feat(check): 响应方式识别(C12)和「这条要不要投标人回答」;评测集纠错
- checklist/response_mode.py:按类别(技术 / 商务)识别招标文件的响应方式(逐条响应 / 默认接受),规则优先; 两种写法冲突按逐条,认不出按逐条并写进覆盖范围。 - check/applicability.py:无效情形(★ / 斜体下划线标注的条款除外)、日期与递交、没要求附材料的评分标准、 无需提供证明或由采购人查询的资格,不逐条找响应,写进覆盖范围;要求附证明材料按原文判断,中标后才交的不算。 - 默认接受的要求没找到响应:一般要求记 implicit_accepted、不出问题;实质性要求记 implicit_review、提示级 人工核对;要附证明材料的不免(迁移 0006)。 - match.py:修同一行单元格又是别的条目候选句时的 KeyError(每条清单项单独记候选句,句子从总表里查)。 - 评测集:注入器同一要求的所有表格行一起改;散文、附件、格式文件里也响应了的要求不挂漏响应类问题; 错位响应必须是另一个参数;复核判不成立的单元不再挂;regen.py 按固定种子重新生成。 开发集 17 份:规矩投标文件报出的问题 639 → 229,误报 21.3 → 12.4 / 100,召回 68% → 65%,致命级 20 → 21 / 32 (docs/architecture/check-eval.md §4)。 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
@@ -211,4 +211,5 @@ samples/
|
||||
- 2026-11-04:查标评测开发集第一版:7 份招标文件、14 份投标文件(风格 A / B 轮换),注入问题 222 个、Fable 复核通过 188 个。查标召回 40%(致命级 49%),带标志词的问题只有 22%;规矩投标文件报出 699 个问题,近一半在不需要响应的要求上(`docs/architecture/check-eval.md` §2)。
|
||||
- 2026-11-04:用户抽查第二批复核 28 条,全部同意(两批合计 45 条)。
|
||||
- 2026-11-04:查标第一轮改进:候选句带同一行的单元格、提示词 check.match v2。开发集 17 份上召回 30% → 50%,带标志词的问题 2% → 54%,误报略降,输入 token +44%;漏响应和响应对象错位变差(`docs/architecture/check-eval.md` §3)。评测集修了待修项、加了 3 份低密度投标文件,第三批复核 55 条、用户抽查 11 条全部同意。
|
||||
- 下一步:看 D01、D09 的漏报明细改提示词;查标第 2 步(响应方式 C12、投标侧结构识别、规则)。
|
||||
- 2026-11-04:查标第二轮:评测集纠错(漏响应、错位的「变差」多是生成器只改了一处);新增响应方式识别(C12,规则优先)和「这条要不要投标人回答」的判断。开发集 17 份上规矩投标文件报出的问题 639 → 229,误报 21.3 → 12.4 / 100,召回 68% → 65%,致命级 20 → 21 / 32;证明材料缺失几乎查不出(扫描件)。第四批复核 45 条,用户抽查 9 条全部同意(四批共 65 条)。见 `docs/architecture/check-eval.md` §4。
|
||||
- 下一步:扫描件兜底和文件要在(C10);响应方式存进清单并可改;查标第 2 步的投标侧结构识别和规则。
|
||||
|
||||
@@ -103,8 +103,34 @@
|
||||
|
||||
**读法**:整行上下文解决了「看不见响应格、偏离格」的问题,带标志词的问题从几乎查不出到过半;误报略降。代价是输入 token 多 44%。**两处变差**:漏响应(D01)和响应对象错位(D09)。推测是同一行或相邻行的内容让模型更容易「找到点什么就判已响应」:漏掉的那一行不在了,检索命中的是相邻要求的行;错位的那一行响应格写的是另一条参数,模型没比对参数名。下一轮在提示词里要求先核对响应格说的是不是同一个参数 / 设备,并单独看 D01、D09 的漏报明细。不需要响应的要求上的误报(329)没有变化,要靠第 ② 项(C12)。
|
||||
|
||||
## 4. 下一步
|
||||
## 4. 第二轮:评测集纠错 + 响应方式和「要不要回答」(2026-11-04)
|
||||
|
||||
1. 看 D01、D09 的漏报明细,改提示词(v3):先核对响应格说的是不是同一个参数 / 设备。
|
||||
2. 第 2 步(投标侧结构识别、规则、C12 响应方式抽取)和第 3 步(整行上下文、偏离表直取、presence 判断)按 §2.2 的顺序改,每改一轮在开发集 14 份上重跑对比。
|
||||
**评测集纠错**:§3 里漏响应、响应对象错位「变差」,逐条看漏报后发现多数是**生成器的错**,不是查标的错:读标清单有重复条目、S04 同一参数列在两张表里、技术方案散文也复述了要求,注入器只改了其中一处,别处还留着正确的响应,模型找到它是对的。改为同一要求的所有表格行一起改;在散文、附件、格式文件里也响应了的要求不挂漏响应 / 数值不满足 / 错位;错位响应必须是另一个参数的。马虎的投标文件按原种子重新生成,新目标 45 个由 Fable 复核(通过 40 个),用户抽查 9 个全部同意(四批合计抽查 65 个,全部同意)。
|
||||
|
||||
**查标改了什么**(第 2 步的 C12 部分):
|
||||
- `checklist/response_mode.py`:按类别(技术 / 商务)识别招标文件的响应方式,规则优先;冲突按逐条,认不出按逐条并标明。开发集 7 份的识别结果和档案一致。(还没存进清单、没做清单页的显示和修改。)
|
||||
- `check/applicability.py`:不需要投标人回答的条目不逐条找响应,写进覆盖范围(无效情形 —— 但 ★ / 斜体下划线标注的条款照常找、日期与递交、没要求附材料的评分标准、无需提供证明或由采购人查询的资格)。要求附证明材料按原文判断(读标的 evidence_required 常常是空的),中标后才交的不算。
|
||||
- 默认接受的要求没找到响应:一般要求记「按招标文件默认接受」、不出问题;实质性要求出提示级「请人工核对」;要附证明材料的不免(迁移 0006 加了两种结论)。
|
||||
|
||||
| 指标(17 份,复核通过的问题) | v1 | v2 + 整行 | + 响应方式和「要不要回答」 |
|
||||
|---|---|---|---|
|
||||
| 查标召回 | 48 / 150(32%) | 102 / 150(68%) | 97 / 150(65%) |
|
||||
| 致命级召回 | 9 / 32 | 20 / 32 | **21 / 32** |
|
||||
| 规矩的投标文件报出的问题 | 699 | 639 | **229** |
|
||||
| 规矩响应的要求上的误报 | 24.7 / 100 | 21.3 / 100 | **12.4 / 100** |
|
||||
| 不需要响应的要求上报的问题 | 334 | 329 | **77** |
|
||||
| 默认接受的要求被报(T06) | 39 | 35 | **1** |
|
||||
| 输入 token | 189 万 | 273 万 | 224 万 |
|
||||
|
||||
按问题类型(最后一列):漏响应 19 / 24、写明不支持 13 / 14、数值不满足 4 / 4、负偏离未说明 39 / 59、评分材料缺失 9 / 14、错位 6 / 13;**证明材料缺失 1 / 17**。
|
||||
|
||||
**读法**:规矩的投标文件上的问题少了三分之二,召回只掉了 3 个点(掉的主要是只在默认接受的文件里、被正确判为「默认接受」的条目,以及证明材料)。证明材料缺失几乎查不出:资格证明、检测报告多是扫描件页,没有文字层,规矩的投标文件上同样报「未找到」——这是下一项(C10 扫描件兜底 + 文件要在的判断)。
|
||||
|
||||
**一处要说明的**:调响应方式识别时,打印结果里顺带出现了留出集 S03、S06、S09(招标文件本身,不是投标文件)的识别结果,其中 S03、S09 和规范 §2 的记录不一致。没有据此改任何规则,留到留出集正式评测时再看。
|
||||
|
||||
## 5. 下一步
|
||||
|
||||
1. 证明材料和格式文件:扫描件兜底(C10)、文件要在(presence)按文件在不在判断。
|
||||
2. 响应方式存进清单、清单页显示并可改(C12 的人工确认点)。
|
||||
3. 第 2 步(投标侧结构识别、规则、C12 响应方式抽取)和第 3 步(整行上下文、偏离表直取、presence 判断)按 §2.2 的顺序改,每改一轮在开发集 14 份上重跑对比。
|
||||
3. 留出集(S03、S06、S09,风格 B + C,封存的词库)和真实国动投标文件,在第 3 步冻结代码后只跑一次。
|
||||
|
||||
@@ -9,7 +9,7 @@
|
||||
1. 没有真实投标文件(真实的国动投标文件只有一份,留作最终检验,见 §4),所以按 S01–S10 的真实招标文件**自己造虚构的投标文件**。
|
||||
2. **规矩的投标文件靠构造保证,不靠复核**:偏离表、响应表、报价表、资格和评分材料清单由脚本从清单生成,技术参数按 `op / value / unit` 算一个满足的值;LLM 只写散文(技术方案、投标函叙述)和按风格改写表格行(截断、同义替换、重新编号),逐行小提示,不整本生成。
|
||||
3. 投标文件对招标文件的**全部要求**响应:「全集」用读标正式清单(`evals/out/runs/fix3/<ID>/items.json`),按「招标侧页码 + 片段」记,不只响应标注集的单元。
