diff --git a/backend/app/engine/nodes/io.py b/backend/app/engine/nodes/io.py index dec7fd82..948ff862 100644 --- a/backend/app/engine/nodes/io.py +++ b/backend/app/engine/nodes/io.py @@ -4,6 +4,7 @@ import json import re from typing import Any +from app.analysis.contract import LEDGER_KIND from app.core.events import EventType from app.engine.context import NodeContext, NodeError from app.engine.expressions import ExpressionError, eval_expression @@ -568,8 +569,12 @@ def _check_citations( if label: hit["caliber"] = label elif cite["kind"] == "cell": - # 查询单元格:Q1.r0.gmv(全局编号,和 doc.catalog 一致) + # 查询单元格:Q1.r0.gmv(全局编号,和 doc.catalog 一致);分析结论结果表的一格同一种写法(A3.r0.share) hit["cell"] = f"{cite['alias']}.r{cite['locator']['row']}.{cite['locator']['column']}" + elif cite["kind"] == LEDGER_KIND: + # 分析结论的具名标量([[a:A3.top_share]]):A3.top_share + key = (cite.get("locator") or {}).get("key") + hit["analysis"] = f"{cite['alias']}.{key}" if key else str(cite["alias"]) else: hit["input"] = cite["locator"].get("field") out["matched"].append({**hit, "segment": seg["id"], "unit": unit["id"], "start": start, "end": end, @@ -1027,37 +1032,51 @@ def _cited_queries(doc: dict[str, Any], catalog: dict[str, Any]) -> tuple[dict[s 单元格([[v:Q1.r0.gmv]],[[table:]] 展开后也是一格一格的引用)按查询数:一格写「Q1 第 1 行「gmv」」, 几格写「Q1 的 3 格」;只拿它当依据([[see:Q1]])的写「Q1(结论依据)」。出具契约据此判这些查询有没有通过 SQL 检查(issuance.sql_check_gaps)。 + + 分析结论按它的工件记(sql_problems.query_problems 顺着结论的输入找到查询):结果表的格同上(A3 第 1 行「share」); + 判定语、具名标量写「A3(判定语)」「A3「top_share」」,当依据([[see:A3]])的写「A3(结论依据)」——引用了一份分析 + 结论,它的输入查询上的问题照样计入缺口。 """ from app.engine.evidence import iter_units cells: dict[str, list[tuple[Any, Any]]] = {} + said: dict[str, list[str]] = {} # 分析结论的判定语、具名标量:{A 编号: [键或 None…]} support: list[str] = [] for _, unit in iter_units(doc): for seg in unit.get("segments") or []: cite = seg.get("cite") or {} - if cite.get("kind") == "cell" and cite.get("status") == "resolved" and cite.get("alias"): + if cite.get("status") != "resolved" or not cite.get("alias"): + continue + if cite.get("kind") == "cell": loc = cite.get("locator") or {} cells.setdefault(str(cite["alias"]), []) if (loc.get("row"), loc.get("column")) not in cells[str(cite["alias"])]: cells[str(cite["alias"])].append((loc.get("row"), loc.get("column"))) + elif cite.get("kind") == LEDGER_KIND: + keys = said.setdefault(str(cite["alias"]), []) + key = (cite.get("locator") or {}).get("key") + if key not in keys: + keys.append(key) for cite in unit.get("see") or []: alias = str((cite or {}).get("alias") or "") - if (catalog.get(alias) or {}).get("kind") == "query" and alias not in support: + if (catalog.get(alias) or {}).get("kind") in ("query", LEDGER_KIND) and alias not in support: support.append(alias) queries: dict[str, list[str]] = {} # 目录里全部查询的编号:合并查询的输入没被直接引用,缺口里也要写得出它是 Q 几 aliases = {str(e["artifact"]): str(e["alias"]) for e in catalog.values() if isinstance(e, dict) and e.get("kind") == "query" and e.get("artifact") and e.get("alias")} - for alias in [*cells, *(a for a in support if a not in cells)]: + for alias in dict.fromkeys([*cells, *said, *support]): artifact = (catalog.get(alias) or {}).get("artifact") if not isinstance(artifact, str) or not artifact: continue hit = cells.get(alias) or [] + refs = [] if len(hit) == 1 and isinstance(hit[0][0], int): - ref = f"{alias} 第 {hit[0][0] + 1} 行「{hit[0][1]}」" - else: - ref = f"{alias} 的 {len(hit)} 格" if hit else f"{alias}(结论依据)" - queries.setdefault(artifact, []).append(ref) + refs.append(f"{alias} 第 {hit[0][0] + 1} 行「{hit[0][1]}」") + elif hit: + refs.append(f"{alias} 的 {len(hit)} 格") + refs.extend(f"{alias}「{key}」" if key