Initial commit: macOS receipt workspace

native-engine/build-engine.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+set -euo pipefail
+ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+PYTHON="$ROOT/.build-tools/bin/python"
+if [ ! -x "$PYTHON" ]; then
+    echo "请先使用 Python 3.12 创建 .build-tools 虚拟环境并安装 native-engine/requirements.txt。" >&2
+    exit 1
+fi
+export PYINSTALLER_CONFIG_DIR="$ROOT/.build-tools/pyinstaller-cache"
+cd "$ROOT/native-engine"
+"$PYTHON" -m PyInstaller --noconfirm --clean --onedir --windowed --name receipt-engine-helper \
+    --osx-bundle-identifier test.reimburse.engine \
+    --distpath "$ROOT/native-engine/dist" --workpath "$ROOT/native-engine/build" \
+    --collect-all rapidocr_onnxruntime --collect-all onnxruntime \
+    --collect-all pypdfium2 --collect-all pypdfium2_raw \
+    --add-data "personal-expense-template.xlsx:." engine.py
+echo "本地处理引擎已打包，可以在 Xcode 中运行 reimburse。"

native-engine/domain.py

@@ -0,0 +1,358 @@
+import re
+import uuid
+from collections import OrderedDict
+from datetime import date
+from decimal import Decimal, InvalidOperation, ROUND_HALF_UP
+
+CATEGORIES = ['餐饮', '住宿', '交通', '办公用品', '招待', '采购', '其他']
+HAN = r'\u3400-\u9fff'
+DATE = r'(20\d{2})[年./-](1[0-2]|0?[1-9])[月./-](3[01]|[12]\d|0?[1-9])日?'
+AMOUNT_PATTERNS = [
+    r'(?:支付总额|订单总额|应付总额|总金额|总计|价税合计|小写|实付金额|支付金额|付款金额|订单金额|合计|金额)[^0-9]{0,10}([0-9]{1,8}(?:[.,][0-9]{1,2})?)',
+    r'(?:¥|￥|RMB|CNY)\s*([0-9]{1,8}(?:[.,][0-9]{1,2})?)',
+    r'(?<![0-9])([0-9]{1,7}[.,][0-9]{2})(?![0-9%])',
+]
+MERCHANT_LABEL = r'销售方名称|销方名称|收款方|收款单位|商户全称|商户名称|商户|付款给|收款人|对方户名|交易对象'
+KEYWORDS = [
+    '餐饮 餐厅 饭店 酒楼 外卖 咖啡 奶茶 茶饮 饮品 火锅 烧烤 快餐 小吃 食品 美团外卖 饿了么 tims 星巴克',
+    '酒店 宾馆 旅馆 住宿 房费 客房 民宿 公寓 入住 退房',
+    '铁路 火车 12306 车次 机票 航空 航班 登机 滴滴 网约车 出租车 打车 货运 运输 加油站 汽油 柴油 停车 高速 地铁 公交',
+    '办公用品 办公 文具 打印 复印 纸张 墨盒 硒鼓 耗材 电脑 键盘 鼠标 显示器 软件 会员订阅 桌椅 文件夹 饮用水 桶装水',
+    '招待 接待 宴请 商务宴请 客户用餐 会议餐',
+    '采购 商品 材料 设备 配件 家具 日用品 百货 商城 京东 淘宝 天猫 超市 便利店 购物',
+]
+
+
+def unique(values):
+    return list(dict.fromkeys(values))
+
+
+def money(value):
+    return Decimal(str(value).replace(',', '.')).quantize(Decimal('.01'), rounding=ROUND_HALF_UP)
+
+
+def first(pattern, text):
+    match = re.search(pattern, text, re.I)
+    return match.group(1).strip() if match else ''
+
+
+def extract(lines):
+    text = '\n'.join(lines)
+    amounts = unique(format(money(value), '.2f') for pattern in AMOUNT_PATTERNS
+                     for value in re.findall(pattern, text, re.I) if money(value) >= 0)
+    dates = []
+    for match in re.finditer(DATE, text):
+        try:
+            dates.append(date(*map(int, match.groups())).isoformat())
+        except ValueError:
+            pass
+    merchants = re.findall(r'(?:' + MERCHANT_LABEL + r')[：:]?\s*([^\n]{2,40})', text)
+
+    def clean(value):
+        return re.sub(r'[|：:]+$', '', re.sub(r'^.*?(?:' + MERCHANT_LABEL.replace('|商户|', '|') + r')[：:]?', '', value)).strip()
+
+    def valid(value):
+        return (2 <= len(value) <= 45 and re.search('[' + HAN + 'A-Za-z]', value)
+                and not re.search('支付成功|交易成功|电子发票|增值税发票|付款方式|订单详情|账单详情|交易详情|商品说明|开票日期|价税合计', value))
+
+    for index, line in enumerate(lines):
+        line = clean(line)
+        if re.search('销售方|销方|收款方|收款单位|商户|付款给|收款人|对方户名|交易对象', line) and index + 1 < len(lines):
+            following = clean(lines[index + 1])
+            if valid(following):
+                merchants.append(following)
+        if re.search('公司|餐厅|饭店|酒店|宾馆|旅馆|商店|店|超市|便利店|中心|服务部|经营部|商行|科技|网络|运输|航空|铁路|医院|药房|加油站|停车场|物业|平台', line) and valid(line):
+            merchants.append(line)
+    return dict(status='success', rawText=text, lines=lines, amounts=amounts,
+                dates=unique(dates), merchants=unique(value.strip().rstrip('：:|') for value in merchants),
+                orderNumbers=unique(re.findall(r'(?:交易单号|商户单号|订单号|流水号|交易号)[：:]?\s*([A-Za-z0-9-]{8,40})', text)),
+                travel=extract_travel(text, unique(dates)), message='')
+
+
+def extract_travel(text, dates):
+    travel = dict(type='', travelerName='', departure='', destination='', departureTime='', transportNumber='')
+    train = re.search('火车票|铁路电子客票|铁路|车次|检票口|12306', text)
+    flight = re.search('航空|航班|登机牌|电子客票行程单|乘机|起飞|客票号', text)
+    if not train and not flight:
+        return travel
+    travel['type'] = 'flight' if flight and not train else 'train'
+    for pattern in [r'(?:旅客姓名|乘车人|乘机人|旅客|姓名)[：:]?\s*([' + HAN + r'·]{2,15})',
+                    r'(?:\d{6,}\*{2,}\d{2,})\s*([' + HAN + r'·]{2,8})']:
+        names = [name for name in re.findall(pattern, text) if len(name) <= 8 and not re.search('税务|日期|发票|铁路|车站|公司|名称|代码|号码', name)]
+        if names:
+            travel['travelerName'] = names[0]
+            break
+    travel['departure'] = first(r'(?:出发地|出发站|始发站|起点|起飞机场)[：:]?\s*([' + HAN + r'A-Za-z0-9]{2,24})', text)
+    travel['destination'] = first(r'(?:目的地|到达站|终点站|终点|到达机场)[：:]?\s*([' + HAN + r'A-Za-z0-9]{2,24})', text)
+    route = re.search(r'([' + HAN + r'A-Za-z]{2,20})(?:站|机场)?\s*(?:→|->|—|至)\s*([' + HAN + r'A-Za-z]{2,20})(?:站|机场)?', text)
+    if route:
+        for index, field in enumerate(['departure', 'destination'], 1):
+            travel[field] = travel[field] or re.sub(r'(站|机场)$', '', route.group(index))
+    if train:
+        stations = unique(re.findall(r'([' + HAN + r']{2,12})站', text))
+        if len(stations) >= 2:
+            travel['departure'] = travel['departure'] or stations[0]
+            travel['destination'] = travel['destination'] or stations[1]
+    clock_pattern = r'([01]?\d|2[0-3]):([0-5]\d)'
+    for match in re.finditer(DATE + r'[^\n]{0,24}?' + clock_pattern, text):
+        if not re.search('开票日期|发票日期', text[max(0, match.start() - 16):match.start()]):
+            year, month, day, hour, minute = map(int, match.groups())
+            travel['departureTime'] = f'{year:04}-{month:02}-{day:02} {hour:02}:{minute:02}'
+            break
+    if not travel['departureTime']:
+        departure_date = ''
+        for match in re.finditer(DATE, text):
+            if not re.search('开票日期|发票日期', text[max(0, match.start() - 16):match.start()]):
+                year, month, day = map(int, match.groups())
