X-Financial/server/src/app/services/user_agent.py

from __future__ import annotations

import json
import re
from datetime import UTC, datetime, timedelta
from decimal import Decimal, InvalidOperation
from typing import Any

from sqlalchemy import or_, select
from sqlalchemy.orm import Session, selectinload

from app.core.agent_enums import AgentAssetStatus, AgentAssetType
from app.models.employee import Employee
from app.models.financial_record import ExpenseClaim
from app.schemas.agent_asset import AgentAssetListItem
from app.schemas.user_agent import (
    UserAgentCitation,
    UserAgentDraftPayload,
    UserAgentExpenseQueryRecord,
    UserAgentQueryPayload,
    UserAgentQueryStatusGroup,
    UserAgentReviewAction,
    UserAgentReviewEditField,
    UserAgentReviewClaimGroup,
    UserAgentReviewDocumentCard,
    UserAgentReviewDocumentField,
    UserAgentReviewPayload,
    UserAgentReviewRiskBrief,
    UserAgentReviewSlotCard,
    UserAgentRequest,
    UserAgentResponse,
    UserAgentSuggestedAction,
)
from app.services.agent_assets import AgentAssetService
from app.services.agent_foundation import AgentFoundationService
from app.services.expense_claims import ExpenseClaimService
from app.services.risk_ontology_bridge import resolve_rule_codes_for_risk_check
from app.services.runtime_chat import RuntimeChatService

SCENARIO_LABELS = {
    "expense": "报销",
    "accounts_receivable": "应收",
    "accounts_payable": "应付",
    "knowledge": "知识",
    "unknown": "通用",
}

RISK_REASON_MAP = {
    "duplicate_expense": "检测到同员工、同金额或近似单据存在重复提交迹象。",
    "location_mismatch": "申报出差地点与票据识别地点可能不一致，需要核对行程或补充说明。",
    "amount_over_limit": "金额超过当前制度或预算阈值，需要补充例外说明。",
    "invoice_anomaly": "票据或附件完整性不满足当前规则要求，需要补件或人工复核。",
    "ar_overdue": "应收账款已出现逾期，存在回款延迟风险。",
    "ap_overdue": "应付付款已出现逾期，可能影响供应商履约或合作关系。",
}

GENERIC_EXPENSE_PROMPTS = {
    "报销",
    "我要报销",
    "我想报销",
    "帮我报销",
    "我要申请报销",
    "发起报销",
    "提交报销",
}

EXPLICIT_DRAFT_KEYWORDS = ("生成", "草稿", "起草", "创建", "发起", "准备")

EXPENSE_TYPE_LABELS = {
    "travel": "差旅费",
    "hotel": "住宿费",
    "transport": "交通费",
    "meal": "餐费",
    "meeting": "会务费",
    "entertainment": "业务招待费",
    "office": "办公费",
    "training": "培训费",
    "communication": "通讯费",
    "welfare": "福利费",
    "other": "其他费用",
}

GROUP_SCENE_LABELS = {
    "travel": "差旅费",
    "entertainment": "业务招待费",
    "meal": "伙食费",
    "transport": "交通费",
    "hotel": "住宿费",
    "office": "办公费",
    "training": "培训费",
    "communication": "通讯费",
    "welfare": "福利费",
    "other": "其他费用",
}

KNOWLEDGE_MODEL_MAIN_TIMEOUT_SECONDS = 3
KNOWLEDGE_MODEL_BACKUP_TIMEOUT_SECONDS = 5
KNOWLEDGE_MODEL_TIMEOUT_SECONDS = KNOWLEDGE_MODEL_BACKUP_TIMEOUT_SECONDS

KNOWLEDGE_DIRECT_ANSWER_HINTS = (
    "是什么",
    "标准",
    "限额",
    "流程",
    "条件",
    "规则",
    "怎么",
    "如何",
    "哪些",
    "需要",
    "是否",
    "区别",
    "范围",
    "额度",
    "金额",
    "多少",
    "多少钱",
    "上限",
)
KNOWLEDGE_QUERY_STOPWORDS = {
    "什么",
    "多少",
    "哪些",
    "怎么",
    "如何",
    "请问",
    "一下",
    "关于",
    "规定",
    "标准",
    "可以",
    "是否",
    "一个",
    "哪些人",
    "目前",
    "当前",
    "一下子",
}
MAX_KNOWLEDGE_QUERY_TERMS = 12
MAX_KNOWLEDGE_DIRECT_EVIDENCE = 4
MAX_KNOWLEDGE_MODEL_HITS = 5
KNOWLEDGE_SECTION_HEADING_PATTERN = re.compile(
    r"^(#\s*.+|##\s*.+|###\s*.+|第[一二三四五六七八九十百零0-9]+[章节条]\s*.*|[一二三四五六七八九十]+、.*|（[一二三四五六七八九十]+）.*|\([一二三四五六七八九十]+\).*)$"
)
KNOWLEDGE_LIST_ITEM_PATTERN = re.compile(r"^[-*•]\s+.+$")
KNOWLEDGE_NUMBERED_ITEM_PATTERN = re.compile(
    r"^(?:(?:\d+[.)、])|(?:[（(][一二三四五六七八九十百零0-9]+[)）])|[①②③④⑤⑥⑦⑧⑨⑩])\s*.+$"
)
KNOWLEDGE_ARTICLE_PATTERN = re.compile(r"^(第[一二三四五六七八九十百零0-9]+条)\s*.*$")

EXPENSE_STATUS_LABELS = {
    "draft": "草稿",
    "submitted": "已提交",
    "review": "审核中",
    "approved": "已通过",
    "rejected": "已驳回",
    "paid": "已付款",
}

EXPENSE_STATUS_GROUP_LABELS = {
    "draft": "草稿",
    "in_progress": "审批中",
    "completed": "审批完成",
    "other": "其他状态",
}

SLOT_LABELS = {
    "expense_type": "报销类型",
    "customer_name": "客户名称",
    "time_range": "发生时间",
    "location": "地点",
    "merchant_name": "酒店/商户",
    "amount": "金额",
    "reason": "事由说明",
    "participants": "参与人员",
    "attachments": "票据附件",
}

DATE_TEXT_PATTERN = re.compile(r"(\d{4}[年/-]\d{1,2}[月/-]\d{1,2}日?)")
AMOUNT_TEXT_PATTERN = re.compile(
    r"(\d+(?:\.\d+)?)\s*(?:万元|万员|万圆|万园|万块|万元整|元整|块钱|块|元|员|圆|园|万)"
)
DOCUMENT_AMOUNT_PATTERN = re.compile(
    r"(?:价税合计|合计金额|费用合计|订单(?:总)?金额|支付(?:金额)?|实付(?:金额)?|实收(?:金额)?|总(?:额|计|价)|票价|金额|车费|消费金额)"
    r"[：:\s￥¥人民币]*([0-9]+(?:[.,][0-9]{1,2})?)"
)
DOCUMENT_CURRENCY_AMOUNT_PATTERN = re.compile(r"[￥¥]\s*([0-9]+(?:[.,][0-9]{1,2})?)")

SOURCE_LABELS = {
    "user_text": "用户描述",
    "user_form": "用户修改",
    "ocr": "票据识别",
    "upload": "上传附件",
    "detail_context": "关联单据",
    "system_context": "系统上下文",
    "inferred": "语义推断",
    "system": "系统判断",
}

SCENE_REQUIRED_SLOT_KEYS = {
    "hotel": {"merchant_name"},
    "meeting": {"location"},
    "entertainment": {"location", "customer_name", "participants"},
}
INFERRED_REASON_LABELS = {
    "travel": "出差行程",
    "hotel": "住宿报销",
    "transport": "交通出行",
    "meal": "餐饮用餐",
    "meeting": "会务活动",
    "entertainment": "客户接待",
    "office": "办公采购",
    "training": "培训学习",
    "communication": "通讯使用",
    "welfare": "员工福利",
    "other": "其他费用",
}
SYSTEM_GENERATED_REASON_PREFIXES = (
    "我上传了",
    "请按当前已识别信息",
    "请把当前上传的票据",
    "请基于当前上传的多张票据",
    "我已核对右侧识别结果",
    "请同步修正逐票据识别结果",
    "我已修改识别信息",
    "查看报销草稿",
    "请解释一下当前这笔报销的合规风险和待补充项",
)
AMOUNT_UNIT_ALIASES = {
    "员": "元",
    "圆": "元",
    "园": "元",
    "块": "元",
    "块钱": "元",
    "元整": "元",
    "万员": "万元",
    "万圆": "万元",
    "万园": "万元",
    "万块": "万元",
    "万元整": "万元",
}


class UserAgentService:
    def __init__(self, db: Session) -> None:
        self.db = db
        self.asset_service = AgentAssetService(db)
        self.runtime_chat_service = RuntimeChatService(db)

    def respond(self, payload: UserAgentRequest) -> UserAgentResponse:
        AgentFoundationService(self.db).ensure_foundation_ready()
        citations = self._build_citations(payload)
        suggested_actions = self._build_suggested_actions(payload)
        risk_flags = self._resolve_risk_flags(payload)
        query_payload = self._build_query_payload(payload)
        draft_payload = (
            self._build_draft_payload(payload)
            if payload.ontology.intent == "draft"
            else None
        )
        review_payload = self._build_review_payload(
            payload,
            citations=citations,
            draft_payload=draft_payload,
        )
        review_answer = self._build_review_body_answer(
            payload,
            review_payload=review_payload,
            draft_payload=draft_payload,
        )

        if payload.degraded and payload.tool_payload.get("message"):
            return UserAgentResponse(
                answer=review_answer or str(payload.tool_payload["message"]),
                citations=citations,
                suggested_actions=suggested_actions,
                query_payload=query_payload,
                draft_payload=draft_payload,
                review_payload=review_payload,
                risk_flags=risk_flags,
                requires_confirmation=payload.requires_confirmation,
            )

        if review_answer:
            return UserAgentResponse(
                answer=review_answer,
                citations=citations,
                suggested_actions=suggested_actions,
                query_payload=query_payload,
                draft_payload=draft_payload,
                review_payload=review_payload,
                risk_flags=risk_flags,
                requires_confirmation=payload.requires_confirmation,
            )

        guided_answer = None
        if draft_payload is None or draft_payload.claim_id is None:
            guided_answer = self._build_guided_answer(payload)
        if guided_answer:
            return UserAgentResponse(
                answer=guided_answer,
                citations=citations,
                suggested_actions=suggested_actions,
                query_payload=query_payload,
                draft_payload=draft_payload,
                review_payload=review_payload,
                risk_flags=risk_flags,
                requires_confirmation=payload.requires_confirmation,
            )

        fast_knowledge_answer = self._build_fast_knowledge_answer(
            payload,
            citations=citations,
        )
        if fast_knowledge_answer:
            return UserAgentResponse(
                answer=fast_knowledge_answer,
                citations=citations,
                suggested_actions=suggested_actions,
                query_payload=query_payload,
                draft_payload=draft_payload,
                review_payload=review_payload,
                risk_flags=risk_flags,
                requires_confirmation=payload.requires_confirmation,
            )

        fallback_answer = self._build_fallback_answer(
            payload,
            citations=citations,
            draft_payload=draft_payload,
        )
        answer = None
        if not self._should_skip_model_answer(payload, review_payload):
            answer = self._generate_answer_with_model(
                payload,
                citations=citations,
                suggested_actions=suggested_actions,
                risk_flags=risk_flags,
                draft_payload=draft_payload,
                fallback_answer=fallback_answer,
            )

        return UserAgentResponse(
            answer=answer or fallback_answer,
            citations=citations,
            suggested_actions=suggested_actions,
            query_payload=query_payload,
            draft_payload=draft_payload,
            review_payload=review_payload,
            risk_flags=risk_flags,
            requires_confirmation=payload.requires_confirmation,
        )

    def _build_fallback_answer(
        self,
        payload: UserAgentRequest,
        *,
        citations: list[UserAgentCitation],
        draft_payload: UserAgentDraftPayload | None,
    ) -> str:
        if str(payload.tool_payload.get("result_type") or "").strip() == "knowledge_search":
            return self._build_explain_answer(payload, citations)

        if payload.ontology.intent in {"query", "compare"}:
            return self._build_query_answer(payload)

        if payload.ontology.intent == "risk_check":
            return self._build_risk_answer(payload, citations)

        if payload.ontology.intent == "draft":
            tool_message = str(payload.tool_payload.get("message") or "").strip()
            if payload.tool_payload.get("draft_limit_reached"):
                return tool_message or "你当前已保存 3 个草稿，请先完成已保存的草稿，才能再次新建草稿。"
            if tool_message and (
                str(payload.tool_payload.get("claim_id") or "").strip()
                or str(payload.tool_payload.get("claim_no") or "").strip()
            ):
                return tool_message
        if payload.ontology.intent == "draft" and draft_payload is not None:
            return (
                f"已生成 {draft_payload.title}，当前仅返回待人工确认的草稿内容，"
                "仍需人工确认后再进入正式流程。"
            )

        return self._build_explain_answer(payload, citations)

    def _build_guided_answer(self, payload: UserAgentRequest) -> str | None:
        if not self._is_generic_expense_prompt(payload):
            return self._build_implicit_expense_draft_guidance(payload)

        attachment_names = self._resolve_attachment_names(payload)
        ocr_summary = str(payload.context_json.get("ocr_summary") or "").strip()
        attachment_hint = ""
        if ocr_summary:
            attachment_hint = f" 我已读取附件 OCR 摘要：{ocr_summary}"
        elif attachment_names:
            attachment_hint = (
                f" 我已带入 {len(attachment_names)} 份附件名称，但目前还不能直接读取附件内容，"
                "仍需要你补充关键信息。"
            )

        return (
            "可以帮你发起报销。请补充费用类型、发生时间、金额、事由和相关对象，"
            "或者直接上传票据附件，我再继续帮你判断能否报、缺什么材料以及生成报销草稿。"
            f"{attachment_hint}"
        )

    def _build_implicit_expense_draft_guidance(
        self,
        payload: UserAgentRequest,
    ) -> str | None:
        if not self._is_implicit_expense_draft_request(payload):
            return None

        amount_text = next(
            (item.value for item in payload.ontology.entities if item.type == "amount"),
            "",
        )
        expense_type = next(
            (
                EXPENSE_TYPE_LABELS.get(item.normalized_value, item.value)
                for item in payload.ontology.entities
                if item.type == "expense_type"
            ),
            "报销",
        )
        time_text = payload.ontology.time_range.raw or "本次"
        amount_hint = f"，金额 {amount_text}" if amount_text else ""

        return (
            f"已识别到一笔{time_text}的{expense_type}支出{amount_hint}。"
            "如果要继续生成报销草稿，还需要补充客户单位、参与人员、费用明细和票据附件。"
            "你也可以继续上传发票或图片，我会把这些信息带入后续对话。"
        )

    def _generate_answer_with_model(
        self,
        payload: UserAgentRequest,
        *,
        citations: list[UserAgentCitation],
        suggested_actions: list[UserAgentSuggestedAction],
        risk_flags: list[str],
        draft_payload: UserAgentDraftPayload | None,
        fallback_answer: str,
    ) -> str | None:
        messages = self._build_model_messages(
            payload,
            citations=citations,
            suggested_actions=suggested_actions,
            risk_flags=risk_flags,
            draft_payload=draft_payload,
            fallback_answer=fallback_answer,
        )
        answer = self._sanitize_model_answer(
            self.runtime_chat_service.complete(
                messages,
                max_tokens=800 if payload.ontology.scenario == "knowledge" else 420,
                temperature=0.2,
                timeout_seconds=(
                    KNOWLEDGE_MODEL_TIMEOUT_SECONDS
                    if payload.ontology.scenario == "knowledge"
                    else None
                ),
                slot_timeouts=(
                    {
                        "main": KNOWLEDGE_MODEL_MAIN_TIMEOUT_SECONDS,
                        "backup": KNOWLEDGE_MODEL_BACKUP_TIMEOUT_SECONDS,
                    }
                    if payload.ontology.scenario == "knowledge"
                    else None
                ),
                max_attempts=1 if payload.ontology.scenario == "knowledge" else None,
            )
        )
        return self._reject_unsupported_location_inference(payload, answer)

    def _sanitize_model_answer(self, answer: str | None) -> str | None:
        if not answer:
            return None

        cleaned = re.sub(r"<think>.*?</think>", "", answer, flags=re.DOTALL | re.IGNORECASE)
        cleaned = cleaned.strip()
        leaked_reasoning_markers = (
            "用户问的是",
            "让我分析一下",
            "实体识别",
            "从对话历史来看",
            "从tool_payload来看",
            "现在问题是",
            "我需要：",
            "关键是我",
        )
        if any(marker in cleaned[:500] for marker in leaked_reasoning_markers):
            return None
        return cleaned or None