else f"{alias}(判定语)" for key in said.get(alias) or []) + queries.setdefault(artifact, []).extend(refs or [f"{alias}(结论依据)"]) return queries, aliases diff --git a/backend/tests/test_contract_analysis_claims.py b/backend/tests/test_contract_analysis_claims.py index b03793f9..d61f6b47 100644 --- a/backend/tests/test_contract_analysis_claims.py +++ b/backend/tests/test_contract_analysis_claims.py @@ -299,3 +299,74 @@ async def test_a_metric_read_from_a_query_cell_is_traced(client, monkeypatch, sh "expression": "cell(nodes.fetch, 0, 'gmv')"}]) issuance = await governed(client, graph) assert "untraced" not in issuance and not any("没追到原始数据" in g for g in issuance["gaps"]) + + +# -------------------------------------------------------------------------- +# 引用了分析结论:输入查询的 SQL 检查问题计入缺口,具名标量记进逐个数字的出处 +# -------------------------------------------------------------------------- + + +@pytest.fixture +def analyses(monkeypatch): + """写作目录里多几条分析结论:{A 编号: 结论工件 id}。台账里的分析结论由入口(分析卡、Agent 的分析工具)记,这里直接 + 补进目录:报告节点和出口复核按同一个 report_catalog 建目录,一处补上。""" + from app.core import artifact_store + from app.engine.evidence import make_eid + from app.engine.nodes import report as report_mod + + added: dict[str, str] = {} + original = report_mod.report_catalog + + def catalog(state, spec, node): + out = original(state, spec, node) + for alias, artifact in added.items(): + out[alias] = {"alias": alias, "kind": "analysis", "artifact": artifact, "item": "drop", + "eid": make_eid("analysis", artifact, {}), "locator": {}, + "result": artifact_store.load(artifact)} + return out + + monkeypatch.setattr(report_mod, "report_catalog", catalog) + return added + + +async def stored_driver(*, checks=None) -> str: + """一份「销售额下降主要来自华东」的拆解结论,输入是一份(可能带着 SQL 检查问题的)查询快照,都存进工件库。""" + from app.analysis.testing import fake_result + from app.core import artifact_store + + query = await artifact_store.put_json({"source": "shop", "sql": "SELECT region, week, SUM(amount) AS gmv FROM orders", + "columns": ["region", "week", "gmv"], "rows": [["华东", "W41", 820]], + "truncated": False, **({"checks": checks} if checks else {})}, + kind="query_snapshot") + result = fake_result(inputs=[{"ref": "Q2", "artifact": query, "role": "data", + "columns": ["region", "week", "gmv"], "rows": "all"}]) + return await artifact_store.put_json(result, kind="analysis_result") + + +FANOUT = {"code": "fanout_sum", "level": "error", "message": "一对多关联后求和,订单金额被重复计算。改法:先汇总再关联"} + + +@pytest.mark.parametrize("text", [ + "[[a:A1]]。[[see:m:gmv]]", + "本周销售额 [[m:gmv]],下降主要来自华东。[[see:m:gmv,A1]]", + "华东占变化的 [[a:A1.top_share|pct]]。[[see:m:gmv]]", +]) +async def test_an_analysis_built_on_a_failing_query_degrades_like_the_query_itself(monkeypatch, analyses, text): + analyses["A1"] = await stored_driver(checks=[FANOUT]) + writes(monkeypatch, text) + issuance = await issued(weekly()) + assert issuance["tier"] == "degraded" + [gap] = [g for g in issuance["gaps"] if "SQL 检查" in g] + assert "一对多关联后求和" in gap and "A1" in gap + analyses["A1"] = await stored_driver() + clean = await issued(weekly()) + assert clean["tier"] == "formal" and not any("SQL 检查" in g for g in clean["gaps"]) + + +async def test_a_named_scalar_is_listed_as_an_analysis_number(monkeypatch, analyses): + analyses["A1"] = await stored_driver() + writes(monkeypatch, "华东占变化的 [[a:A1.top_share|pct]],下降主要来自华东。[[see:m:gmv,A1]]") + issuance = await issued(weekly()) + [hit] = [m for m in issuance["matched"] if m.get("analysis")] + assert hit["analysis"] == "A1.top_share" and hit["token"] == "61.9%" and "input" not in hit + assert "analysis_claims" not in issuance, "判断词挂对了:check 下没有可标注的"