+                departure_date = f'{year:04}-{month:02}-{day:02}'
+                break
+        clock = re.search(clock_pattern, text)
+        travel['departureTime'] = ((departure_date or next(iter(dates), '')) + (' %02d:%02d' % tuple(map(int, clock.groups())) if clock else '')).strip()
+    pattern = r'(?<![A-Z0-9])([GDCZTKSLY]\d{1,5})(?![A-Z0-9])' if train else r'(?<![A-Z0-9])([A-Z0-9]{2}\d{3,4})(?![A-Z0-9])'
+    travel['transportNumber'] = first(pattern, text)
+    return travel
+
+
+def amounts(item):
+    return [money(value) for value in item['ocr']['amounts'] if money(value) > 0] if item['ocr']['status'] == 'success' else []
+
+
+def primary(item):
+    values = item['ocr']['amounts']
+    return money(values[0]) if values and item['ocr']['status'] == 'success' else None
+
+
+def text(item):
+    return item['ocr']['rawText']
+
+
+def common(left, right, field):
+    return [value for value in left['ocr'][field] if value in right['ocr'][field]]
+
+
+def date_distance(invoices, payments):
+    distances = [abs((date.fromisoformat(left) - date.fromisoformat(right)).days)
+                 for invoice in invoices for payment in payments
+                 for left in invoice['ocr']['dates'] for right in payment['ocr']['dates']]
+    return min(distances, default=999999)
+
+
+def similarity(left, right):
+    def normalize(value):
+        return re.sub('[^' + HAN + 'a-z0-9]', '', re.sub('有限责任公司|股份有限公司|有限公司|公司|商户|门店|收款方', '', value.lower()))
+    def grams(value):
+        return {value[index:index + 2] for index in range(len(value) - 1)} if len(value) > 1 else {value}
+    left, right = normalize(left), normalize(right)
+    if not left or not right:
+        return 0
+    if left in right or right in left:
+        return 1
+    return len(grams(left) & grams(right)) / len(grams(left) | grams(right))
+
+
+def merchant_similarity(invoice, payment):
+    return max((similarity(left, right) for left in invoice['ocr']['merchants'] for right in payment['ocr']['merchants']), default=0)
+
+
+def category(items):
+    content = ' '.join(text(item) + ' ' + ' '.join(item['ocr']['merchants']) for item in items).lower()
+    scores = [sum(3 if len(word) >= 4 else 2 if len(word) >= 3 else 1 for word in words.split() if word in content) for words in KEYWORDS]
+    return CATEGORIES[scores.index(max(scores))] if max(scores) else '其他'
+
+
+def add_match(state, invoices, payments, score, reasons):
+    selected_category = category(invoices + payments)
+    for item in invoices + payments:
+        item['matched'] = True
+    state['matches'].append(dict(id=str(uuid.uuid4()), invoices=invoices, payments=payments,
+                                 category=selected_category, matchType='auto', score=score,
+                                 reasons=reasons + ['自动分类：' + selected_category]))
+
+
+def combinations(values, target):
+    sums = {Decimal('0.00'): [[]]}
+    for item, amount in values:
+        snapshot = [(subtotal, [list(group) for group in groups]) for subtotal, groups in sums.items()]
+        for subtotal, groups in snapshot:
+            total = subtotal + amount
+            if total > target:
+                continue
+            destination = sums.setdefault(total, [])
+            for group in list(groups):
+                candidate = group + [item]
+                if candidate not in destination:
+                    destination.append(candidate)
+                    del destination[2:]
+    return [group for group in sums.get(target, []) if len(group) >= 2][:2]
+
+
+def railway(item):
+    return bool(re.search('12306|铁路|火车|车次|中铁', text(item)))
+
+
+def total_label(item):
+    return bool(re.search('支付总额|订单总额|应付总额|总金额|总计|合计', text(item)))
+
+
+def travel_key(item):
+    travel = item['ocr']['travel']
+    return tuple(travel[field].lower() for field in ['type', 'departure', 'destination', 'departureTime', 'transportNumber'])
+
+
+def auto_match(state):
+    invoices, payments = state['invoices'], state['payments']
+    groups = OrderedDict()
+    for invoice in invoices:
+        if invoice['ocr']['travel']['type']:
+            groups.setdefault(travel_key(invoice), []).append(invoice)
+    for group in groups.values():
+        if len(group) < 2 or any(primary(item) is None for item in group):
+            continue
+        total = sum(primary(item) for item in group)
+        evidence = [item for item in payments if not item['matched'] and railway(item) and total in amounts(item)]
+        if evidence and any(re.search('支付成功|付款成功|交易成功|支付方式|交易单号|商户单号', text(item)) for item in evidence) and date_distance(group, evidence) <= 14:
+            reasons = [f'同一行程多人发票合计一致 ¥{total:.2f}', ' + '.join(str(primary(item)) for item in group)]
+            if len(evidence) >= 2:
+                reasons.append('付款凭证与订单截图已共同归入本次行程')
+            add_match(state, group, evidence, 100 if len(evidence) >= 2 else 95, reasons)
+    for payment in payments:
+        total = primary(payment)
+        if payment['matched'] or total is None:
+            continue
+        values = [(item, primary(item)) for item in invoices if not item['matched'] and primary(item) is not None and 0 < primary(item) < total]
+        options = combinations(values, total)
+        if len(options) != 1:
+            continue
+        group, evidence = options[0], [payment]
+        for candidate in payments:
+            if candidate is payment or candidate['matched']:
+                continue
+            same_total = primary(candidate) == total or (total in amounts(candidate) and total_label(candidate))
+            related = (common(payment, candidate, 'orderNumbers') or merchant_similarity(payment, candidate) >= .35
+                       or (railway(payment) and railway(candidate) and (date_distance([payment], [candidate]) <= 7 or total_label(payment) or total_label(candidate))))
+            if same_total and related:
+                evidence.append(candidate)
+        score = 70
+        reasons = [f'多张发票合计一致 ¥{total:.2f}', ' + '.join(str(primary(item)) for item in group)]
+        for condition, reason in [(len(evidence) >= 2, '订单与付款凭证相互印证'),