    @staticmethod
    def _extract_query_location(message: str) -> str:
        match = re.search(r"(?:去|到|前往)([\u4e00-\u9fff]{2,8})(?:出差|开会|培训)", str(message or ""))
        return match.group(1) if match else ""

    def _reject_unsupported_location_inference(
        self,
        payload: UserAgentRequest,
        answer: str | None,
    ) -> str | None:
        del payload
        return answer

    def _build_model_messages(
        self,
        payload: UserAgentRequest,
        *,
        citations: list[UserAgentCitation],
        suggested_actions: list[UserAgentSuggestedAction],
        risk_flags: list[str],
        draft_payload: UserAgentDraftPayload | None,
        fallback_answer: str,
    ) -> list[dict[str, str]]:
        knowledge_question = (
            self._resolve_knowledge_question(payload)
            if payload.ontology.scenario == "knowledge"
            else ""
        )
        facts = {
            "run_id": payload.run_id,
            "user_message": payload.message,
            "ontology": payload.ontology.model_dump(mode="json"),
            "context": {
                "entry_source": payload.context_json.get("entry_source"),
                "user_name": payload.context_json.get("name"),
                "user_role": payload.context_json.get("role"),
                "user_position": payload.context_json.get("position"),
                "user_grade": payload.context_json.get("grade"),
                "user_role_codes": payload.context_json.get("role_codes", []),
                "is_admin": bool(payload.context_json.get("is_admin")),
                "request_context": payload.context_json.get("request_context"),
                "attachment_count": payload.context_json.get("attachment_count"),
                "attachment_names": self._resolve_attachment_names(payload),
                "ocr_summary": payload.context_json.get("ocr_summary", ""),
                "ocr_documents": payload.context_json.get("ocr_documents", []),
                "conversation_id": payload.context_json.get("conversation_id"),
                "conversation_scenario": payload.context_json.get("conversation_scenario"),
                "conversation_intent": payload.context_json.get("conversation_intent"),
                "draft_claim_id": payload.context_json.get("draft_claim_id"),
                "conversation_history": self._resolve_conversation_history(payload),
            },
            "tool_payload": self._build_model_tool_payload(
                payload.tool_payload,
                question=knowledge_question,
            ),
            "citations": [item.model_dump(mode="json") for item in citations],
            "suggested_actions": [item.model_dump(mode="json") for item in suggested_actions],
            "risk_flags": risk_flags,
            "draft_payload": draft_payload.model_dump(mode="json") if draft_payload is not None else None,
            "selected_capability_codes": payload.selected_capability_codes,
            "requires_confirmation": payload.requires_confirmation,
            "fallback_answer": fallback_answer,
        }
        if payload.ontology.scenario == "knowledge":
            facts["knowledge_evidence_blocks"] = self._build_knowledge_evidence_blocks(
                payload.tool_payload,
                question=knowledge_question,
            )
            facts["knowledge_answer_evidence"] = [
                {
                    "title": str(item.get("title") or "").strip(),
                    "heading": str(item.get("heading") or "").strip(),
                    "kind": str(item.get("kind") or "").strip(),
                    "content": str(item.get("content") or "").strip(),
                }
                for item in self._build_knowledge_answer_evidence(payload)
            ]

        if payload.ontology.scenario == "knowledge":
            answer_style_instruction = (
                "你是财务制度知识问答助手。只能依据 facts.tool_payload.hits、facts.knowledge_answer_evidence、citations 与 conversation_history 回答，"
                "不要扩展成通用助手。优先直接回答，不要复述思考过程，不要输出 JSON、代码块或 <think>。"
                "回答风格要像一位真正熟悉制度的财务伙伴：先直接回应用户的核心问题，再用一张简洁表格或短段落说明依据，"
                "最后补充最重要的注意事项。不要写成“已检索到内容”的系统回执，也不要把命中片段连缀成答案。"
                "必须优先回答用户当前这句话本身，不能把制度标题、制度全文或完整标准表当成主答案。"
                "如果用户问的是某次具体行程“一共能报多少”，就先给“当前已能确认的金额”，再用一张很短的表说明项目、"
                "适用标准、计算式和结果；如果总额还缺少住宿晚数、实际票据或其他必要条件，就明确写出“暂不能确认的部分”。"
                "只有用户明确在问“标准有哪些”或“制度全文怎么规定”时，才展开完整标准表。"
                "如果命中的知识已经足够支持计算、比较或归纳，就直接给出结论；金额、标准、天数、补贴等问题要把计算过程写清楚。"
                "适合时请使用 Markdown 二级标题、短段落和表格，让回答更清晰；表格必须保证每一行列数一致，不要出现空白残列。"
                "只能陈述 hits 中明确出现的事实，不能用常识、外部知识或主观推断补齐缺失条件。"
                "回答前先在全部 hits 中寻找与问题最直接相关的章节、表格或条目，不能只依赖排在最前面的片段。"
                "如果 facts.knowledge_answer_evidence 中已经给出更短的高相关证据，优先基于这些证据组织答案，再回看原始 hits 补上下文。"
                "如果某个表格在检索片段中已经被摊平成连续文本，只有在行、列和数值对应关系能够从片段本身明确确认时才能据此计算；"
                "如果列对应关系不清楚，必须说明表格结构在当前片段中不够清晰，不能把第一列或相邻数字想当然套给用户。"
                "如果 hits 中出现“结构化表格补充”，它表示知识归纳阶段已经把原文表格重新整理过，"
                "优先使用这类结构化表格来理解行列关系，再回看原文确认上下文。"
                "facts.knowledge_evidence_blocks 中保留了原始换行和定宽排版；遇到表格时，优先按这些证据块阅读，"
                "必须按表头从左到右逐列对应数值，不能把第一列的数值直接套给后面的列名。"
                "如果完成计算或归纳仍缺少某个关键映射关系、适用条件或数值依据，必须明确说明当前知识库还缺哪一项信息，再给出已能确认的部分。"
                "如果用户问题里没有明确给出某个套用条件，而 hits 或 evidence 里也没有明确出现，就不能自己补一个默认值。"
                "当问题涉及追问时，必须结合 conversation_history 延续上一轮上下文，而不是重新泛化成制度全文摘录。"
                "不要大段粘贴原始命中文本；只提炼与问题直接相关的规则、条件、金额和注意事项。"
                "如果依据仍然不足，明确指出缺少哪一项信息，再给出当前能确认的部分。"
            )
        else:
            answer_style_instruction = "用 2 到 4 段完成回答，优先给结论，再补充最关键的依据与下一步建议。"

        personalization_instruction = (
            "如果 context.user_name 存在，并且当前问题与员工本人适用标准、报销额度、审批权限、职级待遇有关，"
            "开头应自然称呼一次用户，例如“曹笑竹，您好”。"
            "如果需要根据员工身份判断标准，优先参考 context.user_grade 与 context.user_position。"
            "如果问题与用户身份无关，就不要生硬加入姓名、职级或岗位。"
        )

        system_prompt = (
            "你是 X-Financial 的专业财务 AI 助手。"
            "回答必须准确、自然、可执行，不要泄露中间推理。"
            "当知识问题有命中依据时，先给结论，再给结构化说明。"
            "不要把制度全文原样搬出来，不要把检索片段当作最终答案直接粘贴。"
            "如果使用表格，确保列名简洁、数值明确。"
            f"{personalization_instruction}"
            f"{answer_style_instruction}"
        )
        user_prompt = (
            "请严格依据下面的 facts 生成最终答复：\n"
            f"{json.dumps(facts, ensure_ascii=False, indent=2)}"
        )
        return [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": user_prompt},
        ]

    @staticmethod
    def _build_model_tool_payload(
        tool_payload: dict[str, Any],
        *,
        question: str | None = None,
    ) -> dict[str, Any]:
        normalized = dict(tool_payload or {})
        hits = []
        for item in UserAgentService._select_knowledge_model_hits(
            tool_payload,
            question=question,
        ):
            if not isinstance(item, dict):
                continue
            hits.append(
                {
                    "title": str(item.get("title") or "").strip(),
                    "document_name": str(item.get("document_name") or "").strip(),
                    "excerpt": str(item.get("excerpt") or "").strip(),
                    "content": str(item.get("content") or "").strip()[:1200],
                    "tags": list(item.get("tags") or [])[:5],
                    "evidence": list(item.get("evidence") or [])[:3],
                    "code": str(item.get("code") or "").strip(),
                }
            )
        normalized["hits"] = hits
        return normalized

    @staticmethod
    def _build_knowledge_evidence_blocks(
        tool_payload: dict[str, Any],
        *,
        question: str | None = None,
    ) -> str:
        blocks: list[str] = []
        for index, item in enumerate(
            UserAgentService._select_knowledge_model_hits(
                tool_payload,
                question=question,
            )[:3],
            start=1,
        ):
            if not isinstance(item, dict):
                continue
            title = str(item.get("title") or item.get("document_name") or f"证据 {index}").strip()
            code = str(item.get("code") or "").strip()
            content = str(item.get("content") or "").strip()
            if not content:
                continue
            blocks.append(
                "\n".join(
                    [
                        f"[证据 {index}] {title}" + (f" ({code})" if code else ""),
                        "```text",
                        content[:1200],
                        "```",
                    ]
                )
            )
        return "\n\n".join(blocks)

    @staticmethod
    def _select_knowledge_model_hits(
        tool_payload: dict[str, Any],
        *,
        question: str | None = None,
    ) -> list[dict[str, Any]]:
        raw_hits = [
            item
            for item in list(tool_payload.get("hits") or [])
            if isinstance(item, dict)
        ][: max(MAX_KNOWLEDGE_MODEL_HITS + 1, 6)]
        if not raw_hits:
            return []

        query_terms = UserAgentService._extract_knowledge_query_terms(question or "")
        if not query_terms:
            return raw_hits[:MAX_KNOWLEDGE_MODEL_HITS]

        ranked_hits = sorted(
            enumerate(raw_hits),
            key=lambda value: (
                UserAgentService._score_knowledge_model_hit(
                    value[1],
                    query_terms=query_terms,
                    rank_index=value[0],
                ),
                -value[0],
            ),
            reverse=True,
        )
        return [item for _, item in ranked_hits[:MAX_KNOWLEDGE_MODEL_HITS]]

    @staticmethod
    def _score_knowledge_model_hit(
        item: dict[str, Any],
        *,
        query_terms: list[str],
        rank_index: int,
    ) -> int:
        title = str(item.get("title") or item.get("document_name") or "").lower()
        excerpt = str(item.get("excerpt") or "").lower()
        content = str(item.get("content") or "").lower()
        haystack = "\n".join([title, excerpt, content[:1400]])

        matched_terms = [term for term in query_terms if term in haystack]
        score = max(1, 48 - rank_index * 4)
        score += len(matched_terms) * 10
        score += sum(1 for term in matched_terms if term in title) * 8

        leading_marker = UserAgentService._leading_knowledge_appendix_marker(content)
        if leading_marker == "# 章节导航":
            score -= 22
        elif leading_marker == "# 问答线索补充":
            score += 6 if matched_terms else -8
        elif leading_marker == "# 重点章节摘录":
            score += 4 if matched_terms else -4
        elif leading_marker == "# 结构化表格补充":
            score += 8 if matched_terms else -3

        if matched_terms and "|" in content:
            score += 8
        if matched_terms and any(marker in content for marker in ("：", ":")):
            score += 10
        if matched_terms and "\n" in content:
            score += 4
        if matched_terms and any(marker in content for marker in ("附表", "第", "条")):
            score += 4
        if matched_terms and any(marker in content for marker in ("第", "条", "：", "-", "•")):
            score += 4
        if re.search(r"没有.{0,8}(信息|规定|说明|依据)", content):
            score -= 12
        return score

    @staticmethod
    def _leading_knowledge_appendix_marker(content: str) -> str:
        normalized = str(content or "").lstrip()
        for marker in ("# 章节导航", "# 重点章节摘录", "# 问答线索补充", "# 结构化表格补充"):
            index = normalized.find(marker)
            if 0 <= index <= 220:
                return marker
        return ""

    def _build_query_answer(self, payload: UserAgentRequest) -> str:
        scenario = payload.ontology.scenario
        data = payload.tool_payload
        subject = self._resolve_subject(payload)

        if scenario == "expense":
            query_payload = self._build_query_payload(payload)
            scope_label = str(data.get("scope_label") or subject).strip() or subject
            if query_payload is None:
                return f"当前没有查到{scope_label}。你可以补充时间范围、单号或状态继续筛选。"

            window_prefix = (
                f"{query_payload.window_start_date} 至 {query_payload.window_end_date}"
                if query_payload.recent_window_applied
                and query_payload.window_start_date
                and query_payload.window_end_date
                else (
                    f"近 {query_payload.window_days} 日内"
                    if query_payload.recent_window_applied and query_payload.window_days
                    else "当前条件下"
                )
            )
            if query_payload.record_count <= 0:
                if query_payload.older_record_count > 0 and query_payload.window_days:
                    return (
                        f"{window_prefix}没有查到{query_payload.scope_label}。"
                        f"另有 {query_payload.older_record_count} 笔超过 {query_payload.window_days} 日的单据，"
                        "请前往个人报销中心查看。"
                    )
                return f"{window_prefix}没有查到{query_payload.scope_label}。你可以补充时间范围、单号或状态继续筛选。"

            group_lines = [
                f"{item.label} {item.count} 笔"
                for item in query_payload.status_groups
                if item.count > 0
            ]
            answer_parts = [
                f"我先为你列出{window_prefix}的{query_payload.scope_label}，"
                f"共 {query_payload.record_count} 笔，金额合计 {query_payload.total_amount:.2f} 元。"
            ]
            if group_lines:
                answer_parts.append(f"其中包括：{'、'.join(group_lines)}。")

            hint_parts: list[str] = []
            if query_payload.has_more_in_window and query_payload.preview_count < query_payload.record_count:
                hint_parts.append(
                    f"下方先展示最近 {query_payload.preview_count} 笔，你可以直接点击单据查看详情。"
                )
            elif query_payload.records:
                hint_parts.append("下方已列出本次命中的真实单据，可直接点击查看详情。")

            if query_payload.older_record_count > 0 and query_payload.window_days:
                hint_parts.append(
                    f"另有 {query_payload.older_record_count} 笔超过 {query_payload.window_days} 日的单据，"
                    "请前往个人报销中心查看。"
                )

            return " ".join(answer_parts + hint_parts).strip()

        if scenario == "accounts_receivable":
            record_count = int(data.get("record_count") or 0)
            outstanding_amount = float(data.get("outstanding_amount") or 0)
            return (
                f"{subject}共命中 {record_count} 条应收，未回款金额 {outstanding_amount:.2f} 元。"
                "建议结合账龄和客户分布继续排查逾期风险。"
            )

        if scenario == "accounts_payable":
            record_count = int(data.get("record_count") or 0)
            outstanding_amount = float(data.get("outstanding_amount") or 0)
            return (
                f"{subject}共命中 {record_count} 条应付，待付金额 {outstanding_amount:.2f} 元。"
                "如需推进动作，建议先生成付款建议草稿并发起人工确认。"
            )

        return "已完成当前查询，但暂时没有更多结构化结果可展示。"

    def _build_query_payload(
        self,
        payload: UserAgentRequest,
    ) -> UserAgentQueryPayload | None:
        if payload.ontology.scenario != "expense" or payload.ontology.intent not in {"query", "compare"}:
            return None

        result_type = str(payload.tool_payload.get("result_type") or "").strip()
        if result_type and result_type != "expense_claim_list":
            return None

        records: list[UserAgentExpenseQueryRecord] = []
        for item in payload.tool_payload.get("records") or []:
            if not isinstance(item, dict):
                continue
            amount = float(item.get("amount") or 0)
            records.append(
                UserAgentExpenseQueryRecord(
                    claim_id=str(item.get("claim_id") or "").strip(),
                    claim_no=str(item.get("claim_no") or "").strip() or "未编号",
                    employee_name=str(item.get("employee_name") or "").strip(),
                    expense_type=str(item.get("expense_type") or "").strip(),
                    expense_type_label=str(item.get("expense_type_label") or "").strip()
                    or EXPENSE_TYPE_LABELS.get(str(item.get("expense_type") or "").strip(), "报销"),
                    amount=round(amount, 2),
                    status=str(item.get("status") or "").strip(),
                    status_label=str(item.get("status_label") or "").strip()
                    or EXPENSE_STATUS_LABELS.get(str(item.get("status") or "").strip(), "处理中"),
                    status_group=str(item.get("status_group") or "").strip() or "other",
                    status_group_label=str(item.get("status_group_label") or "").strip()
                    or EXPENSE_STATUS_GROUP_LABELS.get(str(item.get("status_group") or "").strip(), "其他状态"),
                    approval_stage=str(item.get("approval_stage") or "").strip() or None,
                    document_date=str(item.get("document_date") or "").strip(),
                    occurred_at=str(item.get("occurred_at") or "").strip(),
                    reason=str(item.get("reason") or "").strip(),
                    location=str(item.get("location") or "").strip(),
                )
            )

