feat: 读标补齐五类清单与项目基本信息，修复国动样本暴露的 4 个问题

- 文字层 v2：页眉页脚带单独成句；写到页尾的句子和下一页首句接续（出处两页各一条） - 结构识别（最小版）：十进制编号、冒号引出的列表找上级条款；上级条款写明无效的，条目照样认定 - 抽取：资格、评分、日期与递交、技术参数、项目基本信息各一份提示词；实质性条款出 v2 （认上级条款、补「做无效响应处理」等写法）；各类取值核对；按首句合并；拆开被合成一条的列举条款；描述标注方式的句子和表头碎片不当条款；* 号标注 - 项目：解析出的名称、编号、采购人、预算、截止时间、分包写回项目，用户改过的不覆盖 - 前端：清单项显示分值、参数要求、日期、证明材料 - 测试 40 个，新增两页虚构样例 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 19:32:25 +08:00
parent ee4bb23488
commit a95e27a290
29 changed files with 1971 additions and 230 deletions
--- a/server/src/bid_doctor/project/extracted.py
+++ b/server/src/bid_doctor/project/extracted.py
@@ -0,0 +1,119 @@
+"""把解析出的项目基本信息和分包写回项目（FR-M2-03、04）。
+
+- AI 值和人工值分开存：用户改过的字段不覆盖；项目名称只在还是「未命名项目」时自动填。
+- 补遗带来的新值先放进待确认（FR-M2-10），用户确认后才生效；首份招标文件直接生效。
+- 分包：解析出包时替换掉建项目时的默认包；已经有包的项目不动（要改由用户在选包页处理）。
+"""
+
+from __future__ import annotations
+
+from datetime import datetime
+from decimal import Decimal
+from typing import Any
+from uuid import UUID
+
+from sqlalchemy import delete, select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from bid_doctor.checklist.extract import Extracted, ProjectInfo, amount_yuan
+from bid_doctor.core.db import TenantContext
+from bid_doctor.core.ids import uuid7
+from bid_doctor.document import citations
+from bid_doctor.document.models import SourceDocument
+from bid_doctor.project import service as projects
+from bid_doctor.project.models import Lot, Project, ProjectField
+
+MONEY = ("budget_total", "ceiling_total")
+COLUMNS = ("name", "code", "purchaser", "agency", "procurement_method", "legal_regime")
+
+
+async def apply(
+    s: AsyncSession,
+    ctx: TenantContext,
+    doc: SourceDocument,
+    info: ProjectInfo,
+    regime_by_rule: str,
+    items: list[Extracted],
+) -> dict[str, UUID]:
+    """写入字段、分包和投标截止时间。返回 {包号: 分包编号}，给清单项挂所属包用。"""
+    p = await projects.get(s, ctx, doc.project_id)
+    values: dict[str, tuple[Any, Any]] = {}
+    for key, f in info.fields.items():
+        value = amount_yuan(f.value, f.unit) if key in MONEY else f.value
+        if key == "legal_regime" and value not in ("gov_procurement", "tendering"):
+            continue
+        values[key] = (value, f.sentence)
+    if "legal_regime" not in values and regime_by_rule != "unknown":
+        values["legal_regime"] = (regime_by_rule, None)
+    deadline = _deadline(items)
+    if deadline:
+        values["deadline_at"] = deadline
+    for key, (value, sentence) in values.items():
+        cite = None
+        if sentence is not None:
+            (c,) = await citations.from_sentences(s, ctx, doc, [sentence], "ai", ["id", "quote", "basis"])
+            cite = c.id
+        await _set_field(s, ctx, p, doc, key, value, cite)
+    return await _lots(s, ctx, p, info)
+
+
+def _deadline(items: list[Extracted]) -> tuple[str, Any] | None:
+    for e in items:
+        sched = e.spec.get("schedule") or {}
+        if e.zone == "formal" and sched.get("date_kind") == "bid_deadline" and sched.get("at"):
+            return sched["at"], e.sentences[0]
+    return None
+
+
+async def _set_field(
+    s: AsyncSession, ctx: TenantContext, p: Project, doc: SourceDocument, key: str, value: Any, cite: UUID | None
+) -> None:
+    f = (
+        await s.execute(select(ProjectField).where(ProjectField.project_id == p.id, ProjectField.field_key == key))
+    ).scalar_one_or_none()
+    if f is None:
+        f = ProjectField(tenant_id=ctx.tenant_id, project_id=p.id, field_key=key)
+        s.add(f)
+    if doc.role in ("addendum", "clarification") and f.ai_value is not None and f.ai_value != value:
+        f.pending_value, f.pending_citation_id = value, cite  # 补遗的新值：等用户确认（FR-M2-10）
+    else:
+        f.ai_value, f.ai_citation_id, f.ai_document_id = value, cite, doc.id
+    await s.flush()
+    if f.human_value is not None or f.pending_value is not None:
+        return
+    if key == "name" and p.name != projects.UNNAMED:
+        return  # 用户建项目时填了名称：以用户的为准
+    if key in COLUMNS:
+        setattr(p, key, value)
+    elif key == "deadline_at":
+        p.deadline_at = datetime.fromisoformat(value)
+
+
+async def _lots(s: AsyncSession, ctx: TenantContext, p: Project, info: ProjectInfo) -> dict[str, UUID]:
+    existing = (await s.execute(select(Lot).where(Lot.project_id == p.id).order_by(Lot.sort_order))).scalars().all()
+    if info.lots and all(lot.is_default for lot in existing):
+        await s.execute(delete(Lot).where(Lot.project_id == p.id, Lot.is_default))
+        for i, lot in enumerate(info.lots):
+            budget, ceiling = lot.get("budget") or {}, lot.get("ceiling") or {}
+            b, c = (
+                amount_yuan(budget.get("value"), budget.get("unit")),
+                amount_yuan(ceiling.get("value"), ceiling.get("unit")),
+            )
+            s.add(
+                Lot(
+                    id=uuid7(),
+                    tenant_id=ctx.tenant_id,
+                    project_id=p.id,
+                    lot_no=str(lot["lot_no"])[:40],
+                    name=lot.get("name"),
+                    origin="ai",
+                    sort_order=i,
+                    edited_fields=[],
+                    budget_amount=Decimal(b) if b else None,
+                    ceiling_amount=Decimal(c) if c else None,
+                    ai_values={k: v for k, v in lot.items() if k != "sentence"},
+                )
+            )
+        await s.flush()
+        existing = (await s.execute(select(Lot).where(Lot.project_id == p.id))).scalars().all()
+    return {lot.lot_no: lot.id for lot in existing}