+                                  (date_distance(group, evidence) <= 7, '行程与付款日期接近'),
+                                  (all(item['ocr']['travel']['type'] for item in group) and len({tuple(item['ocr']['travel'][field] for field in ['departure', 'destination', 'departureTime', 'transportNumber']) for item in group}) == 1, '车次、路线及行程日期一致')]:
+            if condition:
+                score += 10
+                reasons.append(reason)
+        if score >= 90:
+            add_match(state, group, evidence, score, reasons)
+    candidates = []
+    for invoice in invoices:
+        for payment in payments:
+            shared = common(invoice, payment, 'amounts')
+            if not shared or invoice['ocr']['status'] != 'success' or payment['ocr']['status'] != 'success':
+                continue
+            score, reasons = 55, ['金额一致 ¥' + shared[0]]
+            merchant = merchant_similarity(invoice, payment)
+            if merchant >= .35:
+                score += int(25 * merchant + .5)
+                reasons.append(f'商户相似 {int(merchant * 100 + .5)}%')
+            distance = date_distance([invoice], [payment])
+            if distance <= 7:
+                score += 15 if distance <= 3 else 8
+                reasons.append('日期一致' if distance == 0 else f'日期相差 {distance} 天')
+            if common(invoice, payment, 'orderNumbers'):
+                score += 20
+                reasons.append('交易单号一致')
+            for amount in shared:
+                if sum(amount in item['ocr']['amounts'] for item in invoices) == 1 and sum(amount in item['ocr']['amounts'] for item in payments) == 1:
+                    score += 20
+                    reasons.append(f'金额 ¥{amount} 在本批材料中唯一')
+                    break
+            if score >= 70:
+                candidates.append((invoice, payment, min(100, score), reasons))
+    candidates.sort(key=lambda candidate: -candidate[2])
+    for candidate in candidates:
+        invoice, payment, score, reasons = candidate
+        if invoice['matched'] or payment['matched']:
+            continue
+        second = max((other[2] for other in candidates if other is not candidate and (other[0] is invoice or other[1] is payment)
+                      and not other[0]['matched'] and not other[1]['matched']), default=0)
+        if score >= 90 or score - second >= 8:
+            add_match(state, [invoice], [payment], score, reasons)
+    for invoice in invoices:
+        if invoice['matched']:
+            continue
+        available = sorted([(item, primary(item)) for item in payments if not item['matched'] and primary(item) is not None and primary(item) > 0], key=lambda entry: entry[1])
+        options = []
+        for total in amounts(invoice):
+            options.extend((total, group) for group in combinations([(item, amount) for item, amount in available if amount < total], total))
+            if len(options) > 1:
+                break
+        if len(options) == 1:
+            total, evidence = options[0]
+            add_match(state, [invoice], evidence, 95, [f'多笔付款合计一致 ¥{total:.2f}', ' + '.join(str(primary(item)) for item in evidence)])
+    return state
+
+
+def material_date(item, issue=False):
+    content = text(item)
+    if issue:
+        match = re.search(r'(?:申请开票(?:日期|时间)?|开票日期|发票开具日期|开具日期)\s*[：:]?\s*' + DATE, content)
+        if match:
+            try:
+                return date(*map(int, match.groups())).isoformat()
+            except ValueError:
+                pass
+    if item['ocr']['dates']:
+        return item['ocr']['dates'][0]
+    match = re.search(r'(?:^|\D)(\d{1,2})月(\d{1,2})日', content)
+    if match:
+        try:
+            return date(date.today().year, *map(int, match.groups())).isoformat()
+        except ValueError:
+            pass
+    return '9999-12-31'
+
+
+def invoice_amount(item):
+    content = text(item)
+    patterns = [r'价税合计[^\n]{0,35}?(?:小写)?[^0-9]{0,12}([0-9][0-9,]*(?:\.\d{1,2})?)',
+                r'(?:小写|票价|发票金额|合计金额|总金额|应付金额)\s*[（(]?[人民币RMB]*[）)]?\s*[：:]?\s*[¥￥]?\s*([0-9][0-9,]*(?:\.\d{1,2})?)']
+    for pattern in patterns:
+        values = re.findall(pattern, content, re.I)
+        if values:
+            return Decimal(values[-1].replace(',', ''))
+    return max([Decimal(value) for value in item['ocr']['amounts']], default=Decimal(0))
+
+
+def payment_amount(item):
+    for pattern in [r'(?:支付总额|实付金额|实际支付金额|付款金额|实付款|实付价|合计)\s*[：:]?\s*[¥￥]?\s*([0-9][0-9,]*(?:\.\d{1,2})?)',
+                    r'已优惠[^\n]{0,24}?[¥￥]\s*([0-9][0-9,]*(?:\.\d{1,2})?)']:
+        values = re.findall(pattern, text(item), re.I)
+        if values:
+            return abs(Decimal(values[-1].replace(',', '')))
+    return abs(primary(item) or Decimal(0))
+
+
+def payment_total(items):
+    result = Decimal(0)
+    for index, item in enumerate(items):
+        amount = payment_amount(item)
+        duplicate = re.search('支付总额|查看订单支付明细|购票成功|订单详情', text(item)) and any(
+            other_index != index and re.search('支付成功|付款成功|交易成功|支付方式|付款方式|交易单号|商户单号|转账成功|收款方', text(other))
+            and abs(payment_amount(other) - amount) < Decimal('.005') for other_index, other in enumerate(items))
+        if not duplicate:
+            result += amount
+    return result
+
+
+def enrich(state):
+    for item in state['invoices'] + state['payments'] + state['photos']:
+        item['displayAmount'] = float(invoice_amount(item) if item['type'] == 'invoice' else payment_amount(item))
+        item['sortDate'] = material_date(item)
+        item['issueDate'] = material_date(item, True)
+    lookup = {item['id']: item for item in state['invoices'] + state['payments']}
+    for match in state['matches']:
+        match['invoices'] = [lookup[item['id']] for item in match['invoices']]
+        match['payments'] = [lookup[item['id']] for item in match['payments']]
+        match['paymentTotal'] = float(payment_total(match['payments']))
+        match['expenseAmount'] = float(sum(max((Decimal(value) for value in item['ocr']['amounts']), default=Decimal(0)) for item in match['invoices']))
+    state['directoryPaymentTotal'] = float(payment_total(state['payments']))
+    return state