        status_groups: list[UserAgentQueryStatusGroup] = []
        for item in payload.tool_payload.get("status_groups") or []:
            if not isinstance(item, dict):
                continue
            status_groups.append(
                UserAgentQueryStatusGroup(
                    key=str(item.get("key") or "").strip() or "other",
                    label=str(item.get("label") or "").strip() or "其他状态",
                    count=max(0, int(item.get("count") or 0)),
                )
            )

        return UserAgentQueryPayload(
            result_type="expense_claim_list",
            scope_label=str(payload.tool_payload.get("scope_label") or self._resolve_subject(payload)).strip() or "报销单",
            recent_window_applied=bool(payload.tool_payload.get("recent_window_applied")),
            window_days=(
                int(payload.tool_payload["window_days"])
                if payload.tool_payload.get("window_days") not in {None, ""}
                else None
            ),
            window_start_date=(
                str(payload.tool_payload.get("window_start_date") or "").strip() or None
            ),
            window_end_date=(
                str(payload.tool_payload.get("window_end_date") or "").strip() or None
            ),
            record_count=max(0, int(payload.tool_payload.get("record_count") or 0)),
            preview_count=max(0, int(payload.tool_payload.get("preview_count") or len(records))),
            older_record_count=max(0, int(payload.tool_payload.get("older_record_count") or 0)),
            has_more_in_window=bool(payload.tool_payload.get("has_more_in_window") or payload.tool_payload.get("has_more")),
            total_amount=round(float(payload.tool_payload.get("total_amount") or 0), 2),
            status_groups=status_groups,
            records=records,
        )

    def _build_fast_knowledge_answer(
        self,
        payload: UserAgentRequest,
        *,
        citations: list[UserAgentCitation],
    ) -> str | None:
        if payload.ontology.scenario != "knowledge":
            return None
        if str(payload.tool_payload.get("result_type") or "").strip() != "knowledge_search":
            return None

        evidence_items = self._build_knowledge_answer_evidence(payload)
        if not evidence_items:
            return None

        question = self._resolve_knowledge_question(payload)
        if not self._should_use_direct_knowledge_answer(question, evidence_items):
            return None

        return self._render_knowledge_direct_answer(
            payload,
            citations=citations,
            evidence_items=evidence_items,
        )

    def _render_knowledge_direct_answer(
        self,
        payload: UserAgentRequest,
        *,
        citations: list[UserAgentCitation],
        evidence_items: list[dict[str, Any]],
    ) -> str | None:
        if not evidence_items:
            return None

        title = str(
            (citations[0].title if citations else "")
            or evidence_items[0].get("title")
            or "相关制度"
        ).strip()
        user_name = str(payload.context_json.get("name") or "").strip()
        question = self._resolve_knowledge_question(payload)
        query_terms = self._extract_knowledge_query_terms(question)
        ordered_evidence_items = self._prioritize_knowledge_evidence_items(question, evidence_items)
        primary_item = ordered_evidence_items[0]
        primary_heading = self._format_knowledge_heading_label(
            str(primary_item.get("heading") or "").strip()
        )
        primary_lines = self._collect_direct_knowledge_answer_lines(ordered_evidence_items)

        lines: list[str] = []
        if user_name:
            lines.append(f"{user_name}，您好。")
        source_prefix = f"根据《{title}》"
        if primary_heading:
            source_prefix = f"{source_prefix}（{primary_heading}）"

        if str(primary_item.get("kind") or "") == "table":
            lines.append(f"{source_prefix}，当前能直接确认的是：")
            lines.append(self._extract_relevant_table_preview(str(primary_item.get("content") or ""), query_terms))
        else:
            if not primary_lines:
                lines.append(
                    f"{source_prefix}，当前能直接确认的是："
                    f"{self._summarize_knowledge_evidence_content(primary_item, query_terms)}"
                )
            elif len(primary_lines) == 1:
                lines.append(f"{source_prefix}，当前能直接确认的是：{primary_lines[0].strip()}")
            else:
                lines.append(f"{source_prefix}，当前能直接确认的是：")
                lines.extend(primary_lines)

        notes: list[str] = []
        location_note = self._build_missing_location_grounding_note(question, evidence_items)
        if location_note:
            notes.append(location_note)
        if self._question_requires_explicit_condition(question) and not self._answer_evidence_has_numeric_or_condition(evidence_items):
            notes.append("当前命中的证据更偏规则说明或流程约束，还没有直接给出可立即套用的数值或完整条件。")

        if notes:
            lines.append("")
            lines.append("说明：")
            lines.extend(f"- {note}" for note in notes)

        return "\n".join(line for line in lines if line is not None).strip()

    def _prioritize_knowledge_evidence_items(
        self,
        question: str,
        evidence_items: list[dict[str, Any]],
    ) -> list[dict[str, Any]]:
        if not evidence_items or not self._question_requires_explicit_condition(question):
            return evidence_items

        for preferred_kind in ("table", "kv", "clause", "list"):
            for index, item in enumerate(evidence_items):
                if str(item.get("kind") or "") != preferred_kind:
                    continue
                return [item, *evidence_items[:index], *evidence_items[index + 1 :]]

        for index, item in enumerate(evidence_items):
            if re.search(r"\d", str(item.get("content") or "")):
                return [item, *evidence_items[:index], *evidence_items[index + 1 :]]

        return evidence_items

    @staticmethod
    def _resolve_knowledge_question(payload: UserAgentRequest) -> str:
        return str(payload.context_json.get("user_input_text") or payload.message or "").strip()

    @staticmethod
    def _looks_like_structured_knowledge_query(question: str) -> bool:
        normalized = str(question or "").strip()
        if not normalized:
            return False
        return any(keyword in normalized for keyword in KNOWLEDGE_DIRECT_ANSWER_HINTS)

    def _should_use_direct_knowledge_answer(
        self,
        question: str,
        evidence_items: list[dict[str, Any]],
    ) -> bool:
        if not evidence_items:
            return False
        if self._looks_like_structured_knowledge_query(question):
            return True
        return str(evidence_items[0].get("kind") or "") in {"table", "kv", "list", "clause"}

    def _build_knowledge_answer_evidence(
        self,
        payload: UserAgentRequest,
    ) -> list[dict[str, Any]]:
        question = self._resolve_knowledge_question(payload)
        query_terms = self._extract_knowledge_query_terms(question)
        candidates: list[dict[str, Any]] = []

        for hit in self._select_knowledge_model_hits(
            payload.tool_payload,
            question=question,
        ):
            if not isinstance(hit, dict):
                continue
            candidates.extend(self._extract_knowledge_evidence_candidates(hit, query_terms))

        deduped: list[dict[str, Any]] = []
        seen: set[tuple[str, str, str]] = set()
        ranked_candidates = sorted(
            candidates,
            key=lambda value: (
                float(value.get("score") or 0),
                -len(str(value.get("content") or "")),
            ),
            reverse=True,
        )
        top_score = float(ranked_candidates[0].get("score") or 0) if ranked_candidates else 0.0

        for item in ranked_candidates:
            score = float(item.get("score") or 0)
            if deduped and score < max(6.0, top_score - 14):
                continue
            key = (
                str(item.get("title") or "").strip(),
                str(item.get("heading") or "").strip(),
                self._clean_knowledge_segment_text(str(item.get("content") or ""))[:180],
            )
            if key in seen:
                continue
            seen.add(key)
            deduped.append(item)
            if len(deduped) >= MAX_KNOWLEDGE_DIRECT_EVIDENCE:
                break
        return deduped

    def _extract_knowledge_evidence_candidates(
        self,
        hit: dict[str, Any],
        query_terms: list[str],
    ) -> list[dict[str, Any]]:
        title = str(hit.get("title") or hit.get("document_name") or "相关制度").strip()
        content = str(hit.get("content") or "").strip()
        if not content:
            return []

        raw_candidates = self._merge_knowledge_lead_in_segments(
            self._split_knowledge_hit_into_segments(content)
        )
        candidates: list[dict[str, Any]] = []
        for item in raw_candidates:
            score = self._score_knowledge_evidence_candidate(item, query_terms)
            if query_terms and score <= 0:
                continue
            normalized = dict(item)
            normalized["title"] = title
            normalized["score"] = score
            candidates.append(normalized)

        if candidates:
            return candidates

        fallback_text = str(hit.get("excerpt") or "").strip() or self._extract_excerpt(content)
        if not fallback_text:
            return []
        return [
            {
                "title": title,
                "heading": "",
                "kind": "paragraph",
                "content": fallback_text,
                "score": 1,
            }
        ]

    @staticmethod
    def _is_knowledge_lead_in_segment(item: dict[str, str]) -> bool:
        kind = str(item.get("kind") or "").strip()
        content = str(item.get("content") or "").strip()
        return kind in {"kv", "list", "clause"} and content.endswith(("：", ":"))

    @staticmethod
    def _extract_knowledge_marker_family(content: str) -> str:
        normalized = str(content or "").strip()
        if not normalized:
            return ""
        if KNOWLEDGE_ARTICLE_PATTERN.match(normalized):
            return "article"
        if re.match(r"^\d+[.)、]\s*", normalized):
            return "arabic"
        if re.match(r"^[（(][一二三四五六七八九十百零0-9]+[)）]\s*", normalized):
            return "paren"
        if re.match(r"^[①②③④⑤⑥⑦⑧⑨⑩]\s*", normalized):
            return "circled"
        if KNOWLEDGE_LIST_ITEM_PATTERN.match(normalized):
            return "bullet"
        return ""

    @staticmethod
    def _format_knowledge_heading_label(heading: str) -> str:
        parts = [item.strip() for item in str(heading or "").split(">") if item.strip()]
        return " / ".join(parts)

    def _merge_knowledge_lead_in_segments(
        self,
        segments: list[dict[str, str]],
    ) -> list[dict[str, str]]:
        if not segments:
            return []

        merged: list[dict[str, str]] = []
        index = 0
        while index < len(segments):
            current = dict(segments[index])
            if not self._is_knowledge_lead_in_segment(current):
                merged.append(current)
                index += 1
                continue

            base_heading = str(current.get("heading") or "").strip()
            current_marker = self._extract_knowledge_marker_family(str(current.get("content") or ""))
            follow_segments: list[dict[str, str]] = []
            next_index = index + 1

            while next_index < len(segments):
                candidate = segments[next_index]
                if str(candidate.get("heading") or "").strip() != base_heading:
                    break

                candidate_kind = str(candidate.get("kind") or "").strip()
                candidate_content = str(candidate.get("content") or "").strip()
                candidate_marker = self._extract_knowledge_marker_family(candidate_content)
                if not candidate_content or candidate_kind == "table":
                    break
                if current_marker and candidate_marker == current_marker:
                    break
                if self._is_knowledge_lead_in_segment(candidate) and follow_segments:
                    break
                if candidate_kind not in {"list", "paragraph", "kv", "clause"}:
                    break

                follow_segments.append(candidate)
                next_index += 1
                if len(follow_segments) >= 4:
                    break
                if candidate_kind == "paragraph" and len(candidate_content) >= 200:
                    break

            if follow_segments:
                current["content"] = "\n".join(
                    [str(current.get("content") or "").strip()]
                    + [str(item.get("content") or "").strip() for item in follow_segments]
                )
                if any(str(item.get("kind") or "").strip() == "list" for item in follow_segments):
                    current["kind"] = "list"
                merged.append(current)
                index = next_index
                continue

            merged.append(current)
            index += 1

        return merged

    def _split_knowledge_hit_into_segments(self, content: str) -> list[dict[str, str]]:
        segments: list[dict[str, str]] = []
        markdown_headings: list[str] = []
        section_heading = ""
        paragraph_lines: list[str] = []
        table_lines: list[str] = []

        def current_heading() -> str:
            heading_parts = [item for item in markdown_headings if item]
            if section_heading:
                heading_parts.append(section_heading)
            return " > ".join(heading_parts)

        def flush_paragraph() -> None:
            nonlocal paragraph_lines
            if not paragraph_lines:
                return
            merged = " ".join(line.strip() for line in paragraph_lines if line.strip()).strip()
            paragraph_lines = []
            if merged:
                segments.append(
                    {
                        "heading": current_heading(),
                        "kind": "paragraph",
                        "content": merged,
                    }
                )

        def flush_table() -> None:
            nonlocal table_lines
            if not table_lines:
                return
            merged = "\n".join(line.rstrip() for line in table_lines if line.strip()).strip()
            table_lines = []
            if merged:
                segments.append(
                    {
                        "heading": current_heading(),
                        "kind": "table",
                        "content": merged,
                    }
                )

        for raw_line in str(content or "").replace("\r\n", "\n").replace("\r", "\n").splitlines():
            line = raw_line.rstrip()
            stripped = line.strip()

            if not stripped:
                flush_paragraph()
                flush_table()
                continue

            markdown_heading_match = re.match(r"^(#{1,6})\s+(.+)$", stripped)
            if markdown_heading_match:
                flush_paragraph()
                flush_table()
                level = len(markdown_heading_match.group(1))
                heading_text = markdown_heading_match.group(2).strip()
                markdown_headings = markdown_headings[: max(0, level - 1)]
                markdown_headings.append(heading_text)
                section_heading = ""
                continue

            if KNOWLEDGE_SECTION_HEADING_PATTERN.match(stripped) and len(stripped) <= 90:
                flush_paragraph()
                flush_table()
                section_heading = stripped.lstrip("#").strip()
                continue

            if stripped.count("|") >= 2 and "|" in stripped:
                flush_paragraph()
                table_lines.append(stripped)
                continue

            flush_table()

            if KNOWLEDGE_LIST_ITEM_PATTERN.match(stripped):
                flush_paragraph()
                segments.append(
                    {
                        "heading": current_heading(),
                        "kind": "list",
                        "content": stripped,
                    }
                )
                continue

            if KNOWLEDGE_NUMBERED_ITEM_PATTERN.match(stripped):
                flush_paragraph()
                segments.append(
                    {
                        "heading": current_heading(),
                        "kind": "list",
                        "content": stripped,
                    }
                )
                continue

            if KNOWLEDGE_ARTICLE_PATTERN.match(stripped):
                flush_paragraph()
                segments.append(
                    {
                        "heading": current_heading(),
                        "kind": "clause",
                        "content": stripped,
                    }
                )
                continue

            if ("：" in stripped or ":" in stripped) and len(stripped) <= 180:
                flush_paragraph()
                segments.append(
                    {
                        "heading": current_heading(),
                        "kind": "kv",
                        "content": stripped,
                    }
                )
                continue

            paragraph_lines.append(stripped)

        flush_paragraph()
        flush_table()
        return segments

    def _score_knowledge_evidence_candidate(
        self,
        item: dict[str, str],
        query_terms: list[str],
    ) -> int:
        heading = str(item.get("heading") or "").lower()
        content = str(item.get("content") or "").lower()
        kind = str(item.get("kind") or "").strip()
        haystack = "\n".join([heading, content])

        matched_terms = [term for term in query_terms if term in haystack]
        score = len(matched_terms) * 10
        score += sum(1 for term in matched_terms if term in heading) * 6

        if kind == "table":
            score += 10
        elif kind in {"kv", "clause", "list"}:
            score += 8
        elif kind == "paragraph":
            score += 4

        if "问答线索补充" in heading or "重点章节摘录" in heading:
            score += 8
        if "结构化表格补充" in heading:
            score += 10
        if "章节导航" in heading or "目录" in heading:
            score -= 16
        if re.search(r"[.。…]{6,}", content):
            score -= 12
        if any(hint in content for hint in ("应", "需", "不得", "可以", "标准", "条件", "材料", "审批", "流程", "包括")):
            score += 3

        content_length = len(content)
        if content_length > 220:
            score -= min(8, (content_length - 220) // 40)
        return score

    @staticmethod
    def _extract_knowledge_query_terms(question: str) -> list[str]:
        normalized_question = str(question or "").strip().lower()
        if not normalized_question:
            return []

        terms: list[str] = []
        seen: set[str] = set()

        def remember(term: str) -> None:
            normalized = str(term or "").strip().lower()
            if (
                not normalized
                or normalized in seen
                or normalized in KNOWLEDGE_QUERY_STOPWORDS
            ):
                return
            seen.add(normalized)
            terms.append(normalized)

        for item in re.findall(r"[a-z0-9][a-z0-9_\-]{1,}", normalized_question):
            remember(item)

        for block in re.findall(r"[\u4e00-\u9fff]{2,20}", normalized_question):
            if len(block) <= 4:
                remember(block)
                continue
            for size in (4, 3, 2):
                for start in range(0, len(block) - size + 1):
                    remember(block[start : start + size])
                    if len(terms) >= MAX_KNOWLEDGE_QUERY_TERMS:
                        return terms

        return terms[:MAX_KNOWLEDGE_QUERY_TERMS]

    @staticmethod
    def _clean_knowledge_segment_text(content: str) -> str:
        normalized = str(content or "").strip()
        normalized = re.sub(r"^[-*•]\s*", "", normalized)
        normalized = re.sub(r"^(?:\d+[.)、]|[①②③④⑤⑥⑦⑧⑨⑩])\s*", "", normalized)
        normalized = re.sub(r"^[（(][一二三四五六七八九十百零0-9]+[)）]\s*", "", normalized)
        normalized = re.sub(r"\s+", " ", normalized)
        if len(normalized) <= 180:
            return normalized
        return f"{normalized[:177].rstrip()}..."