|
||||
4. 再用确定性的脚本往里**注入问题**得到马虎的投标文件。注入了什么、注在哪里,脚本同时写进**响应清单**,它就是金标准。**问题只挂在人工复核过的标注单元上**(`evals/gold/<ID>.json`,优先实质性条款和资格条件);`optional` 和复核时补的单元不挂问题。
|
||||
4. 再用确定性的脚本往里**注入问题**得到马虎的投标文件。注入了什么、注在哪里,脚本同时写进**响应清单**,它就是金标准。**问题只挂在复核过的标注单元上**(`evals/gold/<ID>.json`,优先实质性条款和资格条件;复核方式见决策 E1)。`optional` 和复核时补的单元也可以挂,按内容复核(四批复核和用户抽查都是这样做的);**上一次复核判为不成立的单元不再挂问题**。
|
||||
5. 评分只对照响应清单,不用检索、偏离表对齐这些查标自己的逻辑来判断对错。
|
||||
|
||||
## 2. 每条要求怎么才算「响应了」:响应方式
|
||||
|
||||
@@ -41,10 +41,18 @@ UNIT_DOWN = {"TB": ("GB", 1024), "GB": ("MB", 1024), "Gbps": ("Mbps", 1000), "s"
|
||||
class Mutator:
|
||||
def __init__(self, spec: dict[str, Any], vocab: dict[str, Any], rng: random.Random) -> None:
|
||||
self.spec, self.vocab, self.rng = spec, vocab, rng
|
||||
self.vols = spec["volumes"]
|
||||
self.unit_defects: dict[str, dict[str, Any]] = {}
|
||||
self.rule_defects: list[dict[str, Any]] = []
|
||||
self.umbrella: set[str] = set()
|
||||
self.vols = spec["volumes"]
|
||||
self.rejected: set[str] = set() # 以前复核判为不成立的单元(同一招标文件的任何投标文件里),不再挂问题
|
||||
# 在表格以外(技术方案散文、附件、格式文件)也响应了的单元:删表格行、改响应值都不构成问题,不挂这几类
|
||||
self.prose = [
|
||||
{t[i : i + 2] for t in [re.sub(r"\W", "", b["text"])] for i in range(len(t) - 1)}
|
||||
for b in self.blocks
|
||||
if b.get("type") == "para" and b.get("form") == "prose"
|
||||
]
|
||||
self.elsewhere = {u for b in self.blocks if b.get("type") != "table" for u in b.get("responds") or []}
|
||||
|
||||
@property
|
||||
def blocks(self) -> list[dict[str, Any]]:
|
||||
@@ -80,7 +88,7 @@ class Mutator:
|
||||
return out
|
||||
|
||||
def free(self, units: list[str]) -> bool:
|
||||
return bool(units) and not any(u in self.unit_defects for u in units)
|
||||
return bool(units) and not any(u in self.unit_defects or u in self.rejected for u in units)
|
||||
|
||||
def mark(
|
||||
self, units: list[str], defect: str, variant: str, granularity: str = "single", worded: bool = True
|
||||
@@ -103,10 +111,11 @@ class Mutator:
|
||||
chunk = rows[start : start + n]
|
||||
if not all(self.free(r.get("responds") or []) for r in chunk if r.get("responds")):
|
||||
continue
|
||||
if any(set(r.get("responds") or []) & self.elsewhere for r in chunk):
|
||||
continue
|
||||
for r in chunk:
|
||||
self.mark(r.get("responds") or [], "D01", "删连续一段", "block", worded=False)
|
||||
del rows[start : start + n]
|
||||
self.drop_everywhere([u for r in chunk for u in r.get("responds") or []])
|
||||
self.drop_everywhere([u for r in chunk for u in r.get("responds") or []], [(t, r) for r in chunk])
|
||||
return
|
||||
|
||||
def row_ops(self, n: int) -> None:
|
||||
@@ -126,10 +135,43 @@ class Mutator:
|
||||
done += 1
|
||||
break
|
||||
|
||||
def drop_everywhere(self, units: list[str]) -> None:
|
||||
"""漏响应要在所有表里都漏:S04 的技术参数同时列在全量偏离表和 ★ 表里,只删一张表等于没漏。"""
|
||||
def in_prose(self, text: str) -> bool:
|
||||
"""技术方案散文里基本复述了这条要求(字 2-gram 有六成以上出现在同一段里):表格里删掉或改错,散文里还在响应。"""
|
||||
t = re.sub(r"\W", "", text)
|
||||
grams = {t[i : i + 2] for i in range(len(t) - 1)}
|
||||
return len(grams) >= 6 and any(len(grams & g) / len(grams) >= 0.6 for g in self.prose)
|
||||
|
||||
def _req_text(self, t: dict[str, Any], r: dict[str, Any]) -> str:
|
||||
at = self.col(t, "requirement")
|
||||
return re.sub(r"\W", "", r["cells"][at]) if at is not None else ""
|
||||
|
||||
def siblings(self, t: dict[str, Any], r: dict[str, Any]) -> list[tuple[dict[str, Any], dict[str, Any]]]:
|
||||
"""同一条要求在投标文件里的所有表格行:挂同一个标注单元,或者要求原文相同(读标清单常把同一要求抽成几条,
|
||||
S04 的参数同时列在两张表里)。注入的问题要在这些行上一起做,否则别处还留着正确的响应,就不是问题了。"""
|
||||
units, text = set(r.get("responds") or []), self._req_text(t, r)
|
||||
out = []
|
||||
for t2 in self.tables():
|
||||
for r2 in t2["rows"]:
|
||||
if r2 is r:
|
||||
continue
|
||||
if units & set(r2.get("responds") or []) or (len(text) >= 6 and self._req_text(t2, r2) == text):
|
||||
out.append((t2, r2))
|
||||
return out
|
||||
|
||||
def propagate(self, t: dict[str, Any], r: dict[str, Any]) -> None:
|
||||
"""把这一行改过的响应、偏离、说明照样写到同一要求的其他行上。"""
|
||||
for t2, r2 in self.siblings(t, r):
|
||||
for field in ("answer", "deviation", "note"):
|
||||
a, b = self.col(t, field), self.col(t2, field)
|
||||
if a is not None and b is not None:
|
||||
r2["cells"][b] = r["cells"][a]
|
||||
|
||||
def drop_everywhere(self, units: list[str], rows: list[tuple[dict[str, Any], dict[str, Any]]] = ()) -> None:
|
||||