native-engine/engine.py

@@ -0,0 +1,118 @@
+import contextlib
+import json
+import os
+import shutil
+import sys
+import traceback
+import uuid
+from pathlib import Path
+
+from domain import auto_match, enrich, extract
+from exports import export_expense, export_ppt, export_travel
+
+PROTOCOL = sys.stdout
+SUPPORTED = {'.jpg', '.jpeg', '.png', '.webp', '.bmp', '.gif', '.pdf'}
+
+
+def emit(value):
+    PROTOCOL.write(json.dumps(value, ensure_ascii=False) + '\n')
+    PROTOCOL.flush()
+
+
+def scan(request):
+    import onnxruntime
+    import pypdfium2 as pdfium
+    from PIL import Image
+    from rapidocr_onnxruntime import RapidOCR
+    root = Path(request['path'])
+    destination = Path(request['workspacePath'])
+    if not root.is_dir():
+        raise ValueError('请选择有效的报账材料文件夹')
+    destination.mkdir(parents=True, exist_ok=True)
+    state = dict(rootPath=request.get('folderName', root.name), invoices=[], payments=[], photos=[], matches=[], warnings=[], directoryPaymentTotal=0)
+    files = []
+    for source in sorted(root.rglob('*')):
+        if source.is_file() and source.suffix.lower() in SUPPORTED:
+            parts = source.relative_to(root).parts[:-1]
+            kind = next((kind for label, kind in [('发票', 'invoice'), ('付款截图', 'payment'), ('实物照片', 'photo')] if label in parts), None)
+            if kind:
+                files.append((source, kind))
+    if not files:
+        raise ValueError('目录中没有可处理材料，请确认包含“发票”“付款截图”子文件夹')
+    emit(dict(event='progress', progress=0, message='正在加载本地 RapidOCR 模型'))
+    onnxruntime.disable_telemetry_events()
+    recognizer = RapidOCR()
+    for index, (source, kind) in enumerate(files):
+        identifier = str(uuid.uuid4())
+        stored = destination / (identifier + source.suffix.lower())
+        shutil.copyfile(source, stored)
+        item = dict(id=identifier, name=source.name, path=str(stored), type=kind, size=stored.stat().st_size, matched=False,
+                    previewPath='', ocr=extract([]))
+        item['ocr']['status'] = 'pending'
+        try:
+            if source.suffix.lower() == '.pdf':
+                with contextlib.closing(pdfium.PdfDocument(str(stored))) as document:
+                    if not len(document):
+                        raise ValueError('PDF 没有可预览页面')
+                    with contextlib.closing(document[0]) as page:
+                        with contextlib.closing(page.render(scale=140 / 72)) as bitmap:
+                            image = bitmap.to_pil().convert('RGB')
+            else:
+                with Image.open(stored) as original:
+                    image = original.convert('RGB')
+            preview = destination / (identifier + '-preview.png')
+            image.save(preview, format='PNG')
+            item['previewPath'] = str(preview)
+            if kind != 'photo':
+                result, _ = recognizer(str(preview if source.suffix.lower() == '.pdf' else stored))
+                item['ocr'] = extract([str(entry[1]).strip() for entry in result or [] if len(entry) >= 3 and str(entry[1]).strip()])
+        except Exception as error:
+            item['ocr']['status'] = 'failed'
+            item['ocr']['message'] = str(error)
+            state['warnings'].append(source.name + '：识别或预览失败，可人工配对')
+        state[{'invoice': 'invoices', 'payment': 'payments', 'photo': 'photos'}[kind]].append(item)
+        emit(dict(event='progress', progress=(index + 1) / len(files) * .95, message=f'已处理 {index + 1}/{len(files)} · {source.name}'))
+    for field, label in [('invoices', '发票'), ('payments', '付款截图'), ('photos', '实物照片')]:
+        if not state[field]:
+            state['warnings'].append('未找到“' + label + '”文件夹或支持的文件')
+    emit(dict(event='progress', progress=.97, message='正在执行自动配对与分类'))
+    return enrich(auto_match(state))
+
+
+def dispatch(request):
+    operation = request['operation']
+    if operation == 'scan':
+        return scan(request)
+    state = request['state']
+    if operation == 'refresh':
+        return enrich(state)
+    if not state['matches']:
+        raise ValueError('至少完成一组核对后才能导出')
+    destination = Path(request['destination'])
+    temporary = destination.with_name('.' + str(uuid.uuid4()) + destination.suffix)
+    try:
+        if operation == 'ppt':
+            export_ppt(state, temporary, request.get('classified', True))
+        elif operation == 'travel':
+            export_travel(state, temporary)
+        elif operation == 'expense':
+            template = Path(__file__).parent / 'personal-expense-template.xlsx'
+            export_expense(state, temporary, template, request.get('purposes', {}), request.get('signatures', []))
+        else:
+            raise ValueError('未知操作：' + operation)
+        os.replace(temporary, destination)
+    finally:
+        temporary.unlink(missing_ok=True)
+    return dict(destination=str(destination))
+
+
+if __name__ == '__main__':
+    try:
+        request = json.loads(sys.stdin.readline())
+        with contextlib.redirect_stdout(sys.stderr):
+            result = dispatch(request)
+        emit(dict(event='result', result=result))
+    except Exception as error:
+        traceback.print_exc(file=sys.stderr)
+        emit(dict(event='error', message=str(error)))
+        sys.exit(1)