    @staticmethod
    def _normalize_knowledge_line(content: str, *, preserve_marker: bool) -> str:
        normalized = str(content or "").strip()
        normalized = re.sub(r"^[-*•]\s*", "", normalized)
        if not preserve_marker:
            normalized = re.sub(r"^(?:\d+[.)、]|[①②③④⑤⑥⑦⑧⑨⑩])\s*", "", normalized)
            normalized = re.sub(r"^[（(][一二三四五六七八九十百零0-9]+[)）]\s*", "", normalized)
        normalized = re.sub(r"\s+", " ", normalized)
        return normalized

    def _split_clean_knowledge_lines(
        self,
        content: str,
        *,
        preserve_marker: bool,
    ) -> list[str]:
        return [
            line
            for line in (
                self._normalize_knowledge_line(item, preserve_marker=preserve_marker)
                for item in str(content or "").splitlines()
            )
            if line
        ]

    def _render_knowledge_evidence_text(self, item: dict[str, Any]) -> str:
        lines = self._split_clean_knowledge_lines(
            str(item.get("content") or ""),
            preserve_marker=True,
        )
        if not lines:
            return ""
        if len(lines) == 1:
            return self._clean_knowledge_segment_text(lines[0])
        return "\n".join(f"  {line}" for line in lines)

    def _collect_direct_knowledge_answer_lines(
        self,
        ordered_evidence_items: list[dict[str, Any]],
    ) -> list[str]:
        if not ordered_evidence_items:
            return []

        primary_item = ordered_evidence_items[0]
        primary_title = str(primary_item.get("title") or "").strip()
        primary_heading = str(primary_item.get("heading") or "").strip()
        primary_kind = str(primary_item.get("kind") or "").strip()

        related_items = [primary_item]
        if primary_kind != "table":
            for item in ordered_evidence_items[1:]:
                if len(related_items) >= 3:
                    break
                if str(item.get("kind") or "").strip() != primary_kind:
                    continue
                if str(item.get("title") or "").strip() != primary_title:
                    continue
                if str(item.get("heading") or "").strip() != primary_heading:
                    continue
                related_items.append(item)

        lines: list[str] = []
        seen: set[str] = set()
        for item in related_items:
            rendered = self._render_knowledge_evidence_text(item)
            for line in rendered.splitlines():
                normalized = str(line or "").strip()
                if not normalized or normalized in seen:
                    continue
                seen.add(normalized)
                lines.append(line)
        return lines

    def _summarize_knowledge_evidence_content(
        self,
        item: dict[str, Any],
        query_terms: list[str],
    ) -> str:
        kind = str(item.get("kind") or "").strip()
        content = str(item.get("content") or "").strip()
        if kind == "table":
            preview = self._extract_relevant_table_preview(content, query_terms)
            preview_rows = [line for line in preview.splitlines() if line.strip()][:4]
            if len(preview_rows) >= 3:
                return "当前命中的直接依据是一张与问题强相关的标准表，已摘出最相关的表头和行。"
            return "当前命中的直接依据是一张与问题强相关的标准表。"
        lines = self._split_clean_knowledge_lines(content, preserve_marker=True)
        if len(lines) >= 2:
            return self._clean_knowledge_segment_text(f"{lines[0]} {' '.join(lines[1:4])}")
        return self._clean_knowledge_segment_text(content)

    @staticmethod
    def _extract_relevant_table_preview(content: str, query_terms: list[str]) -> str:
        lines = [line.strip() for line in str(content or "").splitlines() if line.strip()]
        if len(lines) <= 3:
            return "\n".join(lines)

        header = lines[0]
        divider = lines[1] if len(lines) > 1 else ""
        body = lines[2:] if divider.count("|") >= 2 else lines[1:]

        matched_rows = [
            row
            for row in body
            if any(term in row.lower() for term in query_terms)
        ]
        selected_rows = matched_rows[:3] or body[:2]
        preview_lines = [header]
        if divider:
            preview_lines.append(divider)
        preview_lines.extend(selected_rows)
        return "\n".join(preview_lines).strip()

    @staticmethod
    def _question_requires_explicit_condition(question: str) -> bool:
        normalized = str(question or "").strip()
        return any(keyword in normalized for keyword in ("多少", "金额", "上限", "限额", "标准", "条件", "需要"))

    def _build_missing_location_grounding_note(
        self,
        question: str,
        evidence_items: list[dict[str, Any]],
    ) -> str:
        location = self._extract_query_location(question)
        if not location:
            return ""

        haystack = "\n".join(
            str(item.get("heading") or "") + "\n" + str(item.get("content") or "")
            for item in evidence_items
        )
        if location in haystack:
            return ""
        return (
            f"当前命中的制度依据没有直接写出“{location}”对应的地区档位或映射关系，"
            "因此不能直接把它套用到表格中的某一列。"
        )

    @staticmethod
    def _answer_evidence_has_numeric_or_condition(evidence_items: list[dict[str, Any]]) -> bool:
        for item in evidence_items:
            content = str(item.get("content") or "")
            if re.search(r"\d", content):
                return True
            if any(
                keyword in content
                for keyword in ("应", "需", "不得", "可以", "条件", "材料", "审批", "流程", "标准", "适用")
            ):
                return True
        return False

    def _build_explain_answer(
        self,
        payload: UserAgentRequest,
        citations: list[UserAgentCitation],
    ) -> str:
        if str(payload.tool_payload.get("result_type") or "").strip() == "knowledge_search":
            if citations:
                return self._build_knowledge_search_answer(payload, citations)

            tool_message = str(payload.tool_payload.get("message") or "").strip()
            if tool_message:
                return tool_message

        if citations:
            titles = "、".join(item.title for item in citations[:2])
            summary = citations[0].excerpt or "请结合制度全文进一步确认。"
            return f"已检索到相关依据：{titles}。核心说明：{summary}"

        return (
            f"当前还没有与“{SCENARIO_LABELS.get(payload.ontology.scenario, '当前问题')}”"
            "强匹配的已上线规则引用，建议先人工复核或补充更具体的单据上下文。"
        )

    def _build_knowledge_search_answer(
        self,
        payload: UserAgentRequest,
        citations: list[UserAgentCitation],
    ) -> str:
        hits = [item for item in list(payload.tool_payload.get("hits") or []) if isinstance(item, dict)]
        evidence_items = self._build_knowledge_answer_evidence(payload)
        primary_citation = citations[0] if citations else None
        title = str(
            (primary_citation.title if primary_citation else "")
            or (hits[0].get("title") if hits else "")
            or "相关制度"
        ).strip()
        user_name = str(payload.context_json.get("name") or "").strip()
        prefix = f"{user_name}，您好。\n" if user_name else ""
        if not hits:
            return (
                f"{prefix}我已经从《{title}》中检索到与你这次问题相关的制度依据，"
                "但本次答案生成环节暂时没有成功返回。请稍后重试一次；如果仍然失败，"
                "建议先检查主对话模型的连通性。"
            )

        evidence_lines: list[str] = []
        for item in evidence_items[:3]:
            heading = str(item.get("heading") or "").strip()
            heading_text = f" > {heading}" if heading else ""
            if str(item.get("kind") or "") == "table":
                preview = self._extract_relevant_table_preview(
                    str(item.get("content") or ""),
                    self._extract_knowledge_query_terms(self._resolve_knowledge_question(payload)),
                )
                evidence_lines.append(f"- 《{item.get('title') or title}》{heading_text}：\n{preview}")
                continue
            rendered = self._render_knowledge_evidence_text(item)
            if rendered:
                if "\n" in rendered:
                    evidence_lines.append(f"- 《{item.get('title') or title}》{heading_text}：\n{rendered}")
                else:
                    evidence_lines.append(f"- 《{item.get('title') or title}》{heading_text}：{rendered}")

        if not evidence_lines:
            for item in hits[:2]:
                item_title = str(item.get("title") or item.get("document_name") or "相关制度").strip()
                excerpt = (
                    str(item.get("excerpt") or "").strip()
                    or self._extract_excerpt(str(item.get("content") or ""))
                )
                if not excerpt:
                    continue
                evidence_lines.append(f"- 《{item_title}》：{excerpt}")

        if not evidence_lines:
            return (
                f"{prefix}我已经从《{title}》中检索到与你这次问题相关的制度依据，"
                "但本次答案生成环节暂时没有成功返回。请稍后重试一次；如果仍然失败，"
                "建议先检查主对话模型的连通性。"
            )

        return "\n".join(
            [
                f"{prefix}我已经命中与你这次问题最相关的制度依据，但答案整理阶段本轮没有及时返回。",
                "先给你当前最直接的依据：",
                *evidence_lines,
                "如果你希望我继续把这些依据整理成更完整的结论、步骤或对比说明，可以继续缩小问题范围后再问一次。",
            ]
        ).strip()

    def _build_risk_answer(
        self,
        payload: UserAgentRequest,
        citations: list[UserAgentCitation],
    ) -> str:
        risk_flags = self._resolve_risk_flags(payload)
        platform_messages = self._evaluate_platform_risk_messages(payload)
        if not risk_flags and not platform_messages:
            return "当前未识别到明确风险标签，建议继续查看原始明细或补充更多上下文。"

        reasons = [RISK_REASON_MAP.get(flag, f"{flag} 需要人工进一步确认。") for flag in risk_flags]
        if platform_messages:
            reasons.extend(platform_messages)
        citation_text = (
            f" 参考规则：{'、'.join(item.title for item in citations[:2])}。"
            if citations
            else ""
        )
        signal_count = len(risk_flags) + (1 if platform_messages else 0)
        return (
            f"本次识别到 {signal_count} 类风险信号。"
            f"触发原因：{'；'.join(reasons)}。"
            "建议先复核明细、附件和审批链，再决定是否继续处理。"
            f"{citation_text}"
        )

    def _evaluate_platform_risk_messages(self, payload: UserAgentRequest) -> list[str]:
        claim_id = str(payload.tool_payload.get("claim_id") or "").strip()
        if not claim_id:
            return []

        claim = self.db.scalar(
            select(ExpenseClaim)
            .where(ExpenseClaim.id == claim_id)
            .options(selectinload(ExpenseClaim.items))
        )
        if claim is None:
            return []

        rule_codes = resolve_rule_codes_for_risk_check(
            payload.ontology,
            query_text=payload.message,
        )
        review = ExpenseClaimService(self.db).evaluate_platform_risk_rules(
            claim,
            rule_codes=rule_codes,
        )
        messages: list[str] = []
        for flag in review.get("flags") or []:
            if not isinstance(flag, dict):
                continue
            message = str(flag.get("message") or "").strip()
            if message and message not in messages:
                messages.append(message)
        return messages

    def _build_draft_payload(self, payload: UserAgentRequest) -> UserAgentDraftPayload:
        scenario_label = SCENARIO_LABELS.get(payload.ontology.scenario, "业务")
        subject = self._resolve_subject(payload)
        claim_no = str(payload.tool_payload.get("claim_no") or "").strip() or None
        claim_status = str(payload.tool_payload.get("status") or "").strip() or None
        approval_stage = str(payload.tool_payload.get("approval_stage") or "").strip() or None
        is_submitted = claim_status == "submitted"
        title = f"{scenario_label}处理意见草稿"
        if claim_no:
            title = f"{scenario_label}{'报销单' if is_submitted else '草稿'} {claim_no}"
        if is_submitted:
            body = (
                f"主题：{subject}\n"
                f"结论：报销单已提交，当前节点为 {approval_stage or '审批中'}。\n"
                "建议：后续可在个人报销列表中跟踪审批进度，必要时再补充说明或附件。\n"
                f"原始问题：{payload.message}"
            )
        else:
            body = (
                f"主题：{subject}\n"
                "结论：已根据当前语义解析结果生成草稿，尚未自动执行。\n"
                "建议：请先核对明细、规则命中和所需附件，再由人工确认是否提交正式流程。\n"
                f"原始问题：{payload.message}"
            )
        return UserAgentDraftPayload(
            draft_type=payload.ontology.scenario,
            title=title,
            body=body,
            confirmation_required=not is_submitted,
            claim_id=str(payload.tool_payload.get("claim_id") or "").strip() or None,
            claim_no=claim_no,
            status=claim_status,
            approval_stage=approval_stage,
        )

    def _build_suggested_actions(
        self,
        payload: UserAgentRequest,
    ) -> list[UserAgentSuggestedAction]:
        if payload.ontology.scenario == "knowledge":
            return []

        if self._is_generic_expense_prompt(payload):
            return [
                UserAgentSuggestedAction(
                    label="上传票据",
                    action_type="ask_clarification",
                    description="上传发票、行程单或付款截图，继续识别报销内容。",
                ),
                UserAgentSuggestedAction(
                    label="补充报销信息",
                    action_type="ask_clarification",
                    description="补充费用类型、金额、时间和事由后继续处理。",
                ),
            ]

        if payload.ontology.intent in {"query", "compare"}:
            return [
                UserAgentSuggestedAction(
                    label="查看明细",
                    action_type="open_detail",
                    description="继续查看命中记录和过滤条件。",
                ),
                UserAgentSuggestedAction(
                    label="生成处理意见",
                    action_type="create_draft",
                    description="把当前查询结果整理成可确认草稿。",
                ),
            ]

        if payload.ontology.intent == "risk_check":
            return [
                UserAgentSuggestedAction(
                    label="人工复核风险",
                    action_type="manual_review",
                    description="优先检查明细、附件和规则命中原因。",
                ),
                UserAgentSuggestedAction(
                    label="生成整改建议",
                    action_type="create_draft",
                    description="把风险说明整理成处理意见草稿。",
                ),
            ]

        if payload.ontology.intent == "draft":
            return [
                UserAgentSuggestedAction(
                    label="复制草稿",
                    action_type="copy_draft",
                    description="复制当前草稿后交由人工确认。",
                ),
                UserAgentSuggestedAction(
                    label="补充上下文",
                    action_type="ask_clarification",
                    description="补充单据编号、客户或供应商信息以完善草稿。",
                ),
            ]

        return [
            UserAgentSuggestedAction(
                label="查看规则全文",
                action_type="open_rule",
                description="继续查看引用规则或知识内容。",
            ),
            UserAgentSuggestedAction(
                label="补充问题上下文",
                action_type="ask_clarification",
                description="补充业务对象、时间或单据范围，提升回答准确度。",
            ),
        ]

    def _build_review_payload(
        self,
        payload: UserAgentRequest,
        *,
        citations: list[UserAgentCitation],
        draft_payload: UserAgentDraftPayload | None,
    ) -> UserAgentReviewPayload | None:
        attachment_count = self._resolve_attachment_count(payload)
        ocr_documents = self._resolve_ocr_documents(payload)
        if payload.ontology.scenario != "expense":
            return None
        if payload.ontology.intent not in {"draft", "operate"} and attachment_count <= 0 and not ocr_documents:
            return None