"""漏响应要在所有表里都漏:S04 的技术参数同时列在全量偏离表和 ★ 表里,只删一张表等于没漏;
|
||||
要求原文相同的重复行也一起删。"""
|
||||
drop = {id(r2) for t, r in rows for _, r2 in self.siblings(t, r)} | {id(r) for _, r in rows}
|
||||
for t in self.tables():
|
||||
t["rows"] = [r for r in t["rows"] if not set(r.get("responds") or []) & set(units)]
|
||||
t["rows"] = [r for r in t["rows"] if id(r) not in drop and not set(r.get("responds") or []) & set(units)]
|
||||
|
||||
@staticmethod
|
||||
def col(t: dict[str, Any], field: str) -> int | None:
|
||||
@@ -148,12 +190,14 @@ class Mutator:
|
||||
numeric = tp.get("op") in (">=", "<=", "=") and tp.get("value")
|
||||
if op == "D05" and (numeric or not FUNCTIONAL.search(req_text) or NON_FUNCTIONAL.search(req_text)):
|
||||
return False # 「暂不支持」只套在功能类要求上(数值参数、证书、清单、服务义务不会这么写)
|
||||
if op in ("D01", "D03", "D08", "D09") and (set(units) & self.elsewhere or self.in_prose(req_text)):
|
||||
return False
|
||||
if op == "D01":
|
||||
t["rows"].remove(r)
|
||||
self.drop_everywhere(units)
|
||||
self.drop_everywhere(units, [(t, r)])
|
||||
self.mark(units, "D01", "删一行", worded=False)
|
||||
elif op == "D03":
|
||||
return self._unmet(r, tp, ans, dev)
|
||||
if not self._unmet(r, tp, ans, dev):
|
||||
return False
|
||||
elif op == "D05":
|
||||
param = tp.get("param") or (r["cells"][req][:12] if req is not None else "该功能")
|
||||
r["cells"][ans] = self.phrase("D05", "phrases").format(param=param)
|
||||
@@ -171,11 +215,14 @@ class Mutator:
|
||||
r["cells"][ans] = self.phrase("D08", "phrases")
|
||||
self.mark(units, "D08", r["cells"][ans], worded=False)
|
||||
elif op == "D09":
|
||||
mine = str(tp.get("param") or "")
|
||||
others = [
|
||||
x
|
||||
for _, x in cands
|
||||
if x is not r
|
||||
and (x.get("tp") or {}).get("param")
|
||||
and x["tp"]["param"] != mine # 必须是另一个参数:同名参数(另一台设备的同一指标)的响应可能恰好满足
|
||||
and (not mine or mine not in x["cells"][x.get("anchor", ans)])
|
||||
and x["cells"][x.get("anchor", ans)] != r["cells"][ans]
|
||||
]
|
||||
if not others:
|
||||
@@ -183,6 +230,8 @@ class Mutator:
|
||||
o = self.rng.choice(others)
|
||||
r["cells"][ans] = o["cells"][o.get("anchor", ans)]
|
||||
self.mark(units, "D09", "另一条参数的响应", worded=False)
|
||||
if op != "D01":
|
||||
self.propagate(t, r)
|
||||
return True
|
||||
|
||||
def _unmet(self, r: dict[str, Any], tp: dict[str, Any], ans: int, dev: int | None) -> bool:
|
||||
@@ -205,7 +254,7 @@ class Mutator:
|
||||
if dev is not None and self.rng.random() < 0.5:
|
||||
r["cells"][dev] = "无偏离" # 一半在偏离栏仍写「无偏离」(另一半保持原样,也是无偏离或正偏离)
|
||||
self.mark(r.get("responds") or [], "D03", variant, worded=False)
|
||||
return True
|
||||
return True # 同一要求的其他行由调用方(_apply)同步
|
||||
|
||||
def implicit_deviation(self, n: int) -> None:
|
||||
"""只列偏离的表:挑默认接受的要求写成偏离(D06 负偏离未说明 / D05 写明不支持)。"""
|
||||
@@ -421,6 +470,10 @@ def main() -> None:
|
||||
optional = {u.id for u in units if u.optional}
|
||||
modes = unit_modes(spec) # 注入前算:被删掉的行也要知道它本来该怎么响应
|
||||
m = Mutator(spec, json.loads(VOCAB[a.split].read_text()), random.Random(a.seed)) # noqa: S311 评测数据,不用于安全
|
||||
for f in OUT.glob(f"{tender}-*/manifest.json"):
|
||||
for uid, e in json.loads(f.read_text())["units"].items():
|
||||
if (e.get("review") or {}).get("ok") is False:
|
||||
m.rejected.add(uid) # noqa: S311 评测数据,不用于安全
|
||||
if not a.manifest_only:
|
||||
if a.density == "high":
|
||||
m.d01_block()
|
||||
|
||||
91
evals/check/regen.py
Normal file
91
evals/check/regen.py
Normal file
@@ -0,0 +1,91 @@
|
||||
"""按固定的种子重新生成开发集的马虎投标文件,沿用复核过的结论(同一招标文件里复核过的「单元 + 问题类型」),
|
||||
列出没复核过的新目标(evals/out/check/review/pending-units.json)。
|
||||
|
||||
cd server && ../.venv/bin/python ../evals/check/regen.py
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
import pathlib
|
||||
import subprocess
|
||||
import sys
|
||||
|
||||
EVALS = pathlib.Path(__file__).resolve().parents[1]
|
||||
OUT = EVALS / "out" / "check"
|
||||
# (规矩的投标文件,马虎的投标文件,种子,密度):开发集的固定组合(GUIDELINES §4)
|
||||
SLOPPY = [
|
||||
("S02-B", "S02-B-sloppy", 11, "high"),
|
||||
("S04-A", "S04-A-sloppy", 7, "high"),
|
||||
("S10-B", "S10-B-sloppy", 13, "high"),
|
||||
("S01-A", "S01-A-sloppy", 17, "high"),
|
||||
("S05-B", "S05-B-sloppy", 23, "high"),
|
||||
("S07-A", "S07-A-sloppy", 19, "high"),
|
||||
("S08-B", "S08-B-sloppy", 29, "high"),
|
||||
("S04-B", "S04-B-low", 31, "low"),
|
||||
("S08-A", "S08-A-low", 37, "low"),
|
||||
