diff --git a/AGENTS.md b/AGENTS.md index 0b8bc4a..71c5e2d 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -214,4 +214,5 @@ samples/ - 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。 - 2026-11-04：查标第三轮：文件要在（按标题找文件名，不调模型）、扫描件兜底（C10，没找到的证明材料改为提示级「可能在扫描件里，请人工核对」）、响应方式存进清单并在清单页显示、可改（C12 的人工确认点，迁移 0007、0008）。开发集上规矩投标文件的致命问题 120 → 51，致命级召回不变（22 / 32）；证明材料缺失仍查不出，要靠 OCR / 多模态（S5）。见 `docs/architecture/check-eval.md` §5。 - 2026-11-04：查标第四轮：规则检查（`check/rules.py`，规则集 2026.11-r1，不调模型）——大小写金额、分项合计、超限价、多个报价、别的项目名称、没填的地方、落款日期、投标有效期、附加条件、响应表里的未响应；文字层的句子带上单元格框，表格按单元格分行。请 Fable 从真实投标文件角度审读，按「认不准的不给致命」改了定级和取值范围。开发集 17 份上规则问题 57 / 57 查出、规矩投标文件 0 误报，但规则是看着生成器和开发词库写的，这个分数说明不了泛化，要等留出集和国动真实投标文件。见 `docs/architecture/check-eval.md` §6。 -- 下一步：查标第 2 步剩下的（需人工核对的条目和 `manual-checks` 接口、「报价」类清单项和分项上限 C7），然后第 3 步的 D09 / D06 提示词 v3。 +- 2026-11-04：查标第 2 步收尾：需人工核对（份数、密封、递交、CA 签章加密、解密、保证金，从清单按主题合并，查标页勾选、按项目存，P2）；「报价」类清单项（C7，读标从清单表规则读出分项 / 单价上限）和规则 `PRICE.OVER_CEILING_ITEM`；评测集派生 4 份报价问题投标文件，21 份上规则问题 62 / 62、误报 0。见 `docs/architecture/check-eval.md` §7。 +- 下一步：查标第 3 步的 D09（响应对象错位）/ D06（负偏离未说明）提示词 v3、token 预算和复查复用；第 2 步还差分项和采购需求逐项对照（P13）。 diff --git a/apps/web/src/lib/keys.ts b/apps/web/src/lib/keys.ts index 1646e2d..ddff6dd 100644 --- a/apps/web/src/lib/keys.ts +++ b/apps/web/src/lib/keys.ts @@ -14,5 +14,6 @@ export const qk = { checkRuns: (projectId: string) => ["check-runs", projectId] as const, checkRun: (id: string) => ["check-run", id] as const, findings: (runId: string) => ["findings", runId] as const, + manualChecks: (projectId: string) => ["manual-checks", projectId] as const, itemResults: (runId: string) => ["item-results", runId] as const, }; diff --git a/apps/web/src/lib/queries.ts b/apps/web/src/lib/queries.ts index f6f6e46..1bc6bfc 100644 --- a/apps/web/src/lib/queries.ts +++ b/apps/web/src/lib/queries.ts @@ -25,6 +25,7 @@ export type CheckPreview = Schemas["CheckPreview"]; export type CheckRun = Schemas["CheckRun"]; export type Finding = Schemas["Finding"]; export type CheckItemResult = Schemas["CheckItemResult"]; +export type ManualCheck = Schemas["ManualCheck"]; export const fetchMe = () => unwrap(api.GET("/me")); @@ -125,6 +126,13 @@ export function useCheckRun(id: MaybeRefOrGetter) { }); } +export function useManualChecks(projectId: MaybeRefOrGetter) { + return useQuery({ + queryKey: computed(() => qk.manualChecks(toValue(projectId))), + queryFn: () => unwrap(api.GET("/projects/{project_id}/manual-checks", { params: { path: { project_id: toValue(projectId) } } })), + }); +} + export function useFindings(runId: MaybeRefOrGetter, enabled: MaybeRefOrGetter) { return useQuery({ queryKey: computed(() => qk.findings(toValue(runId))), diff --git a/apps/web/src/pages/ResultsPage.vue b/apps/web/src/pages/ResultsPage.vue index aaadc76..4b9140d 100644 --- a/apps/web/src/pages/ResultsPage.vue +++ b/apps/web/src/pages/ResultsPage.vue @@ -7,7 +7,8 @@ * 说明框在行内展开；处理后显示署名，可以撤销（U 键在第 6 步）； * - 每个问题带招标原文和投标原文两侧出处，「未找到」写明找过哪些分册；点出处在右侧抽屉里看原文； * - 只有最近一次检查的问题可以处理，历史批次只读。 - * TODO(第 6 步)：键盘操作、批量处理「提示」、判断有误反馈、需人工核对、导出报告。 + * - 需人工核对（FR-M3-04、P2）：份数、密封、递交、CA 加密等文件里看不出来的要求，按项目勾选，写勾选人和时间； + * TODO(第 6 步)：键盘操作、批量处理「提示」、判断有误反馈、导出报告。 */ import { unwrap } from "@bid-doctor/api-client"; import { ChevronDown, ChevronRight, X } from "@lucide/vue"; @@ -21,7 +22,7 @@ import { api, errorTitle, isApiError } from "@/lib/api"; import { useCrumbs } from "@/lib/crumbs"; import { qk } from "@/lib/keys"; import { CHECK_STEPS, checkStepIndex, FINDING_ACTION, fmtDateTime, SEVERITY, SEVERITY_CLASS } from "@/lib/labels"; -import { TERMINAL_JOB, useCheckRun, useFindings, useProject, type Citation, type Finding } from "@/lib/queries"; +import { TERMINAL_JOB, useCheckRun, useFindings, useManualChecks, useProject, type Citation, type Finding, type ManualCheck } from "@/lib/queries"; import { toast } from "@/lib/toast"; const props = defineProps<{ projectId: string; runId: string }>(); @@ -116,6 +117,27 @@ function refresh() { void qc.invalidateQueries({ queryKey: qk.projects }); } +/* ---------- 需人工核对 ---------- */ +const manualQ = useManualChecks(() => props.projectId); +const manual = computed(() => manualQ.data.value?.items ?? []); +const manualDone = computed(() => manual.value.filter((m) => m.checked).length); +const manualBusy = ref(null); +const projectReadOnly = computed(() => p.value?.status === "submitted" || p.value?.status === "archived"); + +async function toggleManual(m: ManualCheck, checked: boolean) { + manualBusy.value = m.check_key; + try { + await unwrap( + api.PUT("/projects/{project_id}/manual-checks", { params: { path: { project_id: props.projectId } }, body: { check_key: m.check_key, checked } }), + ); + void qc.invalidateQueries({ queryKey: qk.manualChecks(props.projectId) }); + } catch (e) { + toast(errorTitle(e), { ok: false }); + } finally { + manualBusy.value = null; + } +} + /* ---------- 原文抽屉 ---------- */ const viewer = ref<{ doc: { id: string; label: string }; target: ViewerTarget & { id: string } } | null>(null); function openCitation(c: Citation) { @@ -262,6 +284,30 @@ const errorText = computed(() => { + +
+
+

需人工核对

+ 文件里看不出来，递交前请自己核对。勾选按项目保存，复查后还在。 + 已核对 {{ manualDone }} / {{ manual.length }} +
+
+
+ +

招标文件：{{ m.requirement_text }}

+
+ {{ m.checked_by?.name }} + {{ fmtDateTime(m.checked_at) }}已核对 +
+
+ +
+