native-engine/exports.py

@@ -0,0 +1,232 @@
+import copy
+import io
+import math
+import posixpath
+import zipfile
+from decimal import Decimal
+from pathlib import Path
+from xml.etree import ElementTree as XML
+
+from PIL import Image
+from domain import material_date, text
+
+DEFAULT_SIGNATURES = ['部门长', '剧组出纳', '制片主任', '剧组会计', '执行制片人', '制片人']
+
+
+def image_for(item):
+    return Image.open(item.get('previewPath') or item['path']).convert('RGB')
+
+
+def export_ppt(state, destination, classified):
+    from pptx import Presentation
+    from pptx.util import Pt
+    from pptx.dml.color import RGBColor
+    from pptx.enum.shapes import MSO_SHAPE
+    from pptx.enum.text import PP_ALIGN
+
+    presentation = Presentation()
+    presentation.slide_width, presentation.slide_height = Pt(1280), Pt(720)
+
+    def slide():
+        return presentation.slides.add_slide(presentation.slide_layouts[6])
+
+    def picture(page, item, box):
+        image = image_for(item)
+        left, top, width, height = box
+        scale = min(width / image.width, height / image.height)
+        actual_width, actual_height = int(image.width * scale), int(image.height * scale)
+        content = io.BytesIO()
+        image.save(content, format='PNG')
+        content.seek(0)
+        page.shapes.add_picture(content, Pt(left + (width - actual_width) // 2), Pt(top + (height - actual_height) // 2), Pt(actual_width), Pt(actual_height))
+
+    def payment_page(payments):
+        page = slide()
+        if len(payments) == 1:
+            picture(page, payments[0], (300, 20, 680, 680))
+        else:
+            width = (1200 - 24 * (len(payments) - 1)) // len(payments)
+            for index, item in enumerate(payments):
+                picture(page, item, (40 + index * (width + 24), 20, width, 680))
+
+    def label(page, box, value, size, title=False):
+        shape = page.shapes.add_shape(MSO_SHAPE.RECTANGLE, *(Pt(value) for value in box))
+        shape.fill.solid()
+        shape.fill.fore_color.rgb = RGBColor(245, 248, 247) if title else RGBColor(255, 255, 255)
+        shape.line.color.rgb = RGBColor(216, 226, 222) if title else RGBColor(170, 188, 182)
+        shape.line.width = Pt(1 if title else 1.5)
+        shape.text = value
+        for paragraph in shape.text_frame.paragraphs:
+            paragraph.alignment = PP_ALIGN.CENTER
+            for run in paragraph.runs:
+                run.font.name = 'PingFang SC'
+                run.font.size = Pt(size)
+                run.font.color.rgb = RGBColor(39, 91, 78) if title else RGBColor(112, 132, 126)
+
+    def photos(count):
+        for offset in range(0, count, 4):
+            slots = min(4, count - offset)
+            page = slide()
+            label(page, (55, 28, 1170, 52), f'实物照片粘贴区（请手动粘贴）  {offset + 1}-{min(count, offset + 4)} / {count}', 18, True)
+            for index in range(slots):
+                box = (250, 105, 780, 555) if slots == 1 else (55 + index * 610, 105, 560, 555) if slots == 2 else (55 + index % 2 * 610, 105 + index // 2 * 285, 560, 255)
+                label(page, box, f'实物照片 {offset + index + 1}\n请在此处粘贴', 16)
+
+    def requires_photo(match):
+        return match['category'].strip() not in ['交通', '住宿']
+
+    def simple_batch(batch):
+        if not batch:
+            return
+        for match in batch:
+            picture(slide(), match['invoices'][0], (40, 25, 1200, 670))
+        payment_page([match['payments'][0] for match in batch])
+        photos(sum(len(match['payments']) for match in batch if requires_photo(match)))
+
+    matches = sorted(state['matches'], key=lambda match: (min(material_date(item, True) for item in match['invoices']), match['category'] if classified else ''))
+    batch = []
+    for match in matches:
+        if len(match['invoices']) != 1 or len(match['payments']) != 1:
+            simple_batch(batch)
+            batch = []
+            for item in sorted(match['invoices'], key=material_date):
+                picture(slide(), item, (40, 25, 1200, 670))
+            payments = sorted(match['payments'], key=material_date)
+            for index in range(0, len(payments), 4):
+                payment_page(payments[index:index + 4])
+            if requires_photo(match):
+                photos(len(payments))
+        else:
+            batch.append(match)
+            if len(batch) == 2:
+                simple_batch(batch)
+                batch = []
+    simple_batch(batch)
+    presentation.save(destination)
+
+
+def export_travel(state, destination):
+    import mimetypes
+    mimetypes.knownfiles = []
+    mimetypes.init()
+    from openpyxl import Workbook
+    from openpyxl.styles import Font, PatternFill, Alignment
+    from openpyxl.utils import get_column_letter
+    invoices = [item for match in state['matches'] for item in match['invoices'] if item['ocr']['travel']['type']]
+    if not invoices:
+        raise ValueError('已核对材料中没有识别到火车票或航班行程')
+    workbook = Workbook()
+    sheet = workbook.active
+    sheet.title = '火车及航班行程'
+    sheet.append(['序号', '交通类型', '姓名', '出发地', '目的地', '出发时间', '车次/航班号', '金额', '发票文件'])
+    for cell in sheet[1]:
+        cell.font = Font(bold=True, color='FFFFFF')
+        cell.fill = PatternFill('solid', fgColor='006633')
+        cell.alignment = Alignment(horizontal='center')
+    for index, item in enumerate(invoices, 1):
+        travel = item['ocr']['travel']
+        values = [index, '火车' if travel['type'] == 'train' else '航班', travel['travelerName'], travel['departure'], travel['destination'], travel['departureTime'], travel['transportNumber'], next(iter(item['ocr']['amounts']), ''), item['name']]
+        sheet.append(values)
+        for cell in sheet[sheet.max_row]:
+            if isinstance(cell.value, str):
+                cell.data_type = 's'
+    for index, width in enumerate([8, 12, 16, 18, 18, 22, 18, 14, 38], 1):
+        sheet.column_dimensions[get_column_letter(index)].width = width