("S10-A", "S10-A-low", 41, "low"),
|
||||
]
|
||||
|
||||
|
||||
def reviews(tender: str) -> dict[tuple[str, str], dict]:
|
||||
out = {}
|
||||
for f in OUT.glob(f"{tender}-*/manifest.json"):
|
||||
for uid, e in json.loads(f.read_text())["units"].items():
|
||||
if e.get("review") and e.get("expect") == "defect":
|
||||
out[(uid, e["defect"])] = {k: v for k, v in e["review"].items() if k != "user_spot_check"}
|
||||
return out
|
||||
|
||||
|
||||
def main() -> None:
|
||||
prior = {t: reviews(t) for t in {c[:3] for c, *_ in SLOPPY}}
|
||||
pending = []
|
||||
for clean, sloppy, seed, density in SLOPPY:
|
||||
tender = clean[:3]
|
||||
cmd = [
|
||||
sys.executable,
|
||||
str(EVALS / "check" / "mutate.py"),
|
||||
clean,
|
||||
sloppy,
|
||||
"--split",
|
||||
"dev",
|
||||
"--seed",
|
||||
str(seed),
|
||||
"--density",
|
||||
density,
|
||||
]
|
||||
print(subprocess.run(cmd, capture_output=True, text=True, check=True).stdout.strip())
|
||||
p = OUT / sloppy / "manifest.json"
|
||||
m = json.loads(p.read_text())
|
||||
gold = {x["id"]: x for x in json.loads((EVALS / "gold" / f"{tender}.json").read_text())["items"]}
|
||||
for uid, e in m["units"].items():
|
||||
if e.get("expect") != "defect":
|
||||
continue
|
||||
if e.get("defect") == "D02":
|
||||
e["mode"] = "evidence"
|
||||
r = prior[tender].get((uid, e["defect"]))
|
||||
if r:
|
||||
e["reviewed"], e["review"] = r["ok"], r
|
||||
else:
|
||||
x = gold[uid]
|
||||
pending.append(
|
||||
{
|
||||
"bid": sloppy,
|
||||
"tender": tender,
|
||||
"unit": uid,
|
||||
"cat": e["cat"],
|
||||
"basis": x.get("basis"),
|
||||
"anchors": x.get("anchors"),
|
||||
"note": x.get("note"),
|
||||
"mode": e["mode"],
|
||||
"defect": e["defect"],
|
||||
"variant": e.get("variant"),
|
||||
}
|
||||
)
|
||||
p.write_text(json.dumps(m, ensure_ascii=False, indent=1))
|
||||
(OUT / "review" / "pending-units.json").write_text(json.dumps(pending, ensure_ascii=False, indent=1))
|
||||
print("待复核的新目标:", len(pending))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
@@ -23,7 +23,8 @@ from bid_doctor.ai import prompts
|
||||
from bid_doctor.ai import service as ai_service
|
||||
from bid_doctor.ai.providers.base import near_limit
|
||||
from bid_doctor.ai.service import OutputIncomplete, provider
|
||||
from bid_doctor.check import findings, match
|
||||
from bid_doctor.check import applicability, findings, match
|
||||
from bid_doctor.checklist import response_mode
|
||||
from bid_doctor.core.config import get_settings
|
||||
from bid_doctor.core.db import TenantContext
|
||||
from bid_doctor.document.pipeline import text_layer
|
||||
@@ -120,6 +121,7 @@ async def main() -> None:
|
||||
ap.add_argument("--name", default="base")
|
||||
ap.add_argument("--baseline", action="store_true", help="只检索、不调模型的基线")
|
||||
ap.add_argument("--v1", action="store_true", help="第 1 步的做法:提示词 v1、不带同一行的单元格(对比基线)")
|
||||
ap.add_argument("--no-modes", action="store_true", help="不按响应方式和「要不要回答」筛选(C12 之前的做法)")
|
||||
args = ap.parse_args()
|
||||
d = OUT / args.bid
|
||||
manifest = json.loads((d / "manifest.json").read_text())
|
||||
@@ -137,9 +139,34 @@ async def main() -> None:
|
||||
return None
|
||||
|
||||
ctx = TenantContext(tenant_id=UUID(int=0), user_id=UUID(int=0))
|
||||
all_items = items
|
||||
decisions: dict[UUID, applicability.Decision] = {}
|
||||
if not args.no_modes:
|
||||
modes = response_mode.detect(gold.sentences(manifest["tender"]))
|
||||
decisions = {i: applicability.decide(p, modes) for i, p in items.items()}
|
||||
items = {i: p for i, p in items.items() if decisions[i].match}
|
||||
results, stats = await match.run(ctx, uuid4(), items, docs, progress)
|
||||
if decisions:
|
||||
match.apply_modes(results, items, decisions)
|
||||
rows = []
|
||||
for item_id, item in items.items():
|
||||
for item_id, item in all_items.items():
|
||||
if item_id not in items:
|
||||
rows.append(
|
||||
{
|
||||
"title": item["title"],
|
||||
"sentences": item["sentences"],
|
||||
"categories": item["categories"],
|
||||
"importance": item["importance"],
|
||||
"verdict": f"skipped:{decisions[item_id].skip}",
|
||||
"layer": None,
|
||||
"evidence_status": "none",
|
||||
"reason_code": None,
|
||||
"evidence": [],
|
||||
"detail": {},
|
||||
"finding": None,
|
||||
}
|
||||
)
|
||||
continue
|
||||
res = results[item_id]
|
||||
spec = findings.spec(res, item)
|
||||