@@ -368,6 +414,21 @@ const errorText = computed(() => { background: var(--c-n100); color: var(--text-primary); } +.manual { + margin-top: var(--sp-4); +} +.manual .card-h { + display: flex; + align-items: baseline; + gap: var(--sp-2); +} +.manual .card-h h2 { + margin: 0; + font-size: var(--fs-body); +} +.manual .card-h .r { + margin-left: auto; +} .sign.ign { color: var(--text-secondary); } diff --git a/docs/architecture/check-eval.md b/docs/architecture/check-eval.md index 6e92deb..a74c04e 100644 --- a/docs/architecture/check-eval.md +++ b/docs/architecture/check-eval.md @@ -176,8 +176,16 @@ **读法：这个满分基本说明不了泛化。** 规则是在知道生成器怎么注入、见过开发词库措辞的情况下写的，投标文件版式也都出自同一个生成器。附加条件（P10）和多个报价（P12）的开发词库措辞直接落在规则的写法里，最需要留出集检验。为此做了两件事：规则按招投标文件的通用写法归纳（单元测试和 `bid-mini-forms.pdf` 都用开发词库以外的措辞）；评测集里没有的情形（超限价、跨页分项表、万元单位）只有单元测试。真实投标文件的合并单元格、扫描页、各地模板差异，要等国动真实投标文件来看。P01 / P02 / P03 / P11 / P13–P18 评测集里还没有注入。 -## 7. 下一步 +## 7. 第 2 步收尾：需人工核对、分项上限（2026-11-04） -1. 第 2 步剩下的：需人工核对的条目（份数、密封、递交方式）和 `manual-checks` 接口；「报价」类清单项和分项 / 单价上限（C7、P03）；评测集补注入 P01、P02、P13。 +**需人工核对**（FR-M3-04、决策 P2，`check/manual.py`）：份数、密封、递交、CA 签章加密、解密、保证金这些投标文件里看不出来的要求，从正式清单里挑出来（技术参数、评分项、资格条件不算；写明免收保证金的不列；点名要交的文件归「文件要在」），按主题合并成最多 5 条（开发集 7 份招标文件每份 3～4 条，原始条目 12～48 条），带招标原文出处；查标页勾选，按项目存、写勾选人和时间，复查后还在。接口 `GET / PUT /projects/{id}/manual-checks`，`check_key` 用 `rule:MANUAL.<主题>`，清单改了勾选也还在。不生成「问题」，覆盖范围里写明列了几条。 + +**分项 / 单价上限**（C7，`checklist/price.py` + 规则 `PRICE.OVER_CEILING_ITEM`）：读标时从采购需求的清单表（表头有「名称」和「单价上限 / 分项限价」两栏，表前一句写采购包）逐行读出名称、数量、上限，作为「报价」类正式条目并入清单，出处是表格那一行，不调模型；采购包一级的表不算。查标时这类条目不交给模型，规则拿分项报价表里同名分项的单价比，超了报致命，挂招标侧出处和清单项。名称要求很严（完全相同、去掉括注后相同，或重合度 ≥ 0.8），对不上的写进覆盖范围。开发集只有 S10 有分项上限，5 条全部读对；散文写法（「单价不得超过 × 元」）还不认。 + +**评测集**：从规矩的投标文件派生 4 份只带报价问题的（`evals/check/price_defects.py`：只改一个分项单价，合价、合计、大小写跟着改，算术一致）：S01-B-p01、S04-B-p01（超总价限价 1 元 / 1%）、S10-A-p02（超采购包限价 1%）、S10-A-p03（分项单价超上限 1%）。S10 的分项上限加起来就是采购包限价，总价超限必然有分项超上限，生成器把这条一并记进金标准。21 份上规则问题 62 / 62 查出、误报 0；同样只说明规则和生成器对得上。 + +## 8. 下一步 + +1. 第 2 步还差：分项报价表和采购需求清单逐项对照（P13）、散文写法的单价上限。 2. 第 3 步：D09（响应对象错位）、D06（负偏离未说明）的提示词 v3，token 预算、复查复用、重复条目分组。 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。 diff --git a/docs/architecture/m3-plan.md b/docs/architecture/m3-plan.md index ed159eb..713c7c8 100644 --- a/docs/architecture/m3-plan.md +++ b/docs/architecture/m3-plan.md @@ -10,7 +10,7 @@ |---|---|---| | 第 0 步 评测集 | 试跑完成，继续扩充 | 规范 v0.2、留出集划分已定，留出集词库和风格 C 已封存；生成、注入、渲染、离线查标、评分工具已写；开发集 7 份招标文件、14 份投标文件第一版跑完，结果和发现见 `check-eval.md` §2 | | 第 1 步 链路打通 | ✅ 2026-11-04 | 见下方「第 1 步落地记录」 | -| 第 2 步 规则 | 进行中 | 响应方式（C12）、文件要在、扫描件兜底、规则检查 `rules.py`（报价形式、形式要件、响应表）已上；需人工核对的条目、「报价」类清单项和分项上限（C7）还没做。见 `check-eval.md` §4–§6 | +| 第 2 步 规则 | 基本完成 | 响应方式（C12）、文件要在、扫描件兜底、规则检查 `rules.py`（报价形式、形式要件、响应表）、需人工核对（`manual.py`、`manual-checks` 接口）、「报价」类清单项和分项上限（C7）。还差分项和采购需求逐项对照（P13）、散文写法的单价上限。见 `check-eval.md` §4–§7 | | 第 3 步 AI 响应匹配 | 进行中 | 整行上下文、提示词 v2；D09 / D06、token 预算、复查复用还没做 | | 第 4～6 步 | 未开始 | | diff --git a/docs/architecture/openapi.yaml b/docs/architecture/openapi.yaml index ce124f8..806f177 100644 --- a/docs/architecture/openapi.yaml +++ b/docs/architecture/openapi.yaml @@ -2939,9 +2939,18 @@ components: type: object required: [check_key, label, checked] properties: - check_key: { type: string } - label: { type: string, examples: [正副本份数与装订] } - requirement_text: { type: [string, 'null'] } + check_key: { type: string, description: "rule:MANUAL.<主题>：SUBMIT、SIGN_ENCRYPT、DECRYPT、PAPER、DEPOSIT" } + label: { type: string, examples: [纸质投标文件的份数、装订、密封] } + requirement_text: { type: [string, 'null'], description: 代表这一主题的清单条目标题 } + related_items: + type: array + description: 同一主题的清单条目（公告、须知、无效情形里常各写一遍，按主题合并成一条核对） + items: + type: object + required: [id, title] + properties: + id: { $ref: '#/components/schemas/Id' } + title: { type: string } citation: oneOf: - $ref: '#/components/schemas/Citation' diff --git a/evals/check/price_defects.py b/evals/check/price_defects.py new file mode 100644 index 0000000..0bf7962 --- /dev/null +++ b/evals/check/price_defects.py @@ -0,0 +1,118 @@ +"""从规矩的投标文件派生只带报价问题的投标文件（P01 超总价限价、P02 超采购包限价、P03 分项超单价上限）。 + + cd server && ../.venv/bin/python ../evals/check/price_defects.py <规矩的投标文件> <输出> [--variant ...] + 然后 ../.venv/bin/python ../evals/check/render.py <输出> + +- 只改一个分项的单价，再重算该行合价、合计、开标一览表的小写和大写，算术前后一致：只触发要测的那一条规则； +- 响应清单照抄规矩的那份（单元都是 expect=ok），rule_defects 记这一个问题；不挂标注单元，不需要复核（E1 只管单元）； +- 限价、单价上限取招标档案（evals/out/check/tenders/<招标文件>.json），和读标抽出的值一致时才有意义，生成后自己核对一遍。 +""" + +from __future__ import annotations + +import argparse +import json +import pathlib +import sys +from decimal import ROUND_UP, Decimal +from typing import Any + +ROOT = pathlib.Path(__file__).resolve().parent +OUT = ROOT.parent / "out" / "check" +sys.path.insert(0, str(ROOT)) +from build import rmb_upper # noqa: E402 + +VARIANTS = {"P01": ["超 1 元", "超 1%"], "P02": ["超 1 元", "超 1%"], "P03": ["单价超 1%"]} + + +def fmt(v: Decimal) -> str: + return f"{v:,.2f}" + + +def blocks(spec: dict[str, Any]) -> list[dict[str, Any]]: + return [b for v in spec["volumes"] for b in v["blocks"]] + + +def reprice(spec: dict[str, Any], index: int, unit_price: Decimal) -> Decimal: + """把第 index 个分项的单价改成 unit_price，重算合价、合计、开标一览表；正文里出现的旧总价一并替换。返回新总价。""" + items = spec["price"]["items"] + old_total = Decimal(spec["price"]["total"]) + items[index]["unit_price"] = f"{unit_price:.2f}" + lines = [Decimal(str(i["qty"])) * Decimal(i["unit_price"]) for i in items] + total = sum(lines, Decimal(0)).quantize(Decimal("0.01")) + spec["price"]["total"] = f"{total:.2f}" + table = next(b for b in blocks(spec) if b.get("price_items")) + row = table["rows"][index] + row["cells"][-2], row["cells"][-1] = fmt(unit_price), fmt(lines[index].quantize(Decimal("0.01"))) + pairs = [(fmt(old_total), fmt(total)), (rmb_upper(old_total), rmb_upper(total))] + for b in blocks(spec): + if b.get("type") == "table": + for r in b["rows"]: + r["cells"] = [_swap(c, pairs) for c in r["cells"]] + elif "text" in b: + b["text"] = _swap(b["text"], pairs) + if "lines" in b: + b["lines"] = [_swap(x, pairs) for x in b["lines"]] + return total + + +def _swap(text: str, pairs: list[tuple[str, str]]) -> str: + for a, b in pairs: + text = text.replace(a, b) + return text + + +def main() -> None: + ap = argparse.ArgumentParser() + ap.add_argument("clean") + ap.add_argument("out") + ap.add_argument("kind", choices=sorted(VARIANTS)) + ap.add_argument("--variant") + a = ap.parse_args() + variant = a.variant or VARIANTS[a.kind][0] + src = OUT / a.clean + spec = json.loads((src / "spec.json").read_text()) + manifest = json.loads((src / "manifest.json").read_text()) + profile = json.loads((OUT / "tenders" / f"{spec['tender']}.json").read_text()) + items = spec["price"]["items"] + total = Decimal(spec["price"]["total"]) + forms = [f for f in profile["forms"] if f.get("kind") == "price_items"] + caps = {x["name"]: Decimal(str(x["ceiling"])) for x in (forms[0].get("items") or []) if x.get("ceiling")} + if a.kind in ("P01", "P02"): + ceiling = Decimal(str(spec["ceiling_total"])) + target = ceiling + 1 if variant == "超 1 元" else (ceiling * Decimal("1.01")).quantize(Decimal("0.01")) + i = next((n for n, x in enumerate(items) if Decimal(str(x["qty"])) == 1), 0) + qty = Decimal(str(items[i]["qty"])) + unit = Decimal(items[i]["unit_price"]) + ((target - total) / qty).quantize(Decimal("0.01"), rounding=ROUND_UP) + new_total = reprice(spec, i, unit) + assert new_total > ceiling, (new_total, ceiling) + snippet = fmt(new_total) + else: + i = next(n for n, x in enumerate(items) if x["name"] in caps) + unit = (caps[items[i]["name"]] * Decimal("1.01")).quantize(Decimal("0.01")) + new_total = reprice(spec, i, unit) + snippet = fmt(unit) + out = OUT / a.out + out.mkdir(parents=True, exist_ok=True) + spec["bid"] = a.out + manifest.update( + bid=a.out, + kind="sloppy", + density="price", + rule_defects=[ + {"id": f"{a.kind}-1", "type": a.kind, "variant": variant, "t": snippet}, + # 分项上限加起来常常就是限价：总价超限价时，被抬高的分项也超了上限，一并记 + *( + [{"id": "P03-2", "type": "P03", "variant": "随总价超限", "t": fmt(Decimal(x["unit_price"]))}] + if a.kind != "P03" and (x := items[i])["name"] in caps and Decimal(x["unit_price"]) > caps[x["name"]] + else [] + ), + ], + ) + (out / "spec.json").write_text(json.dumps(spec, ensure_ascii=False, indent=1)) + (out / "manifest.json").write_text(json.dumps(manifest, ensure_ascii=False, indent=1)) + print(f"{a.out}：{a.kind}（{variant}），总价 {fmt(total)} → {fmt(new_total)}，原文片段「{snippet}」") + + +if __name__ == "__main__": + main() diff --git a/evals/check/run_check.py b/evals/check/run_check.py index 7a609e6..5c8acc0 100644 --- a/evals/check/run_check.py +++ b/evals/check/run_check.py @@ -27,6 +27,7 @@ from bid_doctor.ai.providers.base import near_limit from bid_doctor.ai.service import OutputIncomplete, provider from bid_doctor.check import applicability, findings, match, rules from bid_doctor.checklist import response_mode +from bid_doctor.checklist import price from bid_doctor.checklist.extract import amount_yuan from bid_doctor.core.config import get_settings from bid_doctor.core.db import TenantContext @@ -153,6 +154,12 @@ def facts(tender: str, run: str, items: dict[UUID, dict[str, Any]]) -> rules.Fac c = amount_yuan(f.get("ceiling_total"), info["fields"].get("ceiling_total", {}).get("unit")) ceiling = Decimal(c) if c else None days, item = rules.required_validity(items) + lot_no = str(profile.get("bidding_lot") or "").lstrip("0") + caps = [ + rules.ItemCap(c.name, c.cap, c.per) + for c in price.caps(gold.sentences(tender)) + if not c.lot_no or not lot_no or c.lot_no.lstrip("0") == lot_no + ] return rules.Facts( name=f.get("name"), code=f.get("code"), @@ -164,6 +171,7 @@ def facts(tender: str, run: str, items: dict[UUID, dict[str, Any]]) -> rules.Fac ceiling_note=note, validity_days=days, validity_item=item, + caps=caps, ) diff --git a/packages/api-client/src/schema.d.ts b/packages/api-client/src/schema.d.ts index c2abd7c..93574d9 100644 --- a/packages/api-client/src/schema.d.ts +++ b/packages/api-client/src/schema.d.ts @@ -2127,9 +2127,16 @@ export interface components { reused: boolean; }; ManualCheck: { + /** @description rule:MANUAL.<主题>：SUBMIT、SIGN_ENCRYPT、DECRYPT、PAPER、DEPOSIT */ check_key: string; label: string; + /** @description 代表这一主题的清单条目标题 */ requirement_text?: string | null; + /** @description 同一主题的清单条目（公告、须知、无效情形里常各写一遍，按主题合并成一条核对） */ + related_items?: { + id: components["schemas"]["Id"]; + title: string; + }[]; citation?: components["schemas"]["Citation"] | null; checked: boolean; note?: string | null; diff --git a/server/src/bid_doctor/check/api.py b/server/src/bid_doctor/check/api.py index d20e125..668211b 100644 --- a/server/src/bid_doctor/check/api.py +++ b/server/src/bid_doctor/check/api.py @@ -24,6 +24,12 @@ class ResolveIn(BaseModel): note: str | None = Field(default=None, max_length=500) +class ManualCheckIn(BaseModel): + check_key: str = Field(min_length=1, max_length=200) + checked: bool + note: str | None = Field(default=None, max_length=500) + + class BulkResolveIn(BaseModel): finding_ids: list[UUID] = Field(min_length=1, max_length=500) status: Literal["confirmed_ok", "ignored"] @@ -96,3 +102,15 @@ async def resolve_finding(finding_id: UUID, body: ResolveIn, ctx: Ctx) -> dict[s async def reopen_finding(finding_id: UUID, ctx: Ctx) -> dict[str, Any]: async with tenant_tx(ctx) as s: return await service.reopen(s, ctx, finding_id) + + +@router.get("/projects/{project_id}/manual-checks") +async def list_manual_checks(project_id: UUID, ctx: Ctx) -> dict[str, Any]: + async with tenant_tx(ctx) as s: + return {"items": await service.manual_checks(s, ctx, project_id)} + + +@router.put("/projects/{project_id}/manual-checks") +async def set_manual_check(project_id: UUID, body: ManualCheckIn, ctx: Ctx) -> dict[str, Any]: + async with tenant_tx(ctx) as s: + return await service.set_manual_check(s, ctx, project_id, body.check_key, body.checked, body.note) diff --git a/server/src/bid_doctor/check/applicability.py b/server/src/bid_doctor/check/applicability.py index e4d4147..31d8413 100644 --- a/server/src/bid_doctor/check/applicability.py +++ b/server/src/bid_doctor/check/applicability.py @@ -26,6 +26,7 @@ SKIP_LABELS = { "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）", + "price": "报价类要求（分项、单价上限）：由规则比对分项报价表", } @@ -43,6 +44,8 @@ def decide(item: dict[str, Any], modes: dict[str, Mode]) -> Decision: 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 cats == {"price"}: + return Decision(False, skip="price") 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) diff --git a/server/src/bid_doctor/check/manual.py b/server/src/bid_doctor/check/manual.py new file mode 100644 index 0000000..067eb93 --- /dev/null +++ b/server/src/bid_doctor/check/manual.py @@ -0,0 +1,76 @@ +"""需人工核对的条目（FR-M3-04、决策 P2）：份数、密封、递交、CA 签章加密、解密、保证金——投标文件里看不出来， +列出来请用户递交前自己核对；勾选按项目存在 manual_check，复查沿用，报告写勾选人和时间。 + +- 条目只从正式清单里来（带招标原文出处），不凭空生成提醒；技术参数、评分项、资格条件不算； +- 同一主题在清单里常有好几条（公告、须知、无效情形各写一遍），按主题合并成一条，check_key = rule:MANUAL.<主题>， + 清单改了、条目编号变了，勾选也还在； +- 招标文件写明「免收保证金」的，保证金这条不列。 +""" + +from __future__ import annotations + +import re +from dataclasses import dataclass, field +from typing import Any + +TOPICS: tuple[tuple[str, str, re.Pattern[str]], ...] = ( + ("DECRYPT", "开标时在规定时间内解密投标文件", re.compile(r"解密")), + ("SIGN_ENCRYPT", "用 CA 数字证书加盖电子签章、加密投标文件", re.compile(r"CA|电子签章|电子公章|数字证书|加密")), + ("PAPER", "纸质投标文件的份数、装订、密封", re.compile(r"正本|副本|份数|装订|密封|封装|封套|纸质|U盘|光盘")), + ("DEPOSIT", "投标保证金按要求缴纳", re.compile(r"保证金")), + ( + "SUBMIT", + "在投标截止时间前递交（上传）投标文件", + re.compile( + r"(上传|递交|提交|送达).{0,12}(投标|响应|磋商)文件|(投标|响应|磋商)文件.{0,12}(上传|递交|提交|送达)|递交截止" + ), + ), +) +SKIP_CATEGORIES = {"tech_param", "scoring", "qualification"} +NOT_NEEDED = re.compile(r"免收|免交|不收取|无需|无须|不需|不交纳") +NOT_MANUAL = re.compile(r"相似|雷同|同一单位|同一账户|《") # 围串标情形；点名要交的文件（由「文件要在」检查） +PREFERRED = {"SIGN_ENCRYPT": re.compile(r"CA|加密")} # 同一主题里拿哪条做代表 + + +@dataclass(slots=True) +class Topic: + key: str + label: str + items: list[dict[str, Any]] = field(default_factory=list[dict[str, Any]]) + + @property + def check_key(self) -> str: + return f"rule:MANUAL.{self.key}" + + +def topic_of(item: dict[str, Any]) -> str | None: + cats = set(item.get("categories") or []) + title = str(item.get("title") or "") + if cats & SKIP_CATEGORIES or NOT_MANUAL.search(title): + return None + for key, _, pattern in TOPICS: + if pattern.search(title): + return None if key == "DEPOSIT" and NOT_NEEDED.search(title) else key + return None + + +def topics(items: list[dict[str, Any]]) -> list[Topic]: + """按主题合并的需人工核对条目，按 TOPICS 的顺序；每个主题里写明截止时间的、写着主题核心词的条目排前面（做代表）。""" + found: dict[str, Topic] = {} + for item in items: + key = topic_of(item) + if key is None: + continue + label = next(lbl for k, lbl, _ in TOPICS if k == key) + found.setdefault(key, Topic(key, label)).items.append(item) + order = [k for k, _, _ in TOPICS] + out = sorted(found.values(), key=lambda t: order.index(t.key)) + for t in out: + pref = PREFERRED.get(t.key) + t.items.sort( + key=lambda i: ( + not ((i.get("spec") or {}).get("schedule") or {}).get("at"), + bool(pref) and not pref.search(str(i.get("title") or "")), + ) + ) + return out diff --git a/server/src/bid_doctor/check/rules.py b/server/src/bid_doctor/check/rules.py index 8cd6bf3..235550a 100644 --- a/server/src/bid_doctor/check/rules.py +++ b/server/src/bid_doctor/check/rules.py @@ -22,11 +22,16 @@ from typing import Any from uuid import UUID from bid_doctor.check.match import BidDoc +from bid_doctor.document.pipeline import layout +from bid_doctor.document.pipeline.layout import column as _col +from bid_doctor.document.pipeline.layout import rows as rows_of +from bid_doctor.document.pipeline.layout import span as _span from bid_doctor.document.pipeline.text_layer import Sentence, norm LABELS = { "PRICE.OVER_CEILING_TOTAL": "报价和最高限价比对", "PRICE.OVER_CEILING_LOT": "报价和所投采购包的最高限价比对", + "PRICE.OVER_CEILING_ITEM": "分项报价和分项 / 单价上限比对", "PRICE.CAPITAL_MISMATCH": "大写金额和小写金额核对", "PRICE.SUM_MISMATCH": "分项报价合计核对", "PRICE.TOTAL_INCONSISTENT": "投标总价前后一致", @@ -39,7 +44,7 @@ LABELS = { "FORM.RESPONSE_NEGATIVE": "响应表里的未响应、负偏离", } RULES_NOT_YET = ( - "分项报价表和采购需求清单逐项对照、分项 / 单价上限、格式文件固定文字被改、" + "分项报价表和采购需求清单逐项对照、格式文件固定文字被改、" "授权代表和签字人一致、声明函填报、签章有无、低价说明、跨页的响应表" ) @@ -58,9 +63,33 @@ class Facts: ceiling_note: str | None = None # 没有可比的限价时的原因（写进覆盖范围） validity_days: int | None = None validity_item: UUID | None = None + caps: list[ItemCap] = field(default_factory=list["ItemCap"]) # 所投采购包的分项 / 单价上限（「报价」类清单项） refs: dict[str, list[UUID]] = field(default_factory=dict[str, list[UUID]]) +@dataclass(frozen=True, slots=True) +class ItemCap: + """分项上限（checklist/price.py 抽出的「报价」类清单项）：per = unit 按单价比，item 按分项总价比。""" + + name: str + cap: Decimal + per: str + item: UUID | None = None + + +def caps_of(items: dict[UUID, dict[str, Any]]) -> list[ItemCap]: + out = [] + for item_id, p in items.items(): + spec = (p.get("spec") or {}).get("price") or {} + if spec.get("kind") in ("unit_cap", "item_cap") and spec.get("name") and spec.get("cap"): + try: + cap = Decimal(str(spec["cap"])) + except InvalidOperation: + continue + out.append(ItemCap(str(spec["name"]), cap, "unit" if spec["kind"] == "unit_cap" else "item", item_id)) + return out + + @dataclass(slots=True) class Hit: code: str @@ -164,55 +193,9 @@ def yuan(v: Decimal) -> str: return f"{v:,.2f} 元" -# ---------- 版面：页、行、单元格 ---------- -def _box(s: Sentence) -> tuple[float, float, float, float]: - return ( - min(q[0] for q in s.quads), - min(q[1] for q in s.quads), - max(q[2] for q in s.quads), - max(q[3] for q in s.quads), - ) - - +# ---------- 版面：页、行、单元格（document/pipeline/layout.py） ---------- def pages_of(doc: BidDoc) -> dict[int, list[Sentence]]: - out: dict[int, list[Sentence]] = {} - for s in doc.sentences: - out.setdefault(s.page_no, []).append(s) - return out - - -@dataclass(slots=True) -class _Row: - top: float - bottom: float - in_cell: bool - items: list[Sentence] - - -def rows_of(page: list[Sentence]) -> list[list[Sentence]]: - """把一页的句子分成行，行内从左到右。有框线的表格按单元格的上沿分行：同一行的格子上沿相同， - 垂直居中的短格、折成几句的长格、纵向合并的格（只算进它的第一行）都不会串行；其余按上下范围重叠。""" - rows: list[_Row] = [] - - def top_of(s: Sentence) -> float: - return s.cell[1] if s.cell else _box(s)[1] - - for s in sorted((x for x in page if x.quads), key=top_of): - if s.cell: - r = next((r for r in rows if r.in_cell and abs(r.top - s.cell[1]) < 2), None) - if r: - r.items.append(s) - else: - rows.append(_Row(s.cell[1], s.cell[1], True, [s])) - continue - _, top, _, bottom = _box(s) - last = rows[-1] if rows else None - if last and not last.in_cell and top < last.bottom - 1: - last.bottom = max(last.bottom, bottom) - last.items.append(s) - else: - rows.append(_Row(top, bottom, False, [s])) - return [sorted(r.items, key=lambda x: _box(x)[0]) for r in sorted(rows, key=lambda r: r.top)] + return layout.pages(doc.sentences) def row_index(doc: BidDoc) -> dict[int, list[Sentence]]: @@ -220,25 +203,6 @@ def row_index(doc: BidDoc) -> dict[int, list[Sentence]]: return {id(c): row for page in pages_of(doc).values() for row in rows_of(page) for c in row} -def _col(cell: Sentence, cols: dict[str, tuple[float, float]]) -> str | None: - """单元格落在哪一栏：和表头单元格左右范围重叠最多的那一栏；都不重叠（窄数字居中、表头字靠左）取中心最近的。 - cols 要包含表头的每一格，否则别的栏的数字会被归到最近的已知栏。""" - x0, x1 = _span(cell) - best, score = None, 0.0 - for name, (c0, c1) in cols.items(): - overlap = min(x1, c1) - max(x0, c0) - if overlap > score: - best, score = name, overlap - if best is None and cols: - mid = (x0 + x1) / 2 - best = min(cols, key=lambda n: abs((cols[n][0] + cols[n][1]) / 2 - mid)) - return best - - -def _span(c: Sentence) -> tuple[float, float]: - return (c.cell[0], c.cell[2]) if c.cell else (_box(c)[0], _box(c)[2]) - - # ---------- 分项报价表、开标一览表 ---------- @dataclass(slots=True) class ItemRow: @@ -507,6 +471,66 @@ def check_prices( ) +PAREN = re.compile(r"[（(][^）)]*[）)]") + + +def _match_cap(name: str, caps: list[ItemCap]) -> ItemCap | None: + """分项名称对上哪条上限。投标文件的分项名称通常照抄采购需求，所以要求很严：完全相同（或去掉括注后相同）优先； + 否则字对重合度不低于 0.8，或一方包含另一方且长度比不低于 0.8。「服务器」不会对上「服务器机柜」。""" + n = norm(name) + for c in caps: + if norm(c.name) == n or PAREN.sub("", norm(c.name)) == PAREN.sub("", n): + return c + best, score = None, 0.0 + for c in caps: + k = norm(c.name) + x, y = _bigrams(k), _bigrams(n) + dice = 2 * len(x & y) / (len(x) + len(y)) if x and y else 0.0 + contained = (k in n or n in k) and min(len(k), len(n)) / max(len(k), len(n)) >= 0.8 + s = max(dice, 0.8 if contained else 0.0) + if s >= 0.8 and s > score: + best, score = c, s + return best + + +def check_item_caps(facts: Facts, tables: list[PriceTable], hits: list[Hit], notes: list[str]) -> None: + """分项 / 单价上限（第三层限价，FR-M3-06）：分项报价表里同名分项的单价（或分项总价）超过招标文件写的上限。""" + if not facts.caps: + return + if not tables: + notes.append(f"招标文件写了分项 / 单价上限 {len(facts.caps)} 条，投标文件里没认出分项报价表，没有比对") + return + matched = 0 + for t in tables: + for r in t.rows: + name = next((c.text for c in r.cells if not money(c.text, plain=True) and len(norm(c.text)) >= 2), "") + cap = _match_cap(name, facts.caps) if name else None + value = None if cap is None else r.unit_price if cap.per == "unit" else r.line + if cap is None or value is None or r.ambiguous: + continue + matched += 1 + if value <= cap.cap: + continue + what = "单价" if cap.per == "unit" else "分项总价" + hits.append( + Hit( + "PRICE.OVER_CEILING_ITEM", + "price_form", + "fatal", + f"「{name[:20]}」的{what}超过上限", + f"分项报价表里「{name[:30]}」的{what}是 {yuan(value)}，招标文件写的{what}上限是 {yuan(cap.cap)}。" + "超过分项或单价上限通常会导致投标无效，请核对。", + [(t.doc, c) for c in r.cells if money(c.text, plain=True) or c.text == name], + target(name), + basis=f"cap:{cap.item}", + item=cap.item, + detail={"value": str(value), "cap": str(cap.cap), "per": cap.per}, + ) + ) + if matched < len(facts.caps): + notes.append(f"分项 / 单价上限 {len(facts.caps)} 条，在分项报价表里对上名称的 {matched} 条，其余没有比对") + + LOWER_MARK = re.compile(r"小写|[¥￥]|报价|总价|合计|金额|价款") @@ -1065,6 +1089,7 @@ def run(facts: Facts, docs: list[BidDoc], tender_sentences: list[Sentence]) -> t hits: list[Hit] = [] notes: list[str] = [] check_prices(facts, docs, tables, totals, hits, notes) + check_item_caps(facts, tables, hits, notes) check_quotes(docs, tender, tables, totals, hits) check_foreign(facts, docs, tender, hits, notes) check_blanks(docs, hits, notes) diff --git a/server/src/bid_doctor/check/runner.py b/server/src/bid_doctor/check/runner.py index 846ad40..8c0b47b 100644 --- a/server/src/bid_doctor/check/runner.py +++ b/server/src/bid_doctor/check/runner.py @@ -22,7 +22,7 @@ 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 applicability, findings, match, rules +from bid_doctor.check import applicability, findings, manual, match, rules 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 @@ -207,6 +207,13 @@ async def _stages(ctx: TenantContext, run_id: UUID) -> None: for d, x in res.evidence } hits = rules.drop_covered(hits, covered) + n_manual = len(manual.topics(list(snapshot.values()))) + unseen = "份数、密封、递交、CA 签章加密、解密、保证金等文件里看不出来的要求" + rule_cov["not_checked"].append( + f"{unseen}：列入「需人工核对」{n_manual} 条，请递交前自己核对" + if n_manual + else f"{unseen}：清单里没有，没有列「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 await _stage(ctx, run_id, "carry_over") async with tenant_tx(ctx) as s: @@ -262,6 +269,13 @@ async def _facts(s: AsyncSession, ctx: TenantContext, r: CheckRun, snapshot: dic days, item = rules.required_validity(snapshot) if item is not None: refs["validity"] = [UUID(c) for c in snapshot[item].get("citation_ids") or []] + bidding_ids = {x.id for x in bidding} + in_scope = { + i: x for i, x in snapshot.items() if not x.get("lot_id") or not bidding_ids or UUID(x["lot_id"]) in bidding_ids + } + caps = rules.caps_of(in_scope) + for c in caps: + refs[f"cap:{c.item}"] = [UUID(x) for x in snapshot[c.item].get("citation_ids") or []] if c.item else [] refs["deadline"] = refs.pop("deadline_at", []) refs["ceiling"] = refs.pop("ceiling_total", []) return rules.Facts( @@ -275,6 +289,7 @@ async def _facts(s: AsyncSession, ctx: TenantContext, r: CheckRun, snapshot: dic ceiling_note=note, validity_days=days, validity_item=item, + caps=caps, refs=refs, ) diff --git a/server/src/bid_doctor/check/service.py b/server/src/bid_doctor/check/service.py index 693eba5..ec384f6 100644 --- a/server/src/bid_doctor/check/service.py +++ b/server/src/bid_doctor/check/service.py @@ -19,8 +19,8 @@ from sqlalchemy.ext.asyncio import AsyncSession from bid_doctor.audit import service as audit from bid_doctor.billing import service as billing -from bid_doctor.check import rules -from bid_doctor.check.models import CheckItemResult, CheckRun, CheckRunDocument, Finding, FindingCitation +from bid_doctor.check import manual, rules +from bid_doctor.check.models import CheckItemResult, CheckRun, CheckRunDocument, Finding, FindingCitation, ManualCheck from bid_doctor.checklist import edit as checklist_edit from bid_doctor.checklist import service as checklist from bid_doctor.core import queue @@ -36,9 +36,7 @@ from bid_doctor.project import service as projects from bid_doctor.project.models import Project CHECK_TASK = "check_project" -RULE_SET_VERSION = ( - "2026.11-r1" # 规则集有版本，检查批次记下当时用的版本（架构 §3.7）；r1：报价形式、形式要件（rules.py） -) +RULE_SET_VERSION = "2026.11-r2" # 规则集有版本，检查批次记下当时用的版本（架构 §3.7）；r2：加分项 / 单价上限 WAIVE_WINDOW = timedelta(hours=24) ERROR_MESSAGES = { "bid_text_missing": "投标文件里几乎读不出文字（可能都是扫描件），扫描件识别还没有上线", @@ -461,3 +459,66 @@ async def item_results(s: AsyncSession, run_id: UUID) -> list[dict[str, Any]]: } for r in rows ] + + +async