+    sheet.freeze_panes = 'A2'
+    workbook.save(destination)
+
+
+def export_expense(state, destination, template, purposes, signatures):
+    from lxml import etree as ET
+    namespace = 'http://schemas.openxmlformats.org/spreadsheetml/2006/main'
+    relationships = 'http://schemas.openxmlformats.org/officeDocument/2006/relationships'
+    package = 'http://schemas.openxmlformats.org/package/2006/relationships'
+    content_namespace = 'http://schemas.openxmlformats.org/package/2006/content-types'
+    tag = lambda name: '{' + namespace + '}' + name
+    with zipfile.ZipFile(template) as source:
+        parts = {name: source.read(name) for name in source.namelist()}
+    workbook = ET.fromstring(parts['xl/workbook.xml'])
+    relations = ET.fromstring(parts['xl/_rels/workbook.xml.rels'])
+    content_types = ET.fromstring(parts['[Content_Types].xml'])
+    sheets = workbook.find(tag('sheets'))
+    original = next(sheet for sheet in sheets if sheet.get('name') == '1个人报销单')
+    original_index = list(sheets).index(original)
+    relation = next(item for item in relations if item.get('Id') == original.get('{' + relationships + '}id'))
+    sheet_path = posixpath.normpath(posixpath.join('xl', relation.get('Target'))) if not relation.get('Target').startswith('/') else relation.get('Target').lstrip('/')
+    sheet_bytes = parts[sheet_path]
+    names = workbook.find(tag('definedNames'))
+    retained_names = [copy.deepcopy(item) for item in names if item.get('localSheetId') == str(original_index)] if names is not None else []
+    if names is not None:
+        names.clear()
+    sheets.clear()
+    signatures = signatures or DEFAULT_SIGNATURES
+    page_count = math.ceil(len(state['matches']) / 13)
+
+    def set_cell(document, reference, value):
+        cell = document.find('.//' + tag('c') + '[@r="' + reference + '"]')
+        if cell is None:
+            row_number = ''.join(character for character in reference if character.isdigit())
+            row = document.find('.//' + tag('row') + '[@r="' + row_number + '"]')
+            if row is None:
+                raise ValueError('报销模板缺少明细行：' + row_number)
+            cell = ET.SubElement(row, tag('c'), r=reference)
+        for child in list(cell):
+            cell.remove(child)
+        if isinstance(value, (int, float, Decimal)):
+            cell.set('t', 'n')
+            ET.SubElement(cell, tag('v')).text = str(value)
+        else:
+            cell.set('t', 'inlineStr')
+            inline = ET.SubElement(cell, tag('is'))
+            ET.SubElement(inline, tag('t')).text = str(value)
+
+    for page in range(page_count):
+        document = ET.fromstring(sheet_bytes)
+        for row_number in range(9, 22):
+            for column in ['B', 'C', 'G', 'H', 'I']:
+                set_cell(document, column + str(row_number), '')
+            set_cell(document, 'A' + str(row_number), row_number - 8)
+        for index, match in enumerate(state['matches'][page * 13:(page + 1) * 13]):
+            invoice_type = '普票'
+            for invoice in match['invoices']:
+                content = text(invoice)
+                if '专用发票' in content:
+                    invoice_type = '专票'
+                    break
+                if 'invoice' in content.lower():
+                    invoice_type = 'Invoice'
+                    break
+                if '押金' in content and '收据' in content:
+                    invoice_type = '押金收据'
+                    break
+            total = sum(max((abs(Decimal(value.replace(',', ''))) for value in item['ocr']['amounts']), default=Decimal(0)) for item in match['invoices'])
+            for column, value in [('B', invoice_type), ('C', purposes.get(match['id'], '').strip()), ('G', len(match['invoices'])), ('H', total)]:
+                set_cell(document, column + str(index + 9), value)
+        for index, cell in enumerate(['A30', 'D30', 'A31', 'D31', 'A32', 'D32']):
+            value = signatures[index].strip() if index < len(signatures) else ''
+            set_cell(document, cell, value + '：' if value else '')
+        name = '个人报销单' if page_count == 1 else f'个人报销单-{page + 1}'
+        output_path = f'xl/worksheets/native-expense-{page + 1}.xml'
+        relation_id = f'rIdNativeExpense{page + 1}'
+        ET.SubElement(sheets, tag('sheet'), name=name, sheetId=str(page + 1), attrib={'{' + relationships + '}id': relation_id})
+        ET.SubElement(relations, '{' + package + '}Relationship', Id=relation_id, Type=relationships + '/worksheet', Target=output_path[3:])
+        ET.SubElement(content_types, '{' + content_namespace + '}Override', PartName='/' + output_path, ContentType='application/vnd.openxmlformats-officedocument.spreadsheetml.worksheet+xml')
+        parts[output_path] = ET.tostring(document, xml_declaration=True, encoding='UTF-8', standalone=True)
+        rel_path = posixpath.dirname(sheet_path) + '/_rels/' + posixpath.basename(sheet_path) + '.rels'
+        if rel_path in parts:
+            parts['xl/worksheets/_rels/' + posixpath.basename(output_path) + '.rels'] = parts[rel_path]
+        for definition in retained_names:
+            cloned = copy.deepcopy(definition)
+            cloned.set('localSheetId', str(page))
+            cloned.text = (cloned.text or '').replace("'1个人报销单'!", "'" + name + "'!").replace('1个人报销单!', "'" + name + "'!")
+            names.append(cloned)
+    calc = workbook.find(tag('calcPr'))
+    if calc is None:
+        calc = ET.SubElement(workbook, tag('calcPr'))
+    calc.set('fullCalcOnLoad', '1')
+    calc.set('forceFullCalc', '1')
+    for key, document in [('xl/workbook.xml', workbook), ('xl/_rels/workbook.xml.rels', relations), ('[Content_Types].xml', content_types)]:
+        parts[key] = ET.tostring(document, xml_declaration=True, encoding='UTF-8', standalone=True)
+    with zipfile.ZipFile(destination, 'w', zipfile.ZIP_DEFLATED) as output:
+        for name, content in parts.items():
+            output.writestr(name, content)