rows.append(
|
||||
|
||||
31
server/migrations/versions/0006_check_verdict_implicit.py
Normal file
31
server/migrations/versions/0006_check_verdict_implicit.py
Normal file
@@ -0,0 +1,31 @@
|
||||
"""查标结论加「按招标文件默认接受」(决策 C12)。
|
||||
|
||||
- implicit_accepted:一般要求,不出问题;
|
||||
- implicit_review:实质性要求,提示级人工核对。
|
||||
|
||||
Revision ID: 0006
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from alembic import op
|
||||
|
||||
revision = "0006"
|
||||
down_revision = "0005"
|
||||
|
||||
VERDICTS = "'responded', 'not_found', 'negative_deviation', 'suspected_unmet', 'no_value', 'undetermined', 'scanned'"
|
||||
|
||||
|
||||
def upgrade() -> None:
|
||||
op.get_bind().exec_driver_sql(f"""
|
||||
ALTER TABLE check_item_result DROP CONSTRAINT check_item_result_verdict_check;
|
||||
ALTER TABLE check_item_result ADD CONSTRAINT check_item_result_verdict_check
|
||||
CHECK (verdict IN ({VERDICTS}, 'implicit_accepted', 'implicit_review'));
|
||||
""")
|
||||
|
||||
|
||||
def downgrade() -> None:
|
||||
op.get_bind().exec_driver_sql(f"""
|
||||
ALTER TABLE check_item_result DROP CONSTRAINT check_item_result_verdict_check;
|
||||
ALTER TABLE check_item_result ADD CONSTRAINT check_item_result_verdict_check CHECK (verdict IN ({VERDICTS}));
|
||||
""")
|
||||
56
server/src/bid_doctor/check/applicability.py
Normal file
56
server/src/bid_doctor/check/applicability.py
Normal file
@@ -0,0 +1,56 @@
|
||||
"""每条清单项要不要在投标文件里找响应、按什么响应方式判(决策 C12,`docs/architecture/check-eval.md` §2.2 第 ② 项)。
|
||||
|
||||
开发集上规矩的投标文件报出的问题,近一半落在不需要投标人回答的条目上(无效情形的列举、日期、评分标准、
|
||||
采购人自己查询的资格)。这些条目不逐条找响应,写进检查覆盖范围:
|
||||
- 无效情形、评审规定(只有「无效」类别、不是标注出来的条款):由规则检查(第 2 步)和人工核对覆盖;
|
||||
- 日期与递交:由规则检查(落款日期、有效期)和需人工核对覆盖;
|
||||
- 评分标准本身:没写要附什么材料的不找;写了的按证明材料找;
|
||||
- 资格条件里写明「无需提供」或由采购人查询(信用中国)的不找。
|
||||
其余按类别取招标文件的响应方式:技术参数用技术类,其余用商务类。
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import re
|
||||
from dataclasses import dataclass
|
||||
from typing import Any
|
||||
|
||||
from bid_doctor.checklist.response_mode import Mode
|
||||
|
||||
# 要求投标时附证明材料(读标的 evidence_required 常常是空的,按原文判断);中标后 / 签合同时才交的不算
|
||||
EVIDENCE = re.compile(r"(提供|附|出具)[^。;]{0,20}(证明|报告|截图|复印件|扫描件|证书|合同|承诺|声明|社保|发票)")
|
||||
AWARD = re.compile(r"(中标后|成交后|签订合同(前|时)|合同签订(前|时))")
|
||||
NO_PROOF = re.compile(r"无需提供|不需提供|无须提供|采购人[^。;]{0,12}(查询|核查|查验)|信用中国|信用记录")
|
||||
SKIP_LABELS = {
|
||||
"condition": "无效情形和评审规定",
|
||||
"schedule": "日期与递交要求",
|
||||
"scoring": "评分标准(没要求附证明材料)",
|
||||
"no_proof": "资格条件(写明无需提供证明,或由采购人查询)",
|
||||
}
|
||||
|
||||
|
||||
@dataclass(frozen=True, slots=True)
|
||||
class Decision:
|
||||
match: bool
|
||||
mode: str = "explicit" # explicit / implicit
|
||||
skip: str | None = None # 不找响应的原因(SKIP_LABELS 的键)
|
||||
mode_status: str = "found"
|
||||
evidence: bool = False # 要求投标时附证明材料:默认接受也不能免(没附就是没附)
|
||||
|
||||
|
||||
def decide(item: dict[str, Any], modes: dict[str, Mode]) -> Decision:
|
||||
cats = set(item.get("categories") or [])
|
||||
text = f"{item.get('title') or ''}{item.get('requirement_text') or ''}"
|
||||
if cats == {"invalidity"} and item.get("importance_basis") != "marker":
|
||||
return Decision(False, skip="condition") # 「……的,投标无效」这类情形;★ / 斜体下划线标注的条款是要求,照常找
|
||||
if "schedule" in cats and not cats & {"tech_param", "qualification", "scoring"}:
|
||||
return Decision(False, skip="schedule")
|
||||
evidence = bool(item.get("evidence_required") or EVIDENCE.search(text)) and not AWARD.search(text)
|
||||
if cats <= {"scoring"} and not evidence:
|
||||
return Decision(False, skip="scoring")
|
||||
if "qualification" in cats and "tech_param" not in cats and NO_PROOF.search(text):
|
||||
return Decision(False, skip="no_proof")
|
||||
mode = modes.get("technical" if "tech_param" in cats else "commercial")
|
||||
if mode is None:
|
||||
return Decision(True, evidence=evidence)
|
||||
return Decision(True, mode=mode.mode, mode_status=mode.status, evidence=evidence)
|
||||
@@ -52,8 +52,14 @@ def spec(result: Result, item: dict[str, Any]) -> Spec | None:
|
||||
name = str(item.get("title") or "")
|
||||
base = {"importance_basis": item.get("importance_basis"), **result.detail}
|
||||
scope = "、".join(result.detail.get("search_scope") or [])
|
||||
if result.verdict == "responded":
|
||||
if result.verdict in ("responded", "implicit_accepted"):
|
||||
return None
|
||||
if result.verdict == "implicit_review":
|
