "base_total", "total": "2,220", "headline": "2,220", "closed": True}
|
||||
assert inputs[1]["closure"] == {"total_key": "target_total", "total": "2,026", "headline": "2,026", "closed": True}
|
||||
assert inputs[2]["phrase"] == base["phrase"] and inputs[2]["role_label"] == "上一层拆解"
|
||||
query = next(s for s in body["chain"] if s["step"] == "query")
|
||||
assert [0, "region"] in query["highlight"]["cells"] and [1, "gmv"] in query["highlight"]["cells"]
|
||||
assert inputs[1]["query"] == "Q2" and inputs[1]["sealed"] is True and inputs[2]["query"] == "Q1"
|
||||
assert next(s for s in body["chain"] if s["step"] == "analysis_recompute")["status"] == "same"
|
||||
|
||||
|
||||
async def test_the_metric_step_shows_the_comparability(client):
|
||||
async def test_a_drill_down_points_at_its_parent_and_recomputes_from_it(client):
|
||||
"""下钻(华东按渠道再拆):输入里有上一层结论(parent),落到报告目录的 A1;复算按 sha256 取回台账记过的上一层。"""
|
||||
q = artifact_store.content_hash(artifact_store.canonical_json(SNAPSHOT))
|
||||
c = artifact_store.content_hash(artifact_store.canonical_json(CHANNELS))
|
||||
top = await analyzed({"Q2": (SNAPSHOT, q)})
|
||||
child = await analyzed({"Q3": (CHANNELS, c)}, parents={"A1": top}, query="Q3", by=["channel"],
|
||||
parent={"of": "A1", "key": ["华东"]})
|
||||
made = await sealed_run("[[a:A2]]。", result=child, parents=[top], queries=[CHANNELS])
|
||||
body = (await client.get(f"/api/runs/{made['run']}/evidence/segments/{seg_id(made['doc'], 'a:A2')}")).json()
|
||||
inputs = [s for s in body["chain"] if s["step"] == "analysis_input"]
|
||||
assert [s["role"] for s in inputs] == ["parent", "data"]
|
||||
assert inputs[0]["alias"] == "A1" and inputs[0]["phrase"] == top["phrase"] and inputs[0]["sealed"] is True
|
||||
assert inputs[0]["role_label"] == "上一层拆解" and inputs[1]["query"] == "Q2"
|
||||
again = next(s for s in body["chain"] if s["step"] == "analysis_recompute")
|
||||
assert again["status"] == "same", again
|
||||
graph = (await client.get(f"/api/runs/{made['run']}/evidence")).json()
|
||||
edges = {(e["from"], e["to"], e["rel"]) for e in graph["edges"]}
|
||||
assert {("A2", "A1", "derived_from"), ("A2", "Q2", "derived_from"), ("A1", "Q1", "derived_from")} <= edges
|
||||
|
||||
|
||||
#: 本周、上周两份同形查询(只差日期):口径卡跨这两份算环比时自动做可比性检查
|
||||
WINDOW = "SELECT SUM(gmv) AS gmv FROM orders WHERE order_date >= '{a}' AND order_date < '{b}'"
|
||||
LAST_WEEK = {"columns": ["gmv"], "column_types": {"gmv": "number"}, "rows": [[2220]],
|
||||
"sql": WINDOW.format(a="2026-10-05", b="2026-10-12"), "source": "shop", "truncated": False}
|
||||
THIS_WEEK = {**LAST_WEEK, "rows": [[2026]], "sql": WINDOW.format(a="2026-10-12", b="2026-10-19")}
|
||||
|
||||
|
||||
async def comparability_card() -> tuple[dict[str, Any], dict[str, Any]]:
|
||||
"""一张跨两周算环比的口径卡,指标上记着口径卡执行时自动做的可比性检查(结论落成工件,不进台账):(卡, 可比性结论)。"""
|
||||
last = artifact_store.content_hash(artifact_store.canonical_json(LAST_WEEK))
|
||||
this = artifact_store.content_hash(artifact_store.canonical_json(THIS_WEEK))
|
||||
result = await analyzed({"fetch·base": (LAST_WEEK, last), "fetch·target": (THIS_WEEK, this)},
|
||||
{"op": "comparable", "mode": None, "params": {}},
|
||||
base_query="fetch·base", target_query="fetch·target")
|
||||
found = await artifact_store.put_json(result, kind=ARTIFACT_KIND, node_id="card")
|
||||
verdict = result["verdict"]["code"]
|
||||
card = {"kind": "metric_set", "caliber": "周报口径", "caliber_version": "v1", "metrics": [
|
||||
{"id": "wow", "name": "环比", "value": -0.0874, "unit": "",
|
||||
"expression": "cell(nodes.fetch, 0, 'gmv') / cell(nodes.fetch, 0, 'gmv') - 1",
|
||||
"substituted": "2026 / 2220 - 1", "recompute_ok": True,
|
||||
"inputs": [{"path": "cell(nodes.fetch, 0, 'gmv')", "value": 2026, "node_id": "fetch", "via": "tool_cell",
|
||||
"artifact": this, "locator": {"row": 0, "column": "gmv"}},
|
||||
{"path": "cell(nodes.fetch, 0, 'gmv')", "value": 2220, "node_id": "fetch", "via": "tool_cell",
|
||||