native-engine/helper.entitlements

@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0"><dict>
+    <key>com.apple.security.app-sandbox</key><true/>
+    <key>com.apple.security.inherit</key><true/>
+</dict></plist>

native-engine/make_icon.py

@@ -0,0 +1,27 @@
+import json
+from pathlib import Path
+from PIL import Image, ImageDraw
+
+root = Path(__file__).resolve().parents[1] / 'reimburse/Assets.xcassets/AppIcon.appiconset'
+image = Image.new('RGBA', (1024, 1024))
+draw = ImageDraw.Draw(image)
+draw.rounded_rectangle((44, 44, 980, 980), radius=218, fill='#145b54')
+draw.rounded_rectangle((72, 70, 952, 955), radius=194, fill='#1b7368')
+draw.rounded_rectangle((284, 174, 743, 800), radius=42, fill='#f4f0e6')
+draw.rounded_rectangle((350, 268, 665, 294), radius=13, fill='#90b2a8')
+draw.rounded_rectangle((350, 341, 580, 363), radius=11, fill='#b1c7bd')
+draw.rounded_rectangle((350, 405, 634, 427), radius=11, fill='#b1c7bd')
+draw.rounded_rectangle((198, 488, 826, 817), radius=70, fill='#103e39')
+draw.rounded_rectangle((218, 511, 806, 796), radius=54, fill='#224e46')
+draw.rounded_rectangle((617, 582, 854, 726), radius=42, fill='#d6bd82')
+draw.ellipse((662, 630, 709, 677), fill='#4c6351')
+draw.line([(342, 645), (411, 707), (523, 584)], fill='#f4f0e6', width=29, joint='curve')
+entries = []
+for size in [16, 32, 128, 256, 512]:
+    for scale in [1, 2]:
+        filename = f'icon-{size}@{scale}x.png'
+        image.resize((size * scale, size * scale), Image.Resampling.LANCZOS).save(root / filename)
+        entries.append(dict(idiom='mac', size=f'{size}x{size}', scale=f'{scale}x', filename=filename))
+content = json.dumps(dict(images=entries, info=dict(author='xcode', version=1)), indent=2)
+import subprocess
+subprocess.run(['apply_patch', '*** Begin Patch\n*** Delete File: ' + str(root / 'Contents.json') + '\n*** Add File: ' + str(root / 'Contents.json') + '\n' + '\n'.join('+' + line for line in content.splitlines()) + '\n*** End Patch'], check=True)

native-engine/requirements.lock.txt

@@ -0,0 +1,25 @@
+altgraph==0.17.5
+et_xmlfile==2.0.0
+flatbuffers==25.12.19
+lxml==6.1.3
+macholib==1.16.4
+numpy==2.5.1
+onnxruntime==1.28.0
+opencv-python==5.0.0.93
+openpyxl==3.1.5
+packaging==26.3
+pillow==11.3.0
+protobuf==7.35.1
+pyclipper==1.4.0
+pyinstaller==6.16.0
+pyinstaller-hooks-contrib==2026.7
+pypdfium2==4.30.0
+python-pptx==1.0.2
+PyYAML==6.0.3
+rapidocr-onnxruntime==1.4.4
+setuptools==84.0.0
+shapely==2.1.2
+six==1.17.0
+tqdm==4.70.0
+typing_extensions==4.16.0
+xlsxwriter==3.2.9

native-engine/requirements.txt

@@ -0,0 +1,11 @@
+-c requirements.lock.txt
+rapidocr-onnxruntime==1.4.4
+numpy==2.5.1
+onnxruntime==1.28.0
+opencv-python==5.0.0.93
+protobuf==7.35.1
+Pillow==11.3.0
+python-pptx==1.0.2
+openpyxl==3.1.5
+pypdfium2==4.30.0
+pyinstaller==6.16.0