||||
text = (
|
||||
"招标文件规定未列入偏离表的视为接受,投标文件里没有找到对这条要求的明确响应。"
|
||||
"这是实质性要求,请人工核对所投内容确实满足。"
|
||||
)
|
||||
return Spec("manual_review", "hint", f"请人工核对:{name}", text, {**base, "mode": "implicit"})
|
||||
if result.verdict == "not_found" and result.evidence_status == "unverified":
|
||||
claimed = result.detail.get("claimed_pages") or []
|
||||
where = f"{_pages(claimed)}有响应" if claimed else "有响应"
|
||||
|
||||
@@ -74,6 +74,18 @@ class Stats:
|
||||
self.verify_failures[code] = self.verify_failures.get(code, 0) + 1
|
||||
|
||||
|
||||
def apply_modes(results: dict[UUID, Result], items: dict[UUID, dict[str, Any]], modes: dict[UUID, Any]) -> None:
|
||||
"""按招标文件的响应方式改结论(决策 C12):默认接受(implicit)的要求没找到响应不算问题,记「按招标文件默认接受」;
|
||||
实质性要求出提示级「请人工核对」,不出致命。找到了不满足的响应照常报。modes:清单项 → applicability.Decision。"""
|
||||
for item_id, d in modes.items():
|
||||
r = results.get(item_id)
|
||||
if r is None or d.mode != "implicit" or r.verdict != "not_found" or d.evidence:
|
||||
continue # 要附证明材料的要求,默认接受也免不了
|
||||
r.verdict = "implicit_review" if items[item_id].get("importance") == "substantive" else "implicit_accepted"
|
||||
r.evidence_status, r.reason_code = "none", None
|
||||
r.detail = {**r.detail, "mode": "implicit"}
|
||||
|
||||
|
||||
def short_sid(doc: BidDoc, s: Sentence) -> str:
|
||||
return f"d{doc.no}p{s.page_no}s{s.seq}"
|
||||
|
||||
|
||||
@@ -20,16 +20,18 @@ from sqlalchemy.ext.asyncio import AsyncSession
|
||||
from bid_doctor.ai.models import AiCall
|
||||
from bid_doctor.ai.providers.base import TransientError
|
||||
from bid_doctor.billing import service as billing
|
||||
from bid_doctor.check import findings, match
|
||||
from bid_doctor.check import applicability, findings, match
|
||||
from bid_doctor.check.models import CheckItemResult, CheckRun, CheckRunDocument, Finding, FindingCitation
|
||||
from bid_doctor.check.service import RULE_SET_VERSION, open_fatal
|
||||
from bid_doctor.checklist import edit as checklist_edit
|
||||
from bid_doctor.checklist import response_mode
|
||||
from bid_doctor.checklist import service as checklist
|
||||
from bid_doctor.core import events
|
||||
from bid_doctor.core.db import TenantContext, tenant_tx
|
||||
from bid_doctor.core.errors import AppError
|
||||
from bid_doctor.core.ids import uuid7
|
||||
from bid_doctor.document import citations
|
||||
from bid_doctor.document import service as documents
|
||||
from bid_doctor.document.citations import document_label
|
||||
from bid_doctor.document.models import SourceDocument
|
||||
from bid_doctor.document.pipeline import normalize, text_layer
|
||||
@@ -136,6 +138,23 @@ async def _stages(ctx: TenantContext, run_id: UUID) -> None:
|
||||
snapshot = await checklist_edit.version_items(s, r.checklist_version_id)
|
||||
lots = await projects.lot_labels(s, r.project_id)
|
||||
pending = (await checklist.pending_counts(s, [r.project_id])).get(r.project_id, 0)
|
||||
tenders = (
|
||||
await s.execute(
|
||||
select(SourceDocument).where(
|
||||
SourceDocument.project_id == r.project_id,
|
||||
SourceDocument.role.in_(("tender", "addendum", "clarification")),
|
||||
SourceDocument.text_layer_key.is_not(None),
|
||||
)
|
||||
)
|
||||
).scalars()
|
||||
tender_sents = [
|
||||
x
|
||||
for d in tenders
|
||||
for x in text_layer.body_sentences(
|
||||
[text_layer.Page.from_json(pg) for pg in await documents.text_layer(s, ctx, d)]
|
||||
)
|
||||
]
|
||||
modes = response_mode.detect(tender_sents) # 招标文件按类别的响应方式(C12)
|
||||
bidding = set(r.lot_ids)
|
||||
items, other_lot = {}, 0
|
||||
for item_id, payload in snapshot.items():
|
||||
@@ -144,6 +163,12 @@ async def _stages(ctx: TenantContext, run_id: UUID) -> None:
|
||||
items[item_id] = payload
|
||||
else:
|
||||
other_lot += 1 # 不在所投包里的条目不查,写进覆盖范围(C6)
|
||||
decisions = {i: applicability.decide(p, modes) for i, p in items.items()}
|
||||
skipped: dict[str, int] = {}
|
||||
for d in decisions.values():
|
||||
if not d.match and d.skip:
|
||||
skipped[d.skip] = skipped.get(d.skip, 0) + 1
|
||||
items = {i: p for i, p in items.items() if decisions[i].match} # 不需要投标人回答的条目不逐条找响应
|
||||
last = [PROGRESS["match"]]
|
||||
|
||||
async def progress(ratio: float) -> None:
|
||||
@@ -153,6 +178,7 @@ async def _stages(ctx: TenantContext, run_id: UUID) -> None:
|
||||
await _stage(ctx, run_id, "match", pct)
|
||||
|
||||