"artifact": last, "locator": {"row": 0, "column": "gmv"}}],
|
||||
"comparability": {"artifact": found, "verdict": verdict, "phrase": result["phrase"], "reasons": []}}]}
|
||||
return card, result
|
||||
|
||||
|
||||
async def test_the_metric_step_opens_the_comparability_result(client):
|
||||
"""可比性结论不进台账、只从口径卡的指标集里引用:照卡上记的 sha256 取回,随卡封存;点得开它的整条出处链,复算是真的。"""
|
||||
card, result = await comparability_card()
|
||||
made = await sealed_run("环比 [[m:wow]]。", card=card, queries=[LAST_WEEK, THIS_WEEK])
|
||||
body = (await client.get(f"/api/runs/{made['run']}/evidence/segments/{seg_id(made['doc'], 'm:wow')}")).json()
|
||||
view = body["chain"][0]["comparability"]
|
||||
assert view["verdict"] == result["verdict"]["code"] and view["phrase"] == result["phrase"]
|
||||
assert view["sealed"] is True and view["hash_ok"] is True and view["method"] == "comparable@1"
|
||||
steps = view["steps"]
|
||||
assert [s["step"] for s in steps] == ["analysis", "analysis_method", "analysis_calc", "analysis_input",
|
||||
"analysis_input", "analysis_recompute", "analysis_seal"]
|
||||
judged, method, calc, base, target, again, seal = steps
|
||||
assert judged["origin"] == "metric" and judged["cited"] == {"what": "phrase"} and judged["render_ok"] is True
|
||||
assert judged["method_label"] == "可比性检查" and judged["node_id"] == "card"
|
||||
assert [r[0] for r in calc["table"]["rows"]] == ["同口径", "窗口长度", "整期", "星期构成", "数据问题"]
|
||||
assert (base["role"], target["role"]) == ("base", "target")
|
||||
assert base["query"] == "Q2" and target["query"] == "Q3" and base["sealed"] is True
|
||||
assert again["status"] == "same" and again["text"] == "复算一致:按封存的输入重新计算,结论逐项相同"
|
||||
assert seal["via"] == "metric_set" and seal["ledger"] is False and seal["sealed"] is True
|
||||
assert seal["note"].startswith("这份可比性结论不进台账")
|
||||
# 两份输入快照的查询步骤在指标链里(指标的输入就是这两份),可比性的链里不再给
|
||||
assert [s["step"] for s in body["chain"]].count("query") == 2 and "query" not in [s["step"] for s in steps]
|
||||
|
||||
|
||||
async def test_a_card_without_the_comparability_artifact_says_only_what_it_recorded(client):
|
||||
card = {"kind": "metric_set", "caliber": "周报口径", "caliber_version": "v1", "metrics": [
|
||||
{"id": "wow", "name": "环比", "value": -0.087, "unit": "", "expression": "1", "inputs": [],
|
||||
"comparability": {"artifact": "", "verdict": "incomparable", "phrase": "两期不可比:本期只有 3 天数据",
|
||||
@@ -242,6 +350,15 @@ async def test_the_metric_step_shows_the_comparability(client):
|
||||
"artifact": None, "hash_ok": None, "sealed": False}
|
||||
|
||||
|
||||
async def test_the_app_metric_endpoint_opens_the_comparability_too(client):
|
||||
card, _ = await comparability_card()
|
||||
made = await sealed_run("环比 [[m:wow]]。", card=card, queries=[LAST_WEEK, THIS_WEEK])
|
||||
body = (await client.get(f"/api/runs/{made['run']}/evidence/metrics/card/wow")).json()
|
||||
view = body["chain"][0]["comparability"]
|
||||
assert view["sealed"] is True and view["steps"][-2]["status"] == "same"
|
||||
assert [s["query"] for s in view["steps"] if s["step"] == "analysis_input"] == [None, None]
|
||||
|
||||
|
||||
# --------------------------------------------------------------------------
|
||||
# 审计
|
||||
# --------------------------------------------------------------------------
|
||||
@@ -286,7 +403,7 @@ def test_audit_labels_cover_the_new_violation_codes():
|
||||
# --------------------------------------------------------------------------
|
||||
|
||||
SCOPE = {"analysis_scope": {
|
||||
"phenomena": [{"alias": "A1", "op": "decompose", "phrase": "销售额下降 194 万(-8.7%),主要来自华东(-120 万,占 62%)"}],
|
||||
"phenomena": [{"alias": "A1", "op": "decompose", "phrase": "销售额下降 194(-8.7%),主要来自华东(-120,占 61.9%)"}],
|
||||
"decomposed": [{"alias": "A1", "op": "decompose", "mode": "additive", "dims": ["地区"], "verdict": "driver",
|
||||
"phrase": "主要来自华东"}],
|
||||
"excluded": [], "refused": [{"alias": "A1", "op": "baseline", "reason": "历史只有 5 期,至少要 8 期"}],
|
||||
|
||||
Reference in New Issue
Block a user