        document_cards = self._build_review_document_cards(payload, ocr_documents=ocr_documents)
        claim_groups = self._build_review_claim_groups(
            payload,
            document_cards=document_cards,
        )
        slot_cards = self._build_review_slot_cards(
            payload,
            ocr_documents=ocr_documents,
            claim_groups=claim_groups,
        )
        missing_slot_keys = self._resolve_review_missing_slot_keys(
            payload,
            slot_cards=slot_cards,
        )
        risk_briefs = self._build_review_risk_briefs(
            payload,
            citations=citations,
            document_cards=document_cards,
            claim_groups=claim_groups,
        )
        association_choice_pending = self._is_review_association_choice_pending(payload)
        can_proceed = (
            False
            if association_choice_pending
            else self._can_proceed_review(
                payload,
                missing_slot_keys=missing_slot_keys,
                claim_groups=claim_groups,
            )
        )
        confirmation_actions = self._build_review_confirmation_actions(
            payload,
            can_proceed=can_proceed,
            claim_groups=claim_groups,
            draft_payload=draft_payload,
        )
        edit_fields = self._build_review_edit_fields(
            payload,
            draft_payload=draft_payload,
            slot_cards=slot_cards,
        )
        intent_summary = self._build_review_intent_summary(
            payload,
            slot_cards=slot_cards,
            claim_groups=claim_groups,
        )
        body_message = self._build_review_body_message(
            payload,
            slot_cards=slot_cards,
            risk_briefs=risk_briefs,
            can_proceed=can_proceed,
            document_cards=document_cards,
        )

        return UserAgentReviewPayload(
            intent_summary=intent_summary,
            body_message=body_message,
            scenario=payload.ontology.scenario,
            intent=payload.ontology.intent,
            can_proceed=can_proceed,
            missing_slots=[SLOT_LABELS.get(key, key) for key in missing_slot_keys],
            risk_briefs=risk_briefs,
            slot_cards=slot_cards,
            document_cards=document_cards,
            claim_groups=claim_groups,
            confirmation_actions=confirmation_actions,
            edit_fields=edit_fields,
        )

    def _build_review_slot_cards(
        self,
        payload: UserAgentRequest,
        *,
        ocr_documents: list[dict[str, object]],
        claim_groups: list[UserAgentReviewClaimGroup],
    ) -> list[UserAgentReviewSlotCard]:
        entity_map = self._collect_entity_values(payload)
        time_slot = self._build_time_slot(payload)
        location_slot = self._build_location_slot(payload)
        customer_slot = self._build_customer_slot(payload, entity_map=entity_map)
        participants_slot = self._build_participants_slot(payload, entity_map=entity_map)
        amount_slot = self._build_amount_slot(payload, entity_map=entity_map, ocr_documents=ocr_documents)
        expense_type_slot = self._build_expense_type_slot(
            payload,
            entity_map=entity_map,
            ocr_documents=ocr_documents,
        )
        merchant_slot = self._build_merchant_slot(payload, ocr_documents=ocr_documents)
        reason_slot = self._build_reason_slot(
            payload,
            claim_groups=claim_groups,
        )
        attachment_slot = self._build_attachment_slot(payload)
        required_keys = self._resolve_required_review_keys(
            payload,
            primary_expense_type=str(expense_type_slot["normalized_value"] or ""),
            claim_groups=claim_groups,
        )

        cards = [
            self._make_slot_card(
                key="expense_type",
                value=expense_type_slot["value"],
                raw_value=expense_type_slot["raw_value"],
                normalized_value=expense_type_slot["normalized_value"],
                source=expense_type_slot["source"],
                confidence=expense_type_slot["confidence"],
                evidence=expense_type_slot["evidence"],
                required="expense_type" in required_keys,
            ),
            self._make_slot_card(
                key="customer_name",
                value=customer_slot["value"],
                raw_value=customer_slot["raw_value"],
                normalized_value=customer_slot["normalized_value"],
                source=customer_slot["source"],
                confidence=customer_slot["confidence"],
                evidence=customer_slot["evidence"],
                required="customer_name" in required_keys,
            ),
            self._make_slot_card(
                key="time_range",
                value=time_slot["value"],
                raw_value=time_slot["raw_value"],
                normalized_value=time_slot["normalized_value"],
                source=time_slot["source"],
                confidence=time_slot["confidence"],
                evidence=time_slot["evidence"],
                required="time_range" in required_keys,
            ),
            self._make_slot_card(
                key="location",
                value=location_slot["value"],
                raw_value=location_slot["raw_value"],
                normalized_value=location_slot["normalized_value"],
                source=location_slot["source"],
                confidence=location_slot["confidence"],
                evidence=location_slot["evidence"],
                required="location" in required_keys,
            ),
            self._make_slot_card(
                key="merchant_name",
                value=merchant_slot["value"],
                raw_value=merchant_slot["raw_value"],
                normalized_value=merchant_slot["normalized_value"],
                source=merchant_slot["source"],
                confidence=merchant_slot["confidence"],
                evidence=merchant_slot["evidence"],
                required="merchant_name" in required_keys,
            ),
            self._make_slot_card(
                key="amount",
                value=amount_slot["value"],
                raw_value=amount_slot["raw_value"],
                normalized_value=amount_slot["normalized_value"],
                source=amount_slot["source"],
                confidence=amount_slot["confidence"],
                evidence=amount_slot["evidence"],
                required="amount" in required_keys,
            ),
            self._make_slot_card(
                key="reason",
                value=reason_slot["value"],
                raw_value=reason_slot["raw_value"],
                normalized_value=reason_slot["normalized_value"],
                source=reason_slot["source"],
                confidence=reason_slot["confidence"],
                evidence=reason_slot["evidence"],
                required="reason" in required_keys,
            ),
            self._make_slot_card(
                key="participants",
                value=participants_slot["value"],
                raw_value=participants_slot["raw_value"],
                normalized_value=participants_slot["normalized_value"],
                source=participants_slot["source"],
                confidence=participants_slot["confidence"],
                evidence=participants_slot["evidence"],
                required="participants" in required_keys,
            ),
            self._make_slot_card(
                key="attachments",
                value=attachment_slot["value"],
                raw_value=attachment_slot["raw_value"],
                normalized_value=attachment_slot["normalized_value"],
                source=attachment_slot["source"],
                confidence=attachment_slot["confidence"],
                evidence=attachment_slot["evidence"],
                required="attachments" in required_keys,
            ),
        ]
        return cards

    def _build_review_document_cards(
        self,
        payload: UserAgentRequest,
        *,
        ocr_documents: list[dict[str, object]],
    ) -> list[UserAgentReviewDocumentCard]:
        cards: list[UserAgentReviewDocumentCard] = []
        for index, item in enumerate(ocr_documents, start=1):
            classified = self._classify_document(item, payload)
            fields = self._extract_document_fields(item)
            cards.append(
                UserAgentReviewDocumentCard(
                    index=index,
                    filename=str(item.get("filename") or f"document-{index}"),
                    document_type=classified["document_type"],
                    suggested_expense_type=classified["expense_type"],
                    scene_label=GROUP_SCENE_LABELS.get(
                        classified["group_code"],
                        classified["scene_label"],
                    ),
                    summary=str(item.get("summary") or item.get("text") or "").strip(),
                    avg_score=float(item.get("avg_score") or 0.0),
                    preview_kind=str(item.get("preview_kind") or "").strip(),
                    preview_data_url=str(item.get("preview_data_url") or "").strip(),
                    warnings=[str(warning) for warning in item.get("warnings", []) if str(warning).strip()],
                    fields=[
                        UserAgentReviewDocumentField(
                            label=label,
                            value=value,
                            source="ocr",
                        )
                        for label, value in fields.items()
                        if str(value).strip()
                    ],
                )
            )
        return cards

    def _build_review_claim_groups(
        self,
        payload: UserAgentRequest,
        *,
        document_cards: list[UserAgentReviewDocumentCard],
    ) -> list[UserAgentReviewClaimGroup]:
        groups: dict[str, dict[str, object]] = {}
        for card in document_cards:
            group_code = self._normalize_group_code(card.suggested_expense_type)
            bucket = groups.setdefault(
                group_code,
                {
                    "document_indexes": [],
                    "amount_total": 0.0,
                    "expense_type": str(card.suggested_expense_type or group_code).strip() or group_code,
                    "scene_label": GROUP_SCENE_LABELS.get(
                        str(card.suggested_expense_type or group_code).strip() or group_code,
                        GROUP_SCENE_LABELS.get(group_code, "其他费用"),
                    ),
                    "reasons": [],
                },
            )
            bucket["document_indexes"].append(card.index)
            bucket["amount_total"] = float(bucket["amount_total"]) + self._extract_amount_from_card(card)
            bucket["reasons"].append(f"{card.filename} 识别为 {card.scene_label}")
            current_expense_type = str(bucket["expense_type"] or "").strip()
            current_card_type = str(card.suggested_expense_type or "").strip()
            if current_expense_type and current_card_type and current_expense_type != current_card_type:
                bucket["expense_type"] = group_code
                bucket["scene_label"] = GROUP_SCENE_LABELS.get(group_code, "其他费用")

        if not groups:
            expense_type_code = self._collect_entity_values(payload).get("expense_type_code", "other")
            group_code = self._normalize_group_code(expense_type_code)
            groups[group_code] = {
                "document_indexes": [],
                "amount_total": self._resolve_amount_value(payload),
                "expense_type": expense_type_code or "other",
                "scene_label": GROUP_SCENE_LABELS.get(group_code, "其他费用"),
                "reasons": ["当前主要依据用户文本和页面上下文进行分单建议。"],
            }

        claim_groups: list[UserAgentReviewClaimGroup] = []
        for index, (group_code, bucket) in enumerate(groups.items(), start=1):
            title = f"建议报销单 {index}：{bucket['scene_label']}"
            rationale = (
                "；".join(dict.fromkeys(str(item) for item in bucket["reasons"]))
                if bucket["reasons"]
                else "当前仅有单一场景，无需拆单。"
            )
            claim_groups.append(
                UserAgentReviewClaimGroup(
                    group_code=group_code,
                    title=title,
                    expense_type=str(bucket["expense_type"]),
                    scene_label=str(bucket["scene_label"]),
                    document_indexes=list(bucket["document_indexes"]),
                    amount_total=round(float(bucket["amount_total"]), 2),
                    rationale=rationale,
                )
            )
        return claim_groups

    def _build_review_risk_briefs(
        self,
        payload: UserAgentRequest,
        *,
        citations: list[UserAgentCitation],
        document_cards: list[UserAgentReviewDocumentCard],
        claim_groups: list[UserAgentReviewClaimGroup],
    ) -> list[UserAgentReviewRiskBrief]:
        briefs: list[UserAgentReviewRiskBrief] = []
        employee_name = self._collect_entity_values(payload).get("employee_name") or str(
            payload.context_json.get("name") or ""
        ).strip()
        if employee_name:
            since = datetime.now(UTC) - timedelta(days=90)
            stmt = select(ExpenseClaim).where(
                ExpenseClaim.employee_name == employee_name,
                ExpenseClaim.occurred_at >= since,
            )
            recent_claims = list(self.db.scalars(stmt).all())
            if recent_claims:
                risky_count = sum(1 for item in recent_claims if item.risk_flags_json)
                draft_count = sum(1 for item in recent_claims if item.status == "draft")
                briefs.append(
                    UserAgentReviewRiskBrief(
                        title="历史报销画像",
                        level="info",
                        content=(
                            f"{employee_name} 最近 90 天共有 {len(recent_claims)} 笔报销，"
                            f"其中 {risky_count} 笔带风险标记，{draft_count} 笔仍处于草稿态。"
                        ),
                    )
                )
                current_amount = self._resolve_amount_value(payload)
                if current_amount > 0:
                    duplicate_count = sum(
                        1
                        for item in recent_claims
                        if abs(float(item.amount) - current_amount) < 0.01
                    )
                    if duplicate_count:
                        briefs.append(
                            UserAgentReviewRiskBrief(
                                title="金额重复预警",
                                level="warning",
                                content=(
                                    f"近 90 天发现 {duplicate_count} 笔金额相同的报销记录，"
                                    "提交前建议核对是否为重复报销或拆分不当。"
                                ),
                            )
                        )

        if citations:
            briefs.append(
                UserAgentReviewRiskBrief(
                    title="制度注意事项",
                    level="info",
                    content=citations[0].excerpt or f"请先核对 {citations[0].title} 的制度要求。",
                )
            )

        warning_count = sum(len(item.warnings) for item in document_cards)
        if warning_count:
            briefs.append(
                UserAgentReviewRiskBrief(
                    title="票据识别提醒",
                    level="warning",
                    content=f"当前共有 {warning_count} 条票据识别提示，建议逐张确认 OCR 识别字段。",
                )
            )

        if len(claim_groups) > 1:
            briefs.append(
                UserAgentReviewRiskBrief(
                    title="建议拆单",
                    level="high",
                    content=f"系统检测到 {len(claim_groups)} 类费用场景，建议拆成多张报销单后再提交。",
                )
            )

        return briefs[:4]

    def _build_review_confirmation_actions(
        self,
        payload: UserAgentRequest,
        *,
        can_proceed: bool,
        claim_groups: list[UserAgentReviewClaimGroup],
        draft_payload: UserAgentDraftPayload | None,
    ) -> list[UserAgentReviewAction]:
        if self._is_review_association_choice_pending(payload):
            claim_no = str(payload.tool_payload.get("association_candidate_claim_no") or "").strip()
            link_label = f"关联到草稿 {claim_no}" if claim_no else "关联到现有草稿"
            return [
                UserAgentReviewAction(
                    label="取消",
                    action_type="cancel_review",
                    description="放弃当前识别结果，并退出本次核对流程。",
                    emphasis="secondary",
                ),
                UserAgentReviewAction(
                    label="修改识别信息",
                    action_type="edit_review",
                    description="打开结构化模板，按已识别字段逐项修改。",
                    emphasis="secondary",
                ),
                UserAgentReviewAction(
                    label=link_label,
                    action_type="link_to_existing_draft",
                    description=(
                        f"把本次上传票据并入现有草稿 {claim_no}。"
                        if claim_no
                        else "把本次上传票据并入现有草稿。"
                    ),
                    emphasis="primary",
                ),
                UserAgentReviewAction(
                    label="单独建立报销单",
                    action_type="create_new_claim_from_documents",
                    description="基于当前上传的多张票据，新建一张独立的报销草稿。",
                    emphasis="secondary",
                ),
            ]

        primary_action = UserAgentReviewAction(
            label="继续下一步" if can_proceed else "保存为草稿",
            action_type="next_step" if can_proceed else "save_draft",
            description=(
                "当前识别信息已满足继续处理条件，确认后进入下一步。"
                if can_proceed
                else "暂存当前识别结果，后续可以继续补充或修改。"
            ),
            emphasis="primary",
        )
        if len(claim_groups) > 1 and can_proceed:
            primary_action.description = f"系统建议拆分为 {len(claim_groups)} 张报销单，确认后继续下一步。"
        if draft_payload is not None and draft_payload.claim_no and not can_proceed:
            primary_action.description = f"保存后会生成草稿 {draft_payload.claim_no}，后续仍可继续补充。"

        return [
            UserAgentReviewAction(
                label="取消",
                action_type="cancel_review",
                description="放弃当前识别结果，并退出本次核对流程。",
                emphasis="secondary",
            ),
            UserAgentReviewAction(
                label="修改识别信息",
                action_type="edit_review",
                description="打开结构化模板，按已识别字段逐项修改。",
                emphasis="secondary",
            ),
            primary_action,
        ]

    def _build_review_intent_summary(
        self,
        payload: UserAgentRequest,
        *,
        slot_cards: list[UserAgentReviewSlotCard],
        claim_groups: list[UserAgentReviewClaimGroup],
    ) -> str:
        slots = {item.key: item for item in slot_cards}
        expense_type = slots.get("expense_type")
        amount = slots.get("amount")
        time_range = slots.get("time_range")
        location = slots.get("location")
        customer = slots.get("customer_name")

        summary = "我先根据您当前提供的信息整理出一笔报销。"
        if expense_type and expense_type.value:
            summary = f"识别到您希望报销一笔“{expense_type.value}”费用。"
        details: list[str] = []
        if customer and customer.value:
            details.append(f"客户为 {customer.value}")
        if time_range and time_range.value:
            details.append(f"时间为 {time_range.value}")
        if location and location.value:
            details.append(f"地点为 {location.value}")
        if amount and amount.value:
            details.append(f"金额为 {amount.value}")
        reason = slots.get("reason")
        if reason and reason.value:
            details.append(f"事由是 {reason.value}")
        if details:
            return f"{summary} {'，'.join(details)}。"
        return summary