def manual_checks(s: AsyncSession, ctx: TenantContext, project_id: UUID) -> list[dict[str, Any]]: + """需人工核对的条目（按主题合并，来自当前正式清单）和勾选结果（按项目存，复查沿用，P2）。""" + await projects.get(s, ctx, project_id) + topics = manual.topics(await checklist.items(s, project_id, zone="formal")) + rows = { + r.check_key: r + for r in (await s.execute(select(ManualCheck).where(ManualCheck.project_id == project_id))).scalars() + } + names = await identity.display_names(s, [r.checked_by for r in rows.values() if r.checked]) + out = [] + for t in topics: + first = t.items[0] + r = rows.get(t.check_key) + done = bool(r and r.checked) + out.append( + { + "check_key": t.check_key, + "label": t.label, + "requirement_text": first.get("title"), + "citation": (first.get("citations") or [None])[0], + "related_items": [{"id": i["id"], "title": i["title"]} for i in t.items], + "checked": done, + "note": r.note if r else None, + "checked_by": {"id": r.checked_by, "name": names.get(r.checked_by, "")} if r and done else None, + "checked_at": r.checked_at if r and done else None, + } + ) + return out + + +async def set_manual_check( + s: AsyncSession, ctx: TenantContext, project_id: UUID, check_key: str, checked: bool, note: str | None +) -> dict[str, Any]: + p = await projects.get(s, ctx, project_id) + projects.ensure_editable(p) + current = {x["check_key"]: x for x in await manual_checks(s, ctx, project_id)} + if check_key not in current: + raise AppError("validation_failed", errors=[{"field": "check_key", "code": "unknown", "message": "没有这一条"}]) + r = ( + await s.execute( + select(ManualCheck).where(ManualCheck.project_id == project_id, ManualCheck.check_key == check_key) + ) + ).scalar_one_or_none() + before = {"checked": r.checked, "note": r.note} if r else None + note = (note or "").strip() or None + if r is None: + r = ManualCheck(tenant_id=ctx.tenant_id, project_id=project_id, check_key=check_key, checked_by=ctx.user_id) + s.add(r) + r.checked, r.note, r.checked_by, r.checked_at = checked, note, ctx.user_id, datetime.now(UTC) + await audit.record( + s, + ctx, + "manual_check.set", + "project", + project_id, + project_id=project_id, + before=before, + after={"check_key": check_key, "checked": checked, "note": note}, + ) + await s.flush() + return next(x for x in await manual_checks(s, ctx, project_id) if x["check_key"] == check_key) diff --git a/server/src/bid_doctor/checklist/extract.py b/server/src/bid_doctor/checklist/extract.py index 0a2ac29..15e0be1 100644 --- a/server/src/bid_doctor/checklist/extract.py +++ b/server/src/bid_doctor/checklist/extract.py @@ -25,6 +25,7 @@ from uuid import UUID from zoneinfo import ZoneInfo from bid_doctor.ai import service as ai +from bid_doctor.checklist import price from bid_doctor.core.db import TenantContext from bid_doctor.document.pipeline import structure, text_layer from bid_doctor.document.pipeline.text_layer import Sentence, norm @@ -845,6 +846,7 @@ async def run( info = _project_info([pair for got in results[len(jobs) :] for pair in got]) items = merge_items(out) items += fallback(items, doc) + items += price_items(items, doc) for e in items: if e.failures: e.ai_values = {**e.ai_values, "failures": e.failures} @@ -967,6 +969,28 @@ def add_pending_categories(e: Extracted, categories: list[str], failures: list[s e.ai_values = {**e.ai_values, "pending_categories": pending} +def price_items(items: list[Extracted], doc: Doc) -> list[Extracted]: + """「报价」类清单项（C7）：分项限价、单价上限，规则从清单表里读出，不调模型（checklist/price.py）。""" + used = {e.sentences[0].key for e in items if e.sentences} + out = [] + for c in price.caps(doc.sentences): + if c.sentences[0].key in used: + continue + out.append( + Extracted( + task="rule", + categories=["price"], + summary=c.title, + sentences=c.sentences, + zone="formal", + checks_passed=["id", "quote"], + spec=c.spec(), + lot_no=c.lot_no, + ) + ) + return out + + def fallback(items: list[Extracted], doc: Doc) -> list[Extracted]: """规则兜底：标注句没被带「无效」类别的正式条目覆盖 → 生成一条待确认（TV1：标注类召回 ≥ 99%）。 跨页接续的两句算一条。这句已经是别的正式条目（资格、参数……）的首句时，不另生成条目（同一首句只能有一条）， diff --git a/server/src/bid_doctor/checklist/price.py b/server/src/bid_doctor/checklist/price.py new file mode 100644 index 0000000..ded7002 --- /dev/null +++ b/server/src/bid_doctor/checklist/price.py @@ -0,0 +1,129 @@ +"""「报价」类清单项：分项限价、单价上限（决策 C7，FR-M3-06 的第三层限价）。规则抽取，不调模型。 + +招标文件里分项上限几乎都写在采购需求的清单表里：表头有「名称」一栏和「单价上限 / 分项限价 / 最高限价」一栏， +表前一句写着是哪个采购包（「采购包2：……」）。按表头归栏逐行读出名称、数量、上限，每行一条清单项， +出处是表格那一行。总价、采购包两层限价在项目基本信息里，不在这里。 + +- 只认表格（版面工具按单元格分行）；散文里的「单价不得超过 × 元」写法先不认，用户可以在清单里手工新增； +- 表头写「万元」的按万元折算；采购包一级的表（名称栏写「采购包 / 标段」）不算分项； +- 数字只来自招标文件原文（红线 6）。 +""" + +from __future__ import annotations + +import re +from dataclasses import dataclass +from decimal import Decimal, InvalidOperation + +from bid_doctor.document.pipeline import layout +from bid_doctor.document.pipeline.text_layer import Sentence, norm + +CAP_HEAD = re.compile(r"(单价上限|单价最高限价|最高单价|分项限价|分项最高限价|限价|最高限价|单价控制价|控制价)") +NAME_HEAD = re.compile(r"名称|内容|品目|标的") +QTY_HEAD = re.compile(r"^数量") +LOT_HEAD = re.compile(r"采购包|标段|包号") +LOT = re.compile(r"(?:采购包|标段|包)\s*([0-9一二三四五六七八九十]+)") +NUMBER = re.compile(r"\d{1,3}(?:,\d{3})+(?:\.\d+)?|\d+(?:\.\d+)?") +CN_DIGIT = {c: str(i) for i, c in enumerate("零一二三四五六七八九")} + + +@dataclass(slots=True) +class Cap: + """一条分项上限：名称、上限（元）、按单价还是按分项总价、数量、所属采购包、出处句子。""" + + name: str + cap: Decimal + per: str # unit：单价上限；item：分项总价上限 + qty: Decimal | None + lot_no: str | None + sentences: list[Sentence] + + @property + def title(self) -> str: + what = "单价" if self.per == "unit" else "分项总价" + return f"{self.name[:40]}：{what}不超过 {self.cap:,.2f} 元" + + def spec(self) -> dict[str, object]: + return { + "price": { + "kind": "unit_cap" if self.per == "unit" else "item_cap", + "name": self.name, + "cap": f"{self.cap:.2f}", + **({"qty": str(self.qty)} if self.qty is not None else {}), + **({"lot_no": self.lot_no} if self.lot_no else {}), + } + } + + +def _number(text: str) -> Decimal | None: + m = NUMBER.search(norm(text)) + if not m: + return None + try: + return Decimal(m.group(0).replace(",", "")) + except InvalidOperation: + return None + + +def _lot(text: str) -> str | None: + m = LOT.search(text) + if not m: + return None + v = m.group(1) + return "".join(CN_DIGIT.get(c, c) for c in v) if not v.isdigit() else v + + +def caps(sentences: list[Sentence]) -> list[Cap]: + out: list[Cap] = [] + lot: str | None = None + cur: tuple[dict[str, tuple[float, float]], int, str] | None = None # 栏、单位倍数、单价 / 分项 + for _, page in sorted(layout.pages(sentences).items()): + for row in layout.rows(page): + text = norm("".join(c.text for c in row)) + if not any(c.cell for c in row): + lot = _lot(text) or lot # 表前「采购包2：……」 + if cur is not None and len(text) > 30: + cur = None # 表格结束（正文） + continue + head = _head(row) + if head is not None: + cur = head + continue + if cur is None: + continue + cols, scale, per = cur + # 名称栏里取不是纯数字的那一格（序号表头没认成单元格时，序号会被归到名称栏） + name = next( + (c.text for c in row if layout.column(c, cols) == "name" and not NUMBER.fullmatch(norm(c.text))), "" + ) + cap = next((v for c in row if layout.column(c, cols) == "cap" and (v := _number(c.text))), None) + qty = next((v for c in row if layout.column(c, cols) == "qty" and (v := _number(c.text))), None) + if not name or cap is None: + if any(layout.column(c, cols) == "cap" for c in row): + cur = None + continue # 表头折到下一页的残行、合计行 + if LOT.fullmatch(norm(name)) or re.search(r"合计|总计|小计", name): + continue + first = next(c for c in row if c.text == name) # 名称格做首句：清单按首句对齐，序号格容易撞 + out.append(Cap(norm(name), cap * scale, per, qty, lot, [first, *(c for c in row if c is not first)])) + return out + + +def _head(row: list[Sentence]) -> tuple[dict[str, tuple[float, float]], int, str] | None: + cols: dict[str, tuple[float, float]] = {} + scale, per = 1, "item" + for n, c in enumerate(row): + t = norm(c.text) + if CAP_HEAD.search(t) and "cap" not in cols: + cols["cap"] = layout.span(c) + scale = 10_000 if "万元" in t else 1 + per = "unit" if "单价" in t else "item" + elif NAME_HEAD.search(t) and "name" not in cols and not LOT_HEAD.search(t): + cols["name"] = layout.span(c) + elif QTY_HEAD.search(t) and "qty" not in cols: + cols["qty"] = layout.span(c) + else: + cols[f"c{n}"] = layout.span(c) + if not {"cap", "name"} <= cols.keys() or any(LOT_HEAD.search(norm(c.text)) for c in row): + return None + return cols, scale, per diff --git a/server/src/bid_doctor/document/pipeline/layout.py b/server/src/bid_doctor/document/pipeline/layout.py new file mode 100644 index 0000000..4a007d4 --- /dev/null +++ b/server/src/bid_doctor/document/pipeline/layout.py @@ -0,0 +1,80 @@ +"""版面工具：按坐标把文字层的句子分成页、行、栏（读标认表格、查标认投标文件的表格都用）。 + +有框线的表格按单元格的上沿分行（文字层给句子带了单元格框）：同一行的格子上沿相同，垂直居中的短格、 +折成几句的长格、纵向合并的格（只算进它的第一行）都不会串行；其余按句子上下范围重叠分行。 +""" + +from __future__ import annotations + +from dataclasses import dataclass + +from bid_doctor.document.pipeline.text_layer import Sentence + + +def box(s: Sentence) -> tuple[float, float, float, float]: + return ( + min(q[0] for q in s.quads), + min(q[1] for q in s.quads), + max(q[2] for q in s.quads), + max(q[3] for q in s.quads), + ) + + +def span(s: Sentence) -> tuple[float, float]: + """左右范围：在表格里取单元格的，否则取文字的。""" + return (s.cell[0], s.cell[2]) if s.cell else (box(s)[0], box(s)[2]) + + +def pages(sentences: list[Sentence]) -> dict[int, list[Sentence]]: + out: dict[int, list[Sentence]] = {} + for s in sentences: + out.setdefault(s.page_no, []).append(s) + return out + + +@dataclass(slots=True) +class _Row: + top: float + bottom: float + in_cell: bool + items: list[Sentence] + + +def rows(page: list[Sentence]) -> list[list[Sentence]]: + """一页的句子分成行，行内从左到右。""" + out: list[_Row] = [] + + def top_of(s: Sentence) -> float: + return s.cell[1] if s.cell else box(s)[1] + + for s in sorted((x for x in page if x.quads), key=top_of): + if s.cell: + r = next((r for r in out if r.in_cell and abs(r.top - s.cell[1]) < 2), None) + if r: + r.items.append(s) + else: + out.append(_Row(s.cell[1], s.cell[1], True, [s])) + continue + _, top, _, bottom = box(s) + last = out[-1] if out else None + if last and not last.in_cell and top < last.bottom - 1: + last.bottom = max(last.bottom, bottom) + last.items.append(s) + else: + out.append(_Row(top, bottom, False, [s])) + return [sorted(r.items, key=lambda x: box(x)[0]) for r in sorted(out, key=lambda r: r.top)] + + +def column(cell: Sentence, cols: dict[str, tuple[float, float]]) -> str | None: + """单元格落在哪一栏：和表头单元格左右范围重叠最多的那一栏；都不重叠（窄数字居中、表头字靠左）取中心最近的。 + cols 要包含表头的每一格，否则别的栏的数字会被归到最近的已知栏。""" + x0, x1 = span(cell) + best, score = None, 0.0 + for name, (c0, c1) in cols.items(): + overlap = min(x1, c1) - max(x0, c0) + if overlap > score: + best, score = name, overlap + if best is None and cols: + mid = (x0 + x1) / 2 + best = min(cols, key=lambda n: abs((cols[n][0] + cols[n][1]) / 2 - mid)) + return best diff --git a/server/tests/test_check_flow.py b/server/tests/test_check_flow.py index dbdef6e..6901eb5 100644 --- a/server/tests/test_check_flow.py +++ b/server/tests/test_check_flow.py @@ -297,7 +297,7 @@ async def test_rule_findings_are_written_with_citations_and_carried(user: User) # 「让利后总价」和投标总价不同：多半另报了一个价，但不是「备选方案」那种明确的两个报价，按重要 assert rule["PRICE.MULTIPLE_QUOTES"]["severity"] == "important" and rule["FORM.BLANK_FIELD"]["severity"] == "hint" assert any("报价形式" in x for x in run["coverage"]["checked"]) - assert run["coverage"]["rule_set_version"] == "2026.11-r1" + assert run["coverage"]["rule_set_version"] == "2026.11-r2" await user.post(f"/findings/{cap['id']}/resolve", json={"status": "ignored", "note": "大写栏已手工更正"}) again = await run_check(user, project_id, [bid["id"]]) @@ -305,3 +305,37 @@ async def test_rule_findings_are_written_with_citations_and_carried(user: User) f for f in await findings(user, again["id"]) if (f["rule"] or {}).get("code") == "PRICE.CAPITAL_MISMATCH" ) assert cap2["presence"] == "carried" and cap2["status"] == "ignored" + + +async def test_manual_checks_are_kept_per_project(user: User) -> None: + """需人工核对：来自清单（按主题合并），勾选按项目存、写勾选人和时间，复查后还在（决策 P2）。""" + project_id, _, _ = await checked_project(user) + for title in ("须用CA数字证书加密投标文件", "开标后30分钟内在线解密投标文件", "本次采购不收取投标保证金"): + r = await user.post( + f"/projects/{project_id}/checklist/items", json={"categories": ["schedule"], "title": title} + ) + assert