results, stats = await match.run(ctx, run_id, items, bid_docs, progress)
|
||||
match.apply_modes(results, items, decisions)
|
||||
|
||||
await _stage(ctx, run_id, "rules") # 第 2 步:报价形式、形式要件、需人工核对
|
||||
await _stage(ctx, run_id, "verify") # 前两道校验在匹配时逐条做完;这里只是进度节点
|
||||
@@ -174,7 +200,7 @@ async def _stages(ctx: TenantContext, run_id: UUID) -> None:
|
||||
)
|
||||
).one()
|
||||
r.counts = await _counts(s, run_id)
|
||||
r.coverage = _coverage(r, pages_info, items, results, pending, other_lot, stats, tokens)
|
||||
r.coverage = _coverage(r, pages_info, items, results, pending, other_lot, stats, tokens, skipped, modes)
|
||||
await _publish(ctx, r, "running", "summarize", PROGRESS["summarize"])
|
||||
|
||||
|
||||
@@ -352,6 +378,8 @@ def _coverage(
|
||||
other_lot: int,
|
||||
stats: match.Stats,
|
||||
tokens: Any,
|
||||
skipped: dict[str, int],
|
||||
modes: dict[str, response_mode.Mode],
|
||||
) -> dict[str, Any]:
|
||||
"""检查覆盖范围:报告首页、查标页顶部都读这里(FR-M3-09、m3-plan §2.5)。"""
|
||||
undetermined = sum(x.verdict == "undetermined" for x in results.values())
|
||||
@@ -367,6 +395,19 @@ def _coverage(
|
||||
not_checked.append(f"{undetermined} 条没有完成判断(模型漏答或暂时不可用),已列为「请人工核对」")
|
||||
if scanned:
|
||||
not_checked.append(f"{scanned} 页没有文字层(扫描件),这些页上的内容没有检查")
|
||||
for reason, n in skipped.items():
|
||||
not_checked.append(f"{applicability.SKIP_LABELS[reason]} {n} 条:不在投标文件里逐条找响应")
|
||||
labels = {"technical": "技术要求", "commercial": "商务要求"}
|
||||
for m in modes.values():
|
||||
if m.status == "unknown":
|
||||
not_checked.append(f"招标文件没有写明{labels[m.category]}是否要求逐条响应,按逐条响应处理")
|
||||
elif m.status == "conflict":
|
||||
not_checked.append(
|
||||
f"招标文件对{labels[m.category]}的响应方式写法不一致(既要求逐条响应,又写明未列出的视为接受),按逐条响应处理"
|
||||
)
|
||||
implicit = sum(x.verdict in ("implicit_accepted", "implicit_review") for x in results.values())
|
||||
if implicit:
|
||||
checked.append(f"其中 {implicit} 条按招标文件「未列入偏离的视为接受」处理,没有复述不算问题")
|
||||
not_checked.append("报价形式、形式要件、资质匹配:规则检查还没有上线")
|
||||
if not r.checklist_confirmed:
|
||||
not_checked.append("清单未确认,检查结果以当前清单为准")
|
||||
@@ -383,6 +424,16 @@ def _coverage(
|
||||
"evidence_unverified": unverified,
|
||||
},
|
||||
"checklist_confirmed": r.checklist_confirmed,
|
||||
"skipped": skipped,
|
||||
"response_modes": [
|
||||
{
|
||||
"category": m.category,
|
||||
"mode": m.mode,
|
||||
"status": m.status,
|
||||
"basis": [{"page": x.page_no, "text": x.text[:120]} for x in (m.implicit + m.explicit)[:4]],
|
||||
}
|
||||
for m in modes.values()
|
||||
],
|
||||
"legal_regime": {"value": r.legal_regime, "rules_verified": r.legal_regime in REGIMES_VERIFIED},
|
||||
"rule_set_version": RULE_SET_VERSION,
|
||||
"prompt_versions": r.prompt_versions,
|
||||
|
||||
63
server/src/bid_doctor/checklist/response_mode.py
Normal file
63
server/src/bid_doctor/checklist/response_mode.py
Normal file
@@ -0,0 +1,63 @@
|
||||
"""招标文件对投标人怎么响应的规定(决策 C12,`evals/check/GUIDELINES.md` §2):按类别(技术 / 商务)识别响应方式。
|
||||
|
||||
- explicit 逐条响应:「逐条 / 逐项作出响应」「逐一逐条响应填写」「逐条标明满足与否」;
|
||||
- implicit 默认接受:「否则将认为投标人接受」「未列入上表的视作完全接受」「除所列偏离外均视作理解和响应」
|
||||
「其余无偏离内容无需赘述」;
|
||||
- 两种写法都有(S07)按逐条处理,记 conflict,问题说明里引两处原文;一种都没有按逐条处理,记 unknown。
|
||||
规则优先、纯函数;结果带原文句子,查标写进检查覆盖范围。
|
||||
TODO(C12):存进清单、在清单页显示并可改(人工确认点);模型兜底规则没认出的写法。
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import re
|
||||
from dataclasses import dataclass, field
|
||||
|
||||
from bid_doctor.document.pipeline.text_layer import Sentence
|
||||
|
||||
EXPLICIT = re.compile(r"逐(条|项|一)[^。;]{0,15}(响应|应答|标明|填写|填列|作出|做出|说明)")
|
||||
IMPLICIT = re.compile(
|
||||
r"(否则|未列入|未列出|未列明|除[^。;]{0,30}外)[^。;]{0,25}(视为|视作|认为)[^。;]{0,15}(接受|响应|理解)"
|
||||
r"|无需赘述"
|
||||
)
|
||||
TABLE_HINT = re.compile(r"(技术|商务)[^。;]{0,12}(偏离|响应)")
|
||||
LOOKBACK = 12 # 往前找几句,看这句说明属于哪张表(技术 / 商务)
|
||||
CATEGORIES = ("technical", "commercial")
|
||||
|
||||
|
||||
@dataclass(slots=True)
|
||||
class Mode:
|
||||
category: str # technical / commercial
|
||||
mode: str # explicit / implicit
|
||||
status: str # found / conflict / unknown
|
||||
explicit: list[Sentence] = field(default_factory=list[Sentence])
|
||||
implicit: list[Sentence] = field(default_factory=list[Sentence])
|
||||
|
||||
|
||||
def _scope(i: int, sentences: list[Sentence]) -> set[str]:
|
||||
"""这句说明管哪一类:句子自己写了「技术 / 商务」就按它,否则看前面几句里最近的表名;都没有就两类都管。"""
|
||||
own = {c for c, k in (("technical", "技术"), ("commercial", "商务")) if k in sentences[i].text}