    def _build_review_body_answer(
        self,
        payload: UserAgentRequest,
        *,
        review_payload: UserAgentReviewPayload | None,
        draft_payload: UserAgentDraftPayload | None,
    ) -> str | None:
        if review_payload is None:
            return None
        if payload.ontology.scenario != "expense":
            return None
        if payload.ontology.intent not in {"draft", "operate"}:
            return None
        if payload.tool_payload.get("draft_limit_reached"):
            return (
                str(payload.tool_payload.get("message") or "").strip()
                or "你当前已保存 3 个草稿，请先完成已保存的草稿，才能再次新建草稿。"
            )

        review_action = str(payload.context_json.get("review_action") or "").strip()
        if review_action == "save_draft":
            if draft_payload is not None and draft_payload.claim_no:
                return (
                    f"已按您当前确认的信息保存为草稿 {draft_payload.claim_no}。"
                    "后续您可以继续补充缺失项，或修改识别结果后再继续提交。"
                )
            return "已按您当前确认的信息保存为草稿。后续您可以继续补充缺失项，或修改识别结果后再继续提交。"
        if review_action == "link_to_existing_draft":
            document_count = self._resolve_review_document_count(payload)
            if draft_payload is not None and draft_payload.claim_no:
                return (
                    f"已将本次上传的 {document_count} 张票据关联到草稿 {draft_payload.claim_no}。"
                    "您可以继续补充识别字段，确认无误后再提交审批。"
                )
            return "已将本次上传的票据关联到现有草稿。您可以继续补充识别字段，确认无误后再提交审批。"
        if review_action == "create_new_claim_from_documents":
            document_count = self._resolve_review_document_count(payload)
            if draft_payload is not None and draft_payload.claim_no:
                return (
                    f"已按当前上传的 {document_count} 张票据新建报销草稿 {draft_payload.claim_no}。"
                    "您可以继续补充识别字段，确认无误后再提交审批。"
                )
            return "已按当前上传票据新建报销草稿。您可以继续补充识别字段，确认无误后再提交审批。"
        if review_action == "next_step":
            if draft_payload is not None and draft_payload.status == "submitted":
                stage_text = draft_payload.approval_stage or "审批中"
                return f"报销单 {draft_payload.claim_no or ''} 已提交，当前节点为 {stage_text}。".strip()
            if payload.tool_payload.get("submission_blocked"):
                return str(payload.tool_payload.get("message") or "").strip() or "当前报销单暂时还不能提交审批。"
            return (
                f"{self._build_review_intent_summary(payload, slot_cards=review_payload.slot_cards, claim_groups=review_payload.claim_groups)} "
                "当前关键信息已基本齐全，您确认无误后可以继续下一步。"
            )
        if review_action == "edit_review":
            return (
                f"{self._build_review_intent_summary(payload, slot_cards=review_payload.slot_cards, claim_groups=review_payload.claim_groups)} "
                f"{self._build_review_guidance_copy(review_payload, mention_save_draft=True)}"
            )
        return review_payload.body_message or None

    def _build_review_body_message(
        self,
        payload: UserAgentRequest,
        *,
        slot_cards: list[UserAgentReviewSlotCard],
        risk_briefs: list[UserAgentReviewRiskBrief],
        can_proceed: bool,
        document_cards: list[UserAgentReviewDocumentCard],
    ) -> str:
        if self._is_review_association_choice_pending(payload):
            claim_no = str(payload.tool_payload.get("association_candidate_claim_no") or "").strip()
            document_count = len(document_cards) or self._resolve_review_document_count(payload)
            if claim_no:
                return (
                    f"已识别出本次上传的 {document_count} 张票据。"
                    f"系统检测到你已有草稿 {claim_no}，请选择关联到该草稿，或单独建立一张新的报销单。"
                )
            return (
                f"已识别出本次上传的 {document_count} 张票据。"
                "系统检测到你已有可用草稿，请先选择关联到现有草稿，或单独建立一张新的报销单。"
            )

        review_payload = UserAgentReviewPayload(
            intent_summary="",
            body_message="",
            scenario=payload.ontology.scenario,
            intent=payload.ontology.intent,
            can_proceed=can_proceed,
            missing_slots=self._resolve_review_missing_slot_labels(slot_cards),
            risk_briefs=risk_briefs,
            slot_cards=slot_cards,
            document_cards=[],
            claim_groups=[],
            confirmation_actions=[],
            edit_fields=[],
        )
        return (
            f"{self._build_review_intent_summary(payload, slot_cards=slot_cards, claim_groups=[])} "
            f"{self._build_review_guidance_copy(review_payload, mention_save_draft=not can_proceed)}"
        )

    @staticmethod
    def _resolve_review_missing_slot_labels(
        slot_cards: list[UserAgentReviewSlotCard],
    ) -> list[str]:
        return [item.label for item in slot_cards if item.status == "missing"]

    @staticmethod
    def _build_review_guidance_copy(
        review_payload: UserAgentReviewPayload,
        *,
        mention_save_draft: bool,
    ) -> str:
        missing_count = len(review_payload.missing_slots)
        reminder_count = len(review_payload.risk_briefs)

        if review_payload.can_proceed:
            if reminder_count:
                return (
                    f"当前关键信息已基本齐全，但还有 {reminder_count} 条提醒。"
                    "您可以展开下方卡片查看详情，确认无误后继续下一步。"
                )
            return "当前关键信息已基本齐全，您确认无误后可以继续下一步。"

        issue_parts: list[str] = []
        if missing_count:
            issue_parts.append(f"{missing_count} 项信息待补充")
        if reminder_count:
            issue_parts.append(f"{reminder_count} 条提醒")
        issue_summary = "、".join(issue_parts) if issue_parts else "一些细节还需要进一步确认"

        suffix = "；如果想先暂存，也可以点击下方按钮保存草稿。" if mention_save_draft else "。"
        return (
            f"当前还有 {issue_summary}。"
            f"您可以展开下方卡片查看详情，继续补充或修改{suffix}"
        )

    @staticmethod
    def _can_proceed_review(
        payload: UserAgentRequest,
        *,
        missing_slot_keys: list[str],
        claim_groups: list[UserAgentReviewClaimGroup],
    ) -> bool:
        if payload.ontology.ambiguity:
            return False
        if missing_slot_keys:
            return False
        if not claim_groups:
            return False
        return True

    def _build_review_edit_fields(
        self,
        payload: UserAgentRequest,
        *,
        draft_payload: UserAgentDraftPayload | None,
        slot_cards: list[UserAgentReviewSlotCard],
    ) -> list[UserAgentReviewEditField]:
        slot_map = {item.key: item for item in slot_cards}
        employee = self._resolve_employee_profile(payload)
        reporter_name = (
            slot_map.get("reporter_name").value
            if slot_map.get("reporter_name")
            else str(payload.context_json.get("name") or "").strip()
        )
        manager_name = self._resolve_manager_name(employee)
        reason = slot_map.get("reason").value if slot_map.get("reason") else ""
        attachments = "、".join(self._resolve_attachment_names(payload))

        fields = [
            UserAgentReviewEditField(
                key="claim_no",
                label="报销单据编号",
                value=str(draft_payload.claim_no if draft_payload is not None and draft_payload.claim_no else "待生成"),
                placeholder="保存草稿后自动生成",
                required=False,
                group="basic",
            ),
            UserAgentReviewEditField(
                key="expense_type",
                label="报销类型",
                value=slot_map.get("expense_type").value if slot_map.get("expense_type") else "",
                placeholder="例如：业务招待费 / 差旅费",
                group="basic",
            ),
            UserAgentReviewEditField(
                key="occurred_date",
                label="业务发生时间",
                value=slot_map.get("time_range").normalized_value if slot_map.get("time_range") and slot_map.get("time_range").normalized_value else slot_map.get("time_range").value if slot_map.get("time_range") else "",
                placeholder="例如：2026-05-11",
                group="basic",
            ),
            UserAgentReviewEditField(
                key="reporter_name",
                label="报销人",
                value=reporter_name,
                placeholder="请输入报销人姓名",
                group="basic",
            ),
            UserAgentReviewEditField(
                key="manager_name",
                label="直属上司姓名",
                value=manager_name,
                placeholder="请输入直属上司姓名",
                required=False,
                group="basic",
            ),
            UserAgentReviewEditField(
                key="customer_name",
                label="客户名称",
                value=slot_map.get("customer_name").value if slot_map.get("customer_name") else "",
                placeholder="请输入客户名称",
                group="business",
            ),
            UserAgentReviewEditField(
                key="business_location",
                label="业务地点",
                value=slot_map.get("location").normalized_value if slot_map.get("location") and slot_map.get("location").normalized_value else slot_map.get("location").value if slot_map.get("location") else "",
                placeholder="例如：北京 / 客户现场",
                required=False,
                group="business",
            ),
            UserAgentReviewEditField(
                key="merchant_name",
                label="酒店/商户",
                value=slot_map.get("merchant_name").value if slot_map.get("merchant_name") else "",
                placeholder="请输入酒店或商户名称",
                required=False,
                group="business",
            ),
            UserAgentReviewEditField(
                key="amount",
                label="金额",
                value=slot_map.get("amount").normalized_value if slot_map.get("amount") and slot_map.get("amount").normalized_value else slot_map.get("amount").value if slot_map.get("amount") else "",
                placeholder="例如：200.00元",
                group="business",
            ),
            UserAgentReviewEditField(
                key="participants",
                label="参与人员",
                value=slot_map.get("participants").value if slot_map.get("participants") else "",
                placeholder="例如：客户 2 人，我方 1 人",
                group="business",
            ),
            UserAgentReviewEditField(
                key="reason",
                label="事由",
                value=reason,
                placeholder="请输入报销事由",
                field_type="textarea",
                group="business",
            ),
            UserAgentReviewEditField(
                key="attachment_names",
                label="附件清单",
                value=attachments,
                placeholder="例如：发票.jpg、行程单.png",
                required=False,
                field_type="textarea",
                group="attachments",
            ),
        ]
        return fields

    def _resolve_employee_profile(self, payload: UserAgentRequest) -> Employee | None:
        candidates = [
            str(payload.context_json.get("name") or "").strip(),
            str(payload.user_id or "").strip(),
            self._collect_entity_values(payload).get("employee_name", ""),
        ]
        normalized = [item for item in dict.fromkeys(candidates) if item]
        if not normalized:
            return None

        stmt = (
            select(Employee)
            .where(
                or_(
                    Employee.name.in_(normalized),
                    Employee.employee_no.in_(normalized),
                    Employee.email.in_(normalized),
                )
            )
            .limit(1)
        )
        return self.db.scalar(stmt)

    @staticmethod
    def _resolve_manager_name(employee: Employee | None) -> str:
        if employee is None:
            return ""
        if employee.manager is not None and employee.manager.name:
            return employee.manager.name
        if employee.organization_unit is not None and employee.organization_unit.manager_name:
            return employee.organization_unit.manager_name
        return ""

    @staticmethod
    def _extract_message_reason(message: str) -> str:
        for line in str(message or "").splitlines():
            cleaned = line.strip()
            if not cleaned:
                continue
            if cleaned.startswith(("附件名称：", "OCR摘要：", "关联单号：")):
                continue
            return cleaned[:300]
        return ""

    @staticmethod
    def _looks_like_system_generated_reason_message(message: str) -> bool:
        cleaned = str(message or "").strip()
        if not cleaned:
            return False
        compact = re.sub(r"\s+", "", cleaned)
        return compact.startswith(SYSTEM_GENERATED_REASON_PREFIXES)

    def _resolve_reason_source_text(self, payload: UserAgentRequest) -> str:
        explicit_text = payload.context_json.get("user_input_text")
        if isinstance(explicit_text, str):
            return explicit_text.strip()
        if self._looks_like_system_generated_reason_message(payload.message):
            return ""
        return str(payload.message or "").strip()

    @classmethod
    def _resolve_reason_text(cls, message: str) -> str:
        reason = cls._extract_message_reason(message)
        if not reason:
            return ""

        compact = re.sub(r"\s+", "", reason)
        if compact in GENERIC_EXPENSE_PROMPTS:
            return ""

        instruction_prefixes = (
            "帮我生成",
            "请帮我生成",
            "生成",
            "起草",
            "创建",
            "发起",
            "准备",
            "帮我报销",
            "我要报销",
            "我想报销",
        )
        if compact.startswith(instruction_prefixes):
            for separator in ("，", ",", "。", "；", ";", "：", ":"):
                if separator in reason:
                    trailing = reason.split(separator, 1)[1].strip()
                    if trailing:
                        return trailing[:300]
            return ""

        return reason

    @staticmethod
    def _should_skip_model_answer(
        payload: UserAgentRequest,
        review_payload: UserAgentReviewPayload | None,
    ) -> bool:
        if payload.ontology.scenario == "expense" and payload.ontology.intent in {"query", "compare"}:
            return True
        if review_payload is None:
            return False
        return payload.ontology.scenario == "expense" and (
            payload.ontology.intent == "draft"
            or int(payload.context_json.get("attachment_count") or 0) > 0
        )

    def _build_citations(self, payload: UserAgentRequest) -> list[UserAgentCitation]:
        knowledge_citations = self._build_knowledge_citations(payload)
        if payload.ontology.scenario == "knowledge":
            return knowledge_citations[:3]

        rule_citations = self._build_rule_asset_citations(payload)
        if knowledge_citations:
            return (knowledge_citations + rule_citations)[:3]
        return rule_citations

    @staticmethod
    def _build_knowledge_citations(payload: UserAgentRequest) -> list[UserAgentCitation]:
        citations: list[UserAgentCitation] = []
        for item in list(payload.tool_payload.get("hits") or [])[:3]:
            if not isinstance(item, dict):
                continue
            title = str(item.get("title") or item.get("document_name") or "").strip()
            code = str(item.get("code") or item.get("candidate_id") or "").strip()
            if not title or not code:
                continue
            citations.append(
                UserAgentCitation(
                    source_type="knowledge",
                    code=code,
                    title=title,
                    version=str(item.get("version") or "").strip() or None,
                    updated_at=str(item.get("updated_at") or "").strip() or None,
                    excerpt=(
                        str(item.get("excerpt") or "").strip()
                        or str(item.get("content") or "").strip()
                        or None
                    ),
                )
            )
        return citations

    def _build_rule_asset_citations(self, payload: UserAgentRequest) -> list[UserAgentCitation]:
        domain = self._resolve_domain(payload.ontology.scenario)
        items = self.asset_service.list_assets(
            asset_type=AgentAssetType.RULE.value,
            status=AgentAssetStatus.ACTIVE.value,
            domain=domain,
        )
        ranked = self._rank_rule_assets(items, payload)
        citations: list[UserAgentCitation] = []
        for item in ranked[:2]:
            detail = self.asset_service.get_asset(item.id)
            if detail is None:
                continue
            excerpt = self._extract_excerpt(str(detail.current_version_content or ""))
            citations.append(
                UserAgentCitation(
                    source_type="rule",
                    code=detail.code,
                    title=detail.name,
                    version=detail.current_version,
                    updated_at=detail.updated_at.date().isoformat(),
                    excerpt=excerpt,
                )
            )
        return citations

    @staticmethod
    def _resolve_risk_flags(payload: UserAgentRequest) -> list[str]:
        tool_flags = payload.tool_payload.get("risk_flags")
        if isinstance(tool_flags, list) and tool_flags:
            return [str(item) for item in tool_flags]
        return [str(item) for item in payload.ontology.risk_flags]

    @staticmethod
    def _resolve_subject(payload: UserAgentRequest) -> str:
        named_entities = [
            item.value
            for item in payload.ontology.entities
            if item.type in {"employee", "customer", "vendor", "project"}
        ]
        if named_entities:
            return f"{'、'.join(named_entities)} 相关数据"
        return f"{SCENARIO_LABELS.get(payload.ontology.scenario, '当前')}场景数据"

    @staticmethod
    def _is_generic_expense_prompt(payload: UserAgentRequest) -> bool:
        if payload.ontology.scenario != "expense":
            return False
        normalized_message = re.sub(r"\s+", "", payload.message)
        return normalized_message in GENERIC_EXPENSE_PROMPTS

    @staticmethod
    def _is_implicit_expense_draft_request(payload: UserAgentRequest) -> bool:
        if payload.ontology.scenario != "expense" or payload.ontology.intent != "draft":
            return False

        compact_message = re.sub(r"\s+", "", payload.message)
        if any(keyword in compact_message for keyword in EXPLICIT_DRAFT_KEYWORDS):
            return False

        return True

    @staticmethod
    def _resolve_attachment_names(payload: UserAgentRequest) -> list[str]:
        names = payload.context_json.get("attachment_names")
        if not isinstance(names, list):
            return []
        return [str(name) for name in names if str(name).strip()]