r.status_code == 201, r.text + items = (await user.get(f"/projects/{project_id}/manual-checks")).json()["items"] + assert [(i["check_key"], i["checked"]) for i in items] == [ + ("rule:MANUAL.DECRYPT", False), + ("rule:MANUAL.SIGN_ENCRYPT", False), + ] + assert items[1]["requirement_text"] == "须用CA数字证书加密投标文件" + + r = await user.put( + f"/projects/{project_id}/manual-checks", + json={"check_key": "rule:MANUAL.DECRYPT", "checked": True, "note": "已演练"}, + ) + assert r.status_code == 200, r.text + m = r.json() + assert m["checked"] and m["note"] == "已演练" and m["checked_by"]["name"] and m["checked_at"] + bad = await user.put( + f"/projects/{project_id}/manual-checks", json={"check_key": "rule:MANUAL.NOPE", "checked": True} + ) + assert bad.status_code == 422 + + bid_id = (await user.get(f"/projects/{project_id}/check-preview")).json()["bid_documents"][0]["id"] + again = await run_check(user, project_id, [bid_id]) + assert any("列入「需人工核对」2 条" in x for x in again["coverage"]["not_checked"]) + kept = (await user.get(f"/projects/{project_id}/manual-checks")).json()["items"] + assert [i["checked"] for i in kept] == [True, False] diff --git a/server/tests/test_check_units.py b/server/tests/test_check_units.py index 13ff160..0cd292b 100644 --- a/server/tests/test_check_units.py +++ b/server/tests/test_check_units.py @@ -166,3 +166,28 @@ def test_scanned_fallback_turns_missing_evidence_into_a_hint() -> None: and "2 页扫描件" in spec.explanation ) assert results[b].verdict == "not_found" # 技术参数不是证明材料，照常报未找到 + + +def test_manual_topics_merge_by_topic_and_skip_irrelevant() -> None: + """需人工核对（FR-M3-04）：同一主题合并成一条；技术参数、免收保证金、围串标情形、点名要交的文件不算。""" + from bid_doctor.check import manual + + items = [ + { + "title": "投标文件递交截止 2027年3月5日 9:30", + "categories": ["schedule"], + "spec": {"schedule": {"at": "2027-03-05T09:30"}}, + }, + {"title": "电子投标文件须在截止时间前上传", "categories": ["schedule"]}, + {"title": "未在规定时间内解密投标文件的投标无效", "categories": ["invalidity"]}, + {"title": "须提供加盖电子签章的《开标一览表》", "categories": ["format_doc"]}, + {"title": "须用CA数字证书加密投标文件", "categories": ["schedule"]}, + {"title": "本项目免收投标保证金", "categories": ["schedule"]}, + {"title": "数据库支持国密加密存储", "categories": ["tech_param"]}, + {"title": "不同投标人的投标文件装订明显相似的投标无效", "categories": ["invalidity"]}, + {"title": "纸质投标文件正本一份、副本四份，密封递交", "categories": ["schedule"]}, + ] + got = {t.key: [i["title"] for i in t.items] for t in manual.topics(items)} + assert list(got) == ["DECRYPT", "SIGN_ENCRYPT", "PAPER", "SUBMIT"] + assert got["SIGN_ENCRYPT"] == ["须用CA数字证书加密投标文件"] + assert got["SUBMIT"][0].startswith("投标文件递交截止") # 写明截止时间的做代表 diff --git a/server/tests/test_price_caps.py b/server/tests/test_price_caps.py new file mode 100644 index 0000000..03e6666 --- /dev/null +++ b/server/tests/test_price_caps.py @@ -0,0 +1,93 @@ +"""「报价」类清单项（C7）：从招标文件的清单表读分项 / 单价上限，查标时和分项报价表比对。内容虚构。""" + +from __future__ import annotations + +from decimal import Decimal +from uuid import uuid4 + +from bid_doctor.check import applicability, match, rules +from bid_doctor.checklist import price +from bid_doctor.document.pipeline.text_layer import Sentence + +COLS = [(80, 110), (110, 300), (300, 340), (340, 380), (380, 470)] + + +def row(seq: int, y: float, texts: list[str], page: int = 1) -> list[Sentence]: + out = [] + for n, (t, (x0, x1)) in enumerate(zip(texts, COLS, strict=False)): + s = Sentence( + key=f"t:p{page}:s{seq + n}", page_no=page, seq=seq + n, text=t, quads=[[x0 + 2, y + 4, x1 - 2, y + 16]] + ) + s.cell = [x0, y, x1, y + 20] + out.append(s) + return out + + +def line(seq: int, y: float, text: str, page: int = 1) -> Sentence: + return Sentence(key=f"t:p{page}:s{seq}", page_no=page, seq=seq, text=text, quads=[[74, y, 500, y + 12]]) + + +def tender() -> list[Sentence]: + return [ + line(1, 60, "采购包2：数据中心扩容"), + *row(10, 80, ["序号", "分项名称", "数量", "单位", "单价上限（万元）"]), + *row(20, 100, ["1", "服务器", "4", "台", "12.5"]), + *row(30, 120, ["2", "服务器机柜", "2", "个", "0.8"]), + line(40, 160, "采购包一览：各包的最高限价见下表，投标报价超过所投采购包最高限价的投标无效。"), + *row(50, 180, ["序号", "采购包名称", "数量", "单位", "最高限价（元）"]), # 采购包一级：不是分项 + *row(60, 200, ["1", "采购包2", "1", "项", "600000"]), + ] + + +def test_caps_from_tender_table() -> None: + caps = price.caps(tender()) + assert [(c.lot_no, c.name, c.cap, c.per, c.qty) for c in caps] == [ + ("2", "服务器", Decimal("125000.0"), "unit", Decimal(4)), + ("2", "服务器机柜", Decimal("8000.0"), "unit", Decimal(2)), + ] + assert caps[0].title == "服务器：单价不超过 125,000.00 元" and caps[0].sentences[0].text == "服务器" + assert caps[0].spec()["price"] == { + "kind": "unit_cap", + "name": "服务器", + "cap": "125000.00", + "qty": "4", + "lot_no": "2", + } + + +def test_price_items_are_checked_by_rule_not_model() -> None: + d = applicability.decide({"categories": ["price"], "title": "服务器：单价不超过 125,000.00 元"}, {}) + assert (d.match, d.skip) == (False, "price") + + +def bid_row(seq: int, y: float, texts: list[str]) -> list[Sentence]: + out = [] + for n, (t, (x0, x1)) in enumerate( + zip(texts, [(80, 100), (100, 260), (260, 300), (300, 380), (380, 470)], strict=True) + ): + s = Sentence(key=f"b:p1:s{seq + n}", page_no=1, seq=seq + n, text=t, quads=[[x0 + 2, y + 4, x1 - 2, y + 16]]) + s.cell = [x0, y, x1, y + 20] + out.append(s) + return out + + +def test_item_over_cap_is_fatal_and_similar_names_do_not_match() -> None: + bid = [ + *bid_row(1, 80, ["序号", "货物名称", "数量", "单价（元）", "合价（元）"]), + *bid_row(10, 100, ["1", "服务器", "4", "126,000.00", "504,000.00"]), + *bid_row(20, 120, ["2", "服务器机柜", "2", "7,900.00", "15,800.00"]), + ] + item = uuid4() + facts = rules.Facts( + caps=[ + rules.ItemCap("服务器", Decimal(125_000), "unit", item), + rules.ItemCap("服务器机柜", Decimal(8_000), "unit"), + ] + ) + hits, cov = rules.run(facts, [match.BidDoc(uuid4(), 1, "报价分册", bid)], []) + (h,) = [h for h in hits if h.code == "PRICE.OVER_CEILING_ITEM"] + assert h.severity == "fatal" and h.item == item and h.basis == f"cap:{item}" and "服务器" in h.title + assert "126,000.00 元" in h.explanation and "125,000.00 元" in h.explanation + assert not any("其余没有比对" in x for x in cov["not_checked"]) # 两条上限都对上了名称 + + assert rules._match_cap("服务器", [rules.ItemCap("服务器机柜", Decimal(1), "unit")]) is None