|
||||
if own:
|
||||
return own
|
||||
for s in reversed(sentences[max(0, i - LOOKBACK) : i]):
|
||||
if m := TABLE_HINT.search(s.text):
|
||||
return {"technical" if m.group(1) == "技术" else "commercial"}
|
||||
return set(CATEGORIES)
|
||||
|
||||
|
||||
def detect(sentences: list[Sentence]) -> dict[str, Mode]:
|
||||
hits: dict[str, Mode] = {c: Mode(c, "explicit", "unknown") for c in CATEGORIES}
|
||||
for i, s in enumerate(sentences):
|
||||
exp, imp = bool(EXPLICIT.search(s.text)), bool(IMPLICIT.search(s.text))
|
||||
if not exp and not imp:
|
||||
continue
|
||||
for c in _scope(i, sentences):
|
||||
(hits[c].implicit if imp else hits[c].explicit).append(s)
|
||||
for m in hits.values():
|
||||
if m.explicit and m.implicit:
|
||||
m.mode, m.status = "explicit", "conflict"
|
||||
elif m.implicit:
|
||||
m.mode, m.status = "implicit", "found"
|
||||
elif m.explicit:
|
||||
m.mode, m.status = "explicit", "found"
|
||||
return hits
|
||||
@@ -77,3 +77,57 @@ def test_no_value_is_a_hint_and_unverified_is_explained() -> None:
|
||||
)
|
||||
assert uv is not None and uv.severity == "fatal" and "第 3 页" in uv.explanation and "没能核对" in uv.explanation
|
||||
assert findings.spec(match.Result("responded", "model", "verified"), item) is None
|
||||
|
||||
|
||||
def test_response_mode_by_category() -> None:
|
||||
from bid_doctor.checklist import response_mode
|
||||
|
||||
sents = [
|
||||
sent(1, "技术规格偏离表"),
|
||||
sent(2, "投标人应将所有偏离的条目列入上表,未列入上表的视作投标人完全接受。"),
|
||||
sent(3, "商务条款偏离表"),
|
||||
sent(4, "投标人须对商务要求逐条作出响应。"),
|
||||
]
|
||||
m = response_mode.detect(sents)
|
||||
assert (m["technical"].mode, m["technical"].status) == ("implicit", "found")
|
||||
assert (m["commercial"].mode, m["commercial"].status) == ("explicit", "found")
|
||||
both = response_mode.detect([sent(1, "须逐项作出响应。"), sent(2, "否则将认为投标人接受招标文件的要求。")])
|
||||
assert (both["technical"].mode, both["technical"].status) == ("explicit", "conflict") # 写法冲突按逐条(C12)
|
||||
none = response_mode.detect([sent(1, "本项目为货物采购。")])
|
||||
assert (none["commercial"].mode, none["commercial"].status) == ("explicit", "unknown")
|
||||
|
||||
|
||||
def test_applicability_skips_items_that_need_no_response() -> None:
|
||||
from bid_doctor.check import applicability
|
||||
from bid_doctor.checklist import response_mode
|
||||
|
||||
modes = response_mode.detect([sent(1, "未列入偏离表的技术要求视作投标人完全接受。")])
|
||||
decide = applicability.decide
|
||||
assert decide({"categories": ["invalidity"], "importance_basis": "tender_text"}, modes).skip == "condition"
|
||||
assert decide({"categories": ["invalidity"], "importance_basis": "marker"}, modes).match # ★ 条款是要求
|
||||
assert decide({"categories": ["schedule"]}, modes).skip == "schedule"
|
||||
assert decide({"categories": ["scoring"]}, modes).skip == "scoring"
|
||||
assert decide({"categories": ["scoring"], "evidence_required": "业绩合同"}, modes).match
|
||||
assert decide({"categories": ["scoring"], "title": "业绩(2 分):提供近三年类似项目合同复印件"}, modes).match
|
||||
assert (
|
||||
decide({"categories": ["scoring"], "title": "提供检测报告,中标后签订合同前提供原件"}, modes).skip == "scoring"
|
||||
)
|
||||
assert decide({"categories": ["tech_param"], "title": "支持硬件BFD,提供官网截图证明"}, modes).evidence
|
||||
assert decide({"categories": ["qualification"], "title": "未被列入信用中国失信名单"}, modes).skip == "no_proof"
|
||||
d = decide({"categories": ["tech_param"], "title": "内存"}, modes)
|
||||
assert d.match and d.mode == "implicit"
|
||||
|
||||
|
||||
def test_implicit_not_found_is_accepted_but_substantive_gets_a_hint() -> None:
|
||||
from uuid import uuid4
|
||||
|
||||
from bid_doctor.check import applicability
|
||||
|
||||
a, b = uuid4(), uuid4()
|
||||
results = {a: match.Result("not_found", "model"), b: match.Result("not_found", "model")}
|
||||
items = {a: {"importance": "normal"}, b: {"importance": "substantive", "title": "★双电源"}}
|
||||
decisions = {a: applicability.Decision(True, mode="implicit"), b: applicability.Decision(True, mode="implicit")}
|
||||
match.apply_modes(results, items, decisions)
|
||||
assert results[a].verdict == "implicit_accepted" and findings.spec(results[a], items[a]) is None
|
||||
spec = findings.spec(results[b], items[b])
|
||||
assert results[b].verdict == "implicit_review" and spec is not None and spec.severity == "hint"
|
||||
|
||||
Reference in New Issue
Block a user