    @staticmethod
    def _resolve_attachment_count(payload: UserAgentRequest) -> int:
        names = UserAgentService._resolve_attachment_names(payload)
        if names:
            return len(names)
        try:
            return max(0, int(payload.context_json.get("attachment_count") or 0))
        except (TypeError, ValueError):
            return 0

    @staticmethod
    def _resolve_ocr_documents(payload: UserAgentRequest) -> list[dict[str, object]]:
        documents = payload.context_json.get("ocr_documents")
        if not isinstance(documents, list):
            return []
        overrides = payload.context_json.get("review_document_form_values")
        override_map: dict[tuple[int, str], dict[str, object]] = {}
        if isinstance(overrides, list):
            for item in overrides:
                if not isinstance(item, dict):
                    continue
                filename = str(item.get("filename") or "").strip()
                index = int(item.get("index") or 0)
                if not filename and index <= 0:
                    continue
                override_map[(index, filename)] = item
        normalized: list[dict[str, object]] = []
        for index, item in enumerate(documents[:8], start=1):
            if not isinstance(item, dict):
                continue
            normalized_item = dict(item)
            override = override_map.get((index, str(normalized_item.get("filename") or "").strip()))
            if override is None:
                override = override_map.get((index, ""))
            if override is not None:
                summary = str(override.get("summary") or "").strip()
                scene_label = str(override.get("scene_label") or "").strip()
                fields = override.get("fields")
                if summary:
                    normalized_item["summary"] = summary
                if scene_label:
                    normalized_item["scene_label"] = scene_label
                if isinstance(fields, list):
                    normalized_item["document_fields"] = [
                        {
                            "key": str(field.get("key") or field.get("label") or "").strip(),
                            "label": str(field.get("label") or "").strip(),
                            "value": str(field.get("value") or "").strip(),
                        }
                        for field in fields
                        if isinstance(field, dict)
                        and str(field.get("label") or "").strip()
                        and str(field.get("value") or "").strip()
                    ]
            normalized.append(normalized_item)
        return normalized

    @staticmethod
    def _is_review_association_choice_pending(payload: UserAgentRequest) -> bool:
        return bool(payload.tool_payload.get("pending_association_decision"))

    def _resolve_review_document_count(self, payload: UserAgentRequest) -> int:
        return max(
            len(self._resolve_ocr_documents(payload)),
            self._resolve_attachment_count(payload),
        )

    @staticmethod
    def _resolve_conversation_history(payload: UserAgentRequest) -> list[dict[str, object]]:
        history = payload.context_json.get("conversation_history")
        if not isinstance(history, list):
            return []

        normalized: list[dict[str, object]] = []
        for item in history[-8:]:
            if not isinstance(item, dict):
                continue
            role = str(item.get("role") or "").strip()
            content = str(item.get("content") or "").strip()
            if not role or not content:
                continue
            normalized.append({"role": role, "content": content})
        return normalized

    @staticmethod
    def _resolve_domain(scenario: str) -> str | None:
        if scenario == "expense":
            return "expense"
        if scenario == "accounts_receivable":
            return "ar"
        if scenario == "accounts_payable":
            return "ap"
        return None

    @staticmethod
    def _rank_rule_assets(
        items: list[AgentAssetListItem],
        payload: UserAgentRequest,
    ) -> list[AgentAssetListItem]:
        def score(item: AgentAssetListItem) -> tuple[int, str]:
            tags = {str(value) for value in item.scenario_json or []}
            weight = 0
            if payload.ontology.scenario in tags:
                weight += 3
            if payload.ontology.intent in tags:
                weight += 2
            for risk_flag in payload.ontology.risk_flags:
                if risk_flag in tags:
                    weight += 4
            return weight, item.code

        ranked = sorted(items, key=score, reverse=True)
        return [item for item in ranked if score(item)[0] > 0]

    @staticmethod
    def _extract_excerpt(content: str) -> str:
        lines = [line.strip() for line in str(content).splitlines() if line.strip()]
        cleaned: list[str] = []
        for line in lines:
            normalized = re.sub(r"^[#>\-\*\d\.\s`]+", "", line).strip()
            if normalized:
                cleaned.append(normalized)
            if len(cleaned) >= 2:
                break
        return "；".join(cleaned[:2])

    def _collect_entity_values(self, payload: UserAgentRequest) -> dict[str, str]:
        values = {
            "employee_name": "",
            "customer": "",
            "participants": "",
            "amount": "",
            "expense_type": "",
            "expense_type_code": "",
        }
        participants: list[str] = []
        for item in payload.ontology.entities:
            if item.type == "employee" and not values["employee_name"]:
                values["employee_name"] = item.value
            elif item.type == "customer" and not values["customer"]:
                values["customer"] = item.value
            elif item.type == "amount" and item.role != "threshold" and not values["amount"]:
                normalized_amount = str(item.normalized_value or "").strip()
                values["amount"] = f"{normalized_amount}元" if normalized_amount else item.value
            elif item.type == "expense_type" and not values["expense_type_code"]:
                values["expense_type_code"] = item.normalized_value
                values["expense_type"] = EXPENSE_TYPE_LABELS.get(
                    item.normalized_value,
                    item.value,
                )
            elif item.type in {"participant", "person"} and item.value.strip():
                participants.append(item.value.strip())
        if participants:
            values["participants"] = "、".join(dict.fromkeys(participants))
        return values

    def _format_time_range(self, payload: UserAgentRequest) -> str:
        time_range = payload.ontology.time_range
        if time_range.start_date and time_range.end_date:
            if time_range.start_date == time_range.end_date:
                return time_range.start_date
            normalized = f"{time_range.start_date} 至 {time_range.end_date}"
            return normalized
        if time_range.raw:
            return time_range.raw
        return ""

    def _resolve_location_value(self, payload: UserAgentRequest) -> str:
        review_form_values = self._resolve_review_form_values(payload)
        for key in ("business_location", "location"):
            value = str(review_form_values.get(key) or "").strip()
            if value:
                return value

        if str(payload.context_json.get("entry_source") or "").strip() == "detail":
            request_context = payload.context_json.get("request_context")
            if isinstance(request_context, dict):
                for key in ("city", "location"):
                    value = str(request_context.get(key) or "").strip()
                    if value:
                        return value

        labeled_match = re.search(r"(?:业务地点|发生地点|地点)[:：]\s*(?P<value>[^\n，。；]+)", payload.message)
        if labeled_match:
            return labeled_match.group("value").strip()

        city_match = re.search(r"去(?P<city>[\u4e00-\u9fa5]{2,8})(?:出差|拜访|参会|见客户|客户现场)", payload.message)
        if city_match:
            return city_match.group("city").strip()
        if "客户现场" in payload.message.replace(" ", ""):
            return "客户现场"
        return ""

    @staticmethod
    def _resolve_review_form_values(payload: UserAgentRequest) -> dict[str, str]:
        values = payload.context_json.get("review_form_values")
        if not isinstance(values, dict):
            return {}
        normalized: dict[str, str] = {}
        for key, value in values.items():
            cleaned_key = str(key or "").strip()
            if not cleaned_key:
                continue
            normalized[cleaned_key] = str(value or "").strip()
        return normalized

    @staticmethod
    def _build_slot_value(
        *,
        value: str = "",
        raw_value: str = "",
        normalized_value: str = "",
        source: str = "system",
        confidence: float = 0.0,
        evidence: str = "",
    ) -> dict[str, str | float]:
        return {
            "value": str(value or "").strip(),
            "raw_value": str(raw_value or "").strip(),
            "normalized_value": str(normalized_value or "").strip(),
            "source": str(source or "system").strip() or "system",
            "confidence": float(confidence),
            "evidence": str(evidence or "").strip(),
        }

    def _build_time_slot(self, payload: UserAgentRequest) -> dict[str, str | float]:
        review_form_values = self._resolve_review_form_values(payload)
        edited_value = str(
            review_form_values.get("occurred_date")
            or review_form_values.get("time_range")
            or review_form_values.get("business_time")
            or ""
        ).strip()
        if edited_value:
            raw_value = str(review_form_values.get("time_range_raw") or edited_value).strip()
            return self._build_slot_value(
                value=edited_value,
                raw_value=raw_value,
                normalized_value=edited_value,
                source="user_form",
                confidence=1.0,
                evidence="来源于用户修改后的结构化表单。",
            )

        time_range = payload.ontology.time_range
        if time_range.start_date and time_range.end_date:
            normalized_value = (
                time_range.start_date
                if time_range.start_date == time_range.end_date
                else f"{time_range.start_date} 至 {time_range.end_date}"
            )
            raw_value = str(time_range.raw or "").strip()
            return self._build_slot_value(
                value=normalized_value,
                raw_value=raw_value,
                normalized_value=normalized_value,
                source="user_text",
                confidence=0.92,
                evidence="系统已根据当前日期将相对时间换算为标准日期。",
            )

        return self._build_slot_value()

    def _build_location_slot(self, payload: UserAgentRequest) -> dict[str, str | float]:
        review_form_values = self._resolve_review_form_values(payload)
        for key in ("business_location", "location"):
            value = str(review_form_values.get(key) or "").strip()
            if value:
                return self._build_slot_value(
                    value=value,
                    normalized_value=value,
                    source="user_form",
                    confidence=1.0,
                    evidence="来源于用户修改后的结构化表单。",
                )

        if str(payload.context_json.get("entry_source") or "").strip() == "detail":
            request_context = payload.context_json.get("request_context")
            if isinstance(request_context, dict):
                for key in ("city", "location"):
                    value = str(request_context.get(key) or "").strip()
                    if value:
                        return self._build_slot_value(
                            value=value,
                            normalized_value=value,
                            source="detail_context",
                            confidence=0.68,
                            evidence="来源于当前关联单据，仅作为辅助上下文，需要用户再次核对。",
                        )

        value = self._resolve_location_value(payload)
        if value:
            evidence = "用户在文本中明确描述了业务地点。"
            if value == "客户现场":
                evidence = "用户明确提到“客户现场”，但未提供具体城市或地址。"
            return self._build_slot_value(
                value=value,
                normalized_value=value,
                source="user_text",
                confidence=0.82,
                evidence=evidence,
            )
        return self._build_slot_value()

    def _build_customer_slot(
        self,
        payload: UserAgentRequest,
        *,
        entity_map: dict[str, str],
    ) -> dict[str, str | float]:
        review_form_values = self._resolve_review_form_values(payload)
        value = str(review_form_values.get("customer_name") or "").strip()
        if value:
            return self._build_slot_value(
                value=value,
                normalized_value=value,
                source="user_form",
                confidence=1.0,
                evidence="来源于用户修改后的结构化表单。",
            )

        value = entity_map.get("customer", "")
        if value:
            return self._build_slot_value(
                value=value,
                normalized_value=value,
                source="user_text",
                confidence=0.88,
                evidence="用户在原始描述中直接提到了客户对象。",
            )
        return self._build_slot_value()

    def _build_participants_slot(
        self,
        payload: UserAgentRequest,
        *,
        entity_map: dict[str, str],
    ) -> dict[str, str | float]:
        review_form_values = self._resolve_review_form_values(payload)
        value = str(review_form_values.get("participants") or "").strip()
        if value:
            return self._build_slot_value(
                value=value,
                normalized_value=value,
                source="user_form",
                confidence=1.0,
                evidence="来源于用户修改后的结构化表单。",
            )

        value = entity_map.get("participants", "")
        if value:
            return self._build_slot_value(
                value=value,
                normalized_value=value,
                source="user_text",
                confidence=0.8,
                evidence="用户在当前描述中补充了参与人员。",
            )
        return self._build_slot_value()

    def _build_reason_slot(
        self,
        payload: UserAgentRequest,
        *,
        claim_groups: list[UserAgentReviewClaimGroup],
    ) -> dict[str, str | float]:
        review_form_values = self._resolve_review_form_values(payload)
        edited_value = str(review_form_values.get("reason") or "").strip()
        if edited_value:
            return self._build_slot_value(
                value=edited_value,
                raw_value=edited_value,
                normalized_value=edited_value,
                source="user_form",
                confidence=1.0,
                evidence="来源于用户修改后的结构化表单。",
            )

        inferred_reason = self._infer_reason_from_claim_groups(
            claim_groups=claim_groups,
        )
        reason_value = self._resolve_reason_text(self._resolve_reason_source_text(payload))
        if inferred_reason:
            return self._build_slot_value(
                value=inferred_reason,
                raw_value=reason_value or inferred_reason,
                normalized_value=inferred_reason,
                source="ocr",
                confidence=0.82,
                evidence=(
                    "系统已根据票据识别结果预置场景类型；原始描述仍保留为补充说明。"
                    if reason_value
                    else "系统已根据票据识别场景补全通用事由，若需更具体说明可继续修改。"
                ),
            )

        if reason_value:
            return self._build_slot_value(
                value=reason_value,
                raw_value=reason_value,
                normalized_value=reason_value,
                source="user_text",
                confidence=0.76,
                evidence="系统从用户原始描述中提取了本次费用事由，建议继续核对。",
            )
        return self._build_slot_value()

    def _build_amount_slot(
        self,
        payload: UserAgentRequest,
        *,
        entity_map: dict[str, str],
        ocr_documents: list[dict[str, object]],
    ) -> dict[str, str | float]:
        review_form_values = self._resolve_review_form_values(payload)
        edited_amount = str(review_form_values.get("amount") or "").strip()
        if edited_amount:
            normalized = self._normalize_amount_text(edited_amount)
            return self._build_slot_value(
                value=normalized,
                raw_value=edited_amount,
                normalized_value=normalized,
                source="user_form",
                confidence=1.0,
                evidence="来源于用户修改后的结构化表单。",
            )

        amount_value = entity_map.get("amount", "")
        if amount_value:
            normalized = self._normalize_amount_text(amount_value)
            return self._build_slot_value(
                value=normalized,
                raw_value=amount_value,
                normalized_value=normalized,
                source="user_text",
                confidence=0.92,
                evidence="用户在原始描述中直接给出了金额。",
            )

        ocr_total_amount = self._sum_ocr_amounts(ocr_documents)
        if ocr_total_amount > 0:
            normalized = f"{ocr_total_amount:.2f}元"
            return self._build_slot_value(
                value=normalized,
                normalized_value=normalized,
                source="ocr",
                confidence=0.76,
                evidence="金额来自 OCR 汇总结果，仍建议用户核对票据原文。",
            )
        return self._build_slot_value()

    def _build_expense_type_slot(
        self,
        payload: UserAgentRequest,
        *,
        entity_map: dict[str, str],
        ocr_documents: list[dict[str, object]],
    ) -> dict[str, str | float]:
        review_form_values = self._resolve_review_form_values(payload)
        edited_value = str(review_form_values.get("expense_type") or review_form_values.get("reimbursement_type") or "").strip()
        if edited_value:
            normalized_code, normalized_label = self._normalize_expense_type_input(edited_value)
            return self._build_slot_value(
                value=normalized_label,
                raw_value=edited_value,
                normalized_value=normalized_code,
                source="user_form",
                confidence=1.0,
                evidence="来源于用户修改后的结构化表单。",
            )

        expense_type_code = entity_map.get("expense_type_code", "")
        expense_type_value = EXPENSE_TYPE_LABELS.get(expense_type_code, entity_map.get("expense_type", ""))
        if expense_type_value:
            return self._build_slot_value(
                value=expense_type_value,
                raw_value=expense_type_value,
                normalized_value=expense_type_code,
                source="user_text",
                confidence=0.9,
                evidence="系统根据用户描述中的业务场景判断费用类型。",
            )

        inferred_label = self._infer_expense_type_from_documents(payload, ocr_documents) if ocr_documents else ""
        if inferred_label:
            normalized_code, normalized_label = self._normalize_expense_type_input(inferred_label)
            return self._build_slot_value(
                value=normalized_label,
                raw_value=inferred_label,
                normalized_value=normalized_code,
                source="ocr",
                confidence=0.74,
                evidence="系统根据票据内容推断费用类型，仍建议用户确认。",
            )
        return self._build_slot_value()

    def _build_merchant_slot(
        self,
        payload: UserAgentRequest,
        *,
        ocr_documents: list[dict[str, object]],
    ) -> dict[str, str | float]:
        review_form_values = self._resolve_review_form_values(payload)
        edited_value = str(review_form_values.get("merchant_name") or "").strip()
        if edited_value:
            return self._build_slot_value(
                value=edited_value,
                normalized_value=edited_value,
                source="user_form",
                confidence=1.0,
                evidence="来源于用户修改后的结构化表单。",
            )

        merchant_value = self._extract_document_merchant_name(ocr_documents[0]) if ocr_documents else ""
        if merchant_value:
            return self._build_slot_value(
                value=merchant_value,
                normalized_value=merchant_value,
                source="ocr",
                confidence=0.72,
                evidence="商户名称来自 OCR 票据识别结果，仍建议用户核对。",
            )
        return self._build_slot_value()

    def _build_attachment_slot(self, payload: UserAgentRequest) -> dict[str, str | float]:
        review_form_values = self._resolve_review_form_values(payload)
        attachment_names = str(review_form_values.get("attachment_names") or "").strip()
        if attachment_names:
            return self._build_slot_value(
                value=attachment_names,
                normalized_value=attachment_names,
                source="user_form",
                confidence=1.0,
                evidence="来源于用户修改后的结构化表单。",
            )

        count = self._resolve_attachment_count(payload)
        if count > 0:
            names = self._resolve_attachment_names(payload)
            value = "、".join(names) if names else f"{count} 份附件"
            return self._build_slot_value(
                value=value,
                raw_value=value,
                normalized_value=str(count),
                source="upload",
                confidence=1.0,
                evidence="系统已接收到用户上传的附件。",
            )
        return self._build_slot_value()

    @staticmethod
    def _normalize_amount_text(value: str) -> str:
        cleaned = str(value or "").strip()
        if not cleaned:
            return ""
        for alias, canonical in sorted(AMOUNT_UNIT_ALIASES.items(), key=lambda item: len(item[0]), reverse=True):
            cleaned = cleaned.replace(alias, canonical)
        match = AMOUNT_TEXT_PATTERN.search(cleaned)
        if not match:
            return cleaned
        number = float(match.group(1))
        return f"{number:.2f}元"

    @staticmethod
    def _normalize_expense_type_input(value: str) -> tuple[str, str]:
        compact = str(value or "").replace(" ", "")
        if "招待" in compact or ("客户" in compact and any(keyword in compact for keyword in ("吃饭", "用餐", "宴请", "请客"))):
            return "entertainment", "业务招待费"
        if any(keyword in compact for keyword in ("差旅", "出差", "机票", "行程")):
            return "travel", "差旅费"
        if any(keyword in compact for keyword in ("住宿", "酒店", "宾馆")):
            return "hotel", "住宿费"
        if any(keyword in compact for keyword in ("交通", "打车", "网约车", "出租车", "车费", "停车")):
            return "transport", "交通费"
        if any(keyword in compact for keyword in ("餐费", "用餐", "午餐", "晚餐", "早餐", "伙食")):
            return "meal", "餐费"
        if "会务" in compact:
            return "meeting", "会务费"
        if any(keyword in compact for keyword in ("办公费", "办公用品", "文具", "耗材", "办公耗材", "打印纸", "办公设备", "键盘", "鼠标", "白板")):
            return "office", "办公费"
        if any(keyword in compact for keyword in ("培训费", "培训", "讲师费", "课时费", "课程费")):
            return "training", "培训费"
        if any(keyword in compact for keyword in ("通讯费", "话费", "流量费", "宽带费")):
            return "communication", "通讯费"
        if any(keyword in compact for keyword in ("福利费", "团建", "慰问", "节日福利", "体检费")):
            return "welfare", "福利费"
        return "other", str(value or "").strip() or "其他费用"

    def _resolve_required_review_keys(
        self,
        payload: UserAgentRequest,
        *,
        primary_expense_type: str,
        claim_groups: list[UserAgentReviewClaimGroup],
    ) -> set[str]:
        required = {"expense_type", "time_range", "amount", "reason", "attachments"}
        scene_codes = {
            str(item.group_code or "").strip()
            for item in claim_groups
            if str(item.group_code or "").strip()
        }
        if primary_expense_type:
            scene_codes.add(primary_expense_type)

        for scene_code in scene_codes:
            required.update(SCENE_REQUIRED_SLOT_KEYS.get(scene_code, set()))

        compact_message = re.sub(r"\s+", "", self._resolve_reason_source_text(payload) or payload.message)
        if "entertainment" in scene_codes or (
            "客户" in compact_message and any(keyword in compact_message for keyword in ("招待", "吃饭", "用餐", "宴请", "请客"))
        ):
            required.update({"customer_name", "participants"})

        return required

    @staticmethod
    def _infer_reason_from_claim_groups(
        *,
        claim_groups: list[UserAgentReviewClaimGroup],
    ) -> str:
        if len(claim_groups) == 1:
            document_indexes = list(claim_groups[0].document_indexes or [])
            if not document_indexes:
                return ""

            expense_type = str(claim_groups[0].expense_type or "").strip()
            group_code = str(claim_groups[0].group_code or "").strip()
            if expense_type:
                return INFERRED_REASON_LABELS.get(expense_type, "") or str(claim_groups[0].scene_label or "").strip()
            if group_code:
                return INFERRED_REASON_LABELS.get(group_code, "") or str(claim_groups[0].scene_label or "").strip()
        return ""

    @staticmethod
    def _resolve_review_missing_slot_keys(
        payload: UserAgentRequest,
        *,
        slot_cards: list[UserAgentReviewSlotCard],
    ) -> list[str]:
        required_keys = {item.key for item in slot_cards if item.required}
        slot_map = {item.key: item for item in slot_cards}
        missing_keys = {
            item.key
            for item in slot_cards
            if item.required and (item.status == "missing" or not str(item.value).strip())
        }
        for key in payload.ontology.missing_slots:
            normalized_key = str(key or "").strip()
            if (
                normalized_key
                and normalized_key in required_keys
                and (
                    normalized_key not in slot_map
                    or slot_map[normalized_key].status == "missing"
                    or not str(slot_map[normalized_key].value).strip()
                )
            ):
                missing_keys.add(normalized_key)

        ordered_keys: list[str] = []
        for item in slot_cards:
            if item.required and item.key in missing_keys and item.key not in ordered_keys:
                ordered_keys.append(item.key)
        return ordered_keys

    def _make_slot_card(
        self,
        *,
        key: str,
        value: str,
        raw_value: str,
        normalized_value: str,
        source: str,
        confidence: float,
        evidence: str,
        required: bool = True,
    ) -> UserAgentReviewSlotCard:
        is_missing = required and not str(value).strip()
        source_key = source if source in SOURCE_LABELS else "system"
        return UserAgentReviewSlotCard(
            key=key,
            label=SLOT_LABELS.get(key, key),
            value=str(value or "").strip(),
            raw_value=str(raw_value or "").strip(),
            normalized_value=str(normalized_value or "").strip(),
            source=source,
            source_label=SOURCE_LABELS.get(source_key, "系统判断"),
            confidence=confidence,
            required=required,
            confirmed=not is_missing and source in {"user_text", "user_form"},
            status="missing" if is_missing else "identified" if source in {"user_text", "user_form"} else "inferred",
            hint=f"建议补充 {SLOT_LABELS.get(key, key)}。"
            if is_missing and required
            else ("该字段来自系统辅助上下文，建议你再核对一次。" if source in {"detail_context", "ocr"} else ""),
            evidence=evidence,
        )

    def _classify_document(
        self,
        item: dict[str, object],
        payload: UserAgentRequest,
    ) -> dict[str, str]:
        provided_type = str(item.get("document_type") or "").strip().lower()
        expense_type_code = self._collect_entity_values(payload).get("expense_type_code", "")
        has_customer = bool(self._collect_entity_values(payload).get("customer"))
        if provided_type:
            if provided_type in {"flight_itinerary", "train_ticket"}:
                return {
                    "document_type": provided_type,
                    "expense_type": "travel",
                    "group_code": "travel",
                    "scene_label": "差旅票据",
                }
            if provided_type == "hotel_invoice":
                return {
                    "document_type": provided_type,
                    "expense_type": "hotel",
                    "group_code": "travel",
                    "scene_label": "住宿票据",
                }
            if provided_type in {"taxi_receipt", "parking_toll_receipt"}:
                return {
                    "document_type": provided_type,
                    "expense_type": "transport",
                    "group_code": "travel",
                    "scene_label": "交通票据",
                }
            if provided_type == "meal_receipt":
                group_code = "entertainment" if expense_type_code == "entertainment" or has_customer else "meal"
                return {
                    "document_type": provided_type,
                    "expense_type": group_code,
                    "group_code": group_code,
                    "scene_label": "餐饮票据",
                }
            if provided_type == "office_invoice":
                return {
                    "document_type": provided_type,
                    "expense_type": "office",
                    "group_code": "office",
                    "scene_label": "办公用品票据",
                }
            if provided_type == "meeting_invoice":
                return {
                    "document_type": provided_type,
                    "expense_type": "meeting",
                    "group_code": "meeting",
                    "scene_label": "会务票据",
                }
            if provided_type == "training_invoice":
                return {
                    "document_type": provided_type,
                    "expense_type": "training",
                    "group_code": "training",
                    "scene_label": "培训票据",
                }

        text = " ".join(
            [
                str(item.get("filename") or ""),
                str(item.get("summary") or ""),
                str(item.get("text") or ""),
            ]
        ).lower()
        compact = text.replace(" ", "")

        if any(keyword in compact for keyword in ("机票", "航班", "火车", "高铁", "行程单")):
            return {
                "document_type": "travel_ticket",
                "expense_type": "travel",
                "group_code": "travel",
                "scene_label": "差旅票据",
            }
        if any(keyword in compact for keyword in ("酒店", "住宿", "宾馆")):
            return {
                "document_type": "hotel_invoice",
                "expense_type": "hotel",
                "group_code": "travel",
                "scene_label": "住宿票据",
            }
        if any(keyword in compact for keyword in ("打车", "出租车", "滴滴", "网约车", "过路费", "停车")):
            return {
                "document_type": "transport_receipt",
                "expense_type": "transport",
                "group_code": "travel",
                "scene_label": "交通票据",
            }
        if any(keyword in compact for keyword in ("餐", "饭店", "酒楼", "酒家", "餐饮", "meal")):
            group_code = "entertainment" if expense_type_code == "entertainment" or has_customer else "meal"
            return {
                "document_type": "meal_receipt",
                "expense_type": group_code,
                "group_code": group_code,
                "scene_label": "餐饮票据",
            }
        if any(keyword in compact for keyword in ("办公用品", "文具", "耗材", "办公耗材", "打印纸", "键盘", "鼠标", "白板", "墨盒", "硒鼓")):
            return {
                "document_type": "other",
                "expense_type": "office",
                "group_code": "office",
                "scene_label": "办公用品票据",
            }
        return {
            "document_type": "other",
            "expense_type": expense_type_code or "other",
            "group_code": self._normalize_group_code(expense_type_code or "other"),
            "scene_label": "其他票据",
        }

    @staticmethod
    def _normalize_group_code(expense_type_code: str) -> str:
        if expense_type_code in {"travel", "hotel", "transport"}:
            return "travel"
        if expense_type_code in {"entertainment", "meal", "office", "training", "communication", "welfare"}:
            return expense_type_code
        return "other"

    def _extract_document_fields(self, item: dict[str, object]) -> dict[str, str]:
        raw_fields = item.get("document_fields")
        normalized_fields: dict[str, str] = {}
        if isinstance(raw_fields, list):
            for field in raw_fields:
                if not isinstance(field, dict):
                    continue
                key = str(field.get("key") or "").strip()
                label = str(field.get("label") or "").strip()
                value = str(field.get("value") or "").strip()
                if not value:
                    continue
                normalized_label = self._normalize_document_field_label(key=key, label=label)
                display_label = normalized_label or label
                normalized_value = self._normalize_document_field_value(
                    label=display_label,
                    value=value,
                )
                if display_label and normalized_value:
                    normalized_fields.setdefault(display_label, normalized_value)

        text = " ".join([str(item.get("summary") or ""), str(item.get("text") or "")]).strip()
        amount_value = self._extract_amount_text_from_value(text)
        if amount_value and "金额" not in normalized_fields:
            normalized_fields["金额"] = amount_value
        date_match = DATE_TEXT_PATTERN.search(text)
        if date_match and "时间" not in normalized_fields:
            normalized_fields["时间"] = date_match.group(1)

        merchant = self._extract_document_merchant_name_from_text(text)
        if merchant and "商户/酒店" not in normalized_fields:
            normalized_fields["商户/酒店"] = merchant
        return normalized_fields

    @staticmethod
    def _normalize_document_field_label(*, key: str, label: str) -> str:
        compact_key = str(key or "").strip().lower().replace("_", "")
        compact_label = str(label or "").replace(" ", "")
        if compact_key in {
            "amount",
            "totalamount",
            "paymentamount",
            "paidamount",
            "actualamount",
        } or any(
            token in compact_label
            for token in ("金额", "价税合计", "合计", "总额", "总计", "票价", "支付金额", "实付金额", "实收金额")
        ):
            return "金额"
        if compact_key in {"date", "time", "issuedat", "invoicedate"} or any(
            token in compact_label for token in ("日期", "时间", "开票日期", "发生时间")
        ):
            return "时间"
        if compact_key in {"merchant", "merchantname", "sellername", "vendorname"} or any(
            token in compact_label for token in ("商户", "酒店", "销售方", "开票方", "收款方")
        ):
            return "商户/酒店"
        return label

    def _normalize_document_field_value(self, *, label: str, value: str) -> str:
        normalized_label = str(label or "").strip()
        raw_value = str(value or "").strip()
        if not normalized_label or not raw_value:
            return ""
        if normalized_label == "金额":
            return self._extract_amount_text_from_value(raw_value) or raw_value
        if normalized_label == "时间":
            match = DATE_TEXT_PATTERN.search(raw_value)
            return match.group(1) if match else raw_value
        return raw_value

    def _extract_amount_text_from_value(self, value: str) -> str:
        raw_value = str(value or "").strip()
        if not raw_value:
            return ""
        best_amount: Decimal | None = None
        for pattern in (DOCUMENT_AMOUNT_PATTERN, DOCUMENT_CURRENCY_AMOUNT_PATTERN, AMOUNT_TEXT_PATTERN):
            for match in pattern.finditer(raw_value):
                try:
                    candidate = Decimal(str(match.group(1)).replace(",", "."))
                except (InvalidOperation, TypeError):
                    continue
                if candidate <= Decimal("0.00"):
                    continue
                if best_amount is None or candidate > best_amount:
                    best_amount = candidate
        if best_amount is None:
            return ""
        return f"{best_amount.quantize(Decimal('0.01')):.2f}元"

    def _extract_document_merchant_name(self, item: dict[str, object]) -> str:
        fields = self._extract_document_fields(item)
        merchant = str(fields.get("商户/酒店") or "").strip()
        if merchant:
            return merchant
        text = " ".join([str(item.get("summary") or ""), str(item.get("text") or "")]).strip()
        return self._extract_document_merchant_name_from_text(text)

    @staticmethod
    def _extract_document_merchant_name_from_text(text: str) -> str:
        for keyword in ("酒店", "宾馆", "饭店", "酒楼", "餐厅", "航空", "铁路", "滴滴"):
            if keyword in text:
                return keyword
        return ""

    @staticmethod
    def _extract_amount_from_card(card: UserAgentReviewDocumentCard) -> float:
        for item in card.fields:
            if item.label != "金额":
                continue
            try:
                normalized_value = str(item.value).replace("元", "").replace("￥", "").replace("¥", "").strip()
                return float(normalized_value)
            except ValueError:
                return 0.0
        return 0.0

    def _resolve_amount_value(self, payload: UserAgentRequest) -> float:
        for item in payload.ontology.entities:
            if item.type == "amount" and item.role != "threshold":
                try:
                    return float(item.normalized_value)
                except ValueError:
                    return 0.0
        return 0.0

    def _sum_ocr_amounts(self, ocr_documents: list[dict[str, object]]) -> float:
        total = 0.0
        for item in ocr_documents:
            fields = self._extract_document_fields(item)
            amount_text = str(fields.get("金额") or "").replace("元", "").replace("￥", "").replace("¥", "").strip()
            if not amount_text:
                continue
            try:
                total += float(amount_text)
            except ValueError:
                continue
        return total

    def _infer_expense_type_from_documents(
        self,
        payload: UserAgentRequest,
        ocr_documents: list[dict[str, object]],
    ) -> str:
        labels: list[str] = []
        for item in ocr_documents:
            classified = self._classify_document(item, payload)
            label = GROUP_SCENE_LABELS.get(classified["group_code"], "")
            if label and label not in labels:
                labels.append(label)
        return " + ".join(labels[:3])