native-engine/tests/smoke_engine.py

@@ -0,0 +1,63 @@
+import argparse
+import json
+import subprocess
+import sys
+import shutil
+from pathlib import Path
+from PIL import Image
+
+
+def invoke(command, request):
+    process = subprocess.run(command, input=json.dumps(request, ensure_ascii=False) + '\n', capture_output=True, text=True, timeout=180)
+    events = [json.loads(line) for line in process.stdout.splitlines() if line.strip()]
+    errors = [event for event in events if event['event'] == 'error']
+    if process.returncode or errors:
+        raise RuntimeError(str(errors) + '\n' + process.stderr)
+    results = [event['result'] for event in events if event['event'] == 'result']
+    if len(results) != 1:
+        raise AssertionError('处理引擎没有返回唯一结果')
+    return results[0]
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--binary')
+    parser.add_argument('--output', required=True)
+    arguments = parser.parse_args()
+    root = Path(__file__).resolve().parents[3]
+    output = Path(arguments.output).resolve()
+    output.mkdir(parents=True, exist_ok=True)
+    engine = Path(__file__).resolve().parents[1] / 'engine.py'
+    command = [arguments.binary] if arguments.binary else [sys.executable, str(engine)]
+    state = invoke(command, dict(operation='scan', path=str(root / 'fapiao/demo-materials'), workspacePath=str(output / 'materials'), folderName='演示报账材料'))
+    assert len(state['invoices']) == 2
+    assert len(state['payments']) == 2
+    assert len(state['photos']) == 1
+    assert len(state['matches']) == 2
+    assert {match['category'] for match in state['matches']} == {'餐饮', '住宿'}
+    assert state['directoryPaymentTotal'] == 596.5
+    assert sum(match['expenseAmount'] for match in state['matches']) == 596.5
+    assert not state['warnings']
+    (output / 'state.json').write_text(json.dumps(state, ensure_ascii=False, indent=2))
+    for operation, extension in [('ppt', 'pptx'), ('expense', 'xlsx')]:
+        destination = output / ('演示报账材料.' + extension)
+        invoke(command, dict(operation=operation, state=state, destination=str(destination), classified=True, purposes={match['id']: match['category'] + '费用' for match in state['matches']}, signatures=['经办人', '财务']))
+        assert destination.stat().st_size > 1000
+    pdf_input = output / 'pdf-input'
+    (pdf_input / '发票').mkdir(parents=True, exist_ok=True)
+    (pdf_input / '付款截图').mkdir(parents=True, exist_ok=True)
+    with Image.open(root / 'fapiao/demo-materials/发票/住宿_差旅发票_A001.png') as original:
+        original.convert('RGB').save(pdf_input / '发票/两页发票.pdf', save_all=True, append_images=[Image.new('RGB', (800, 500), 'white')])
+    (pdf_input / '发票/损坏文件.pdf').write_bytes(b'not a pdf')
+    shutil.copyfile(root / 'fapiao/demo-materials/付款截图/酒店支付记录_B009.png', pdf_input / '付款截图/付款.png')
+    pdf_state = invoke(command, dict(operation='scan', path=str(pdf_input), workspacePath=str(output / 'pdf-materials')))
+    assert len(pdf_state['invoices']) == 2
+    assert len(pdf_state['matches']) == 1
+    assert pdf_state['matches'][0]['expenseAmount'] == 468
+    assert sum(item['ocr']['status'] == 'failed' for item in pdf_state['invoices']) == 1
+    invoke(command, dict(operation='ppt', state=pdf_state, destination=str(output / 'PDF首页贴票.pptx')))
+    print('完整链路通过：5 份材料、2 组自动核对、总额 596.50；PPT、个人报销单、PDF 首页与损坏文件降级均通过。')
+
+
+if __name__ == '__main__':
+    main()

native-engine/tests/test_domain.py

@@ -0,0 +1,109 @@
+import copy
+import sys
+import unittest
+from decimal import Decimal
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+from domain import auto_match, category, combinations, enrich, extract, invoice_amount, payment_total
+
+
+def material(identifier, kind, amount, content='', dates=None, merchants=None):
+    ocr = extract(content.splitlines())
+    ocr.update(amounts=[amount] if isinstance(amount, str) else amount, dates=dates or [], merchants=merchants or [])
+    return dict(id=identifier, name=identifier + '.png', path=identifier, previewPath='', type=kind,
+                size=1, matched=False, ocr=ocr)
+
+
+def workspace(invoices, payments):
+    return dict(rootPath='测试', invoices=invoices, payments=payments, photos=[], matches=[], warnings=[])
+
+
+def train(identifier, traveler):
+    return material(identifier, 'invoice', '1058.00', f'铁路电子客票\n乘车人：{traveler}\n北京南 → 上海虹桥\n2026年06月24日 15:00\nG21次', ['2026-06-24'], ['中国铁路'])
+
+
+class ExtractionTests(unittest.TestCase):
+    def test_train(self):
+        data = extract(['铁路电子客票', '乘车人：张三', '北京南 → 上海虹桥', '2026年08月12日 09:30', 'G101次', '票价 ￥553.00'])
+        self.assertEqual(data['travel'], dict(type='train', travelerName='张三', departure='北京南', destination='上海虹桥', departureTime='2026-08-12 09:30', transportNumber='G101'))
+
+    def test_flight(self):
+        travel = extract(['航空电子客票行程单', '旅客姓名：李四', '出发地：广州', '目的地：成都', '2026-08-15 14:20', '航班号 CZ3401'])['travel']
+        self.assertEqual((travel['type'], travel['travelerName'], travel['departure'], travel['destination'], travel['transportNumber']), ('flight', '李四', '广州', '成都', 'CZ3401'))
+
+    def test_invoice_date_is_not_departure(self):
+        travel = extract(['电子发票（铁路电子客票） 北京市税务局', '发票号码：26119110010005751560 开票日期：2026年06月26日', '北京南站 G21 上海虹桥站', '2026年06月24日 15:00开 17车04C号', '2107021990****0234 王天行', '票价：￥1058.00'])['travel']
+        self.assertEqual(travel['travelerName'], '王天行')
+        self.assertEqual(travel['departureTime'], '2026-06-24 15:00')
+
+    def test_invalid_date_does_not_abort_material(self):
+        self.assertEqual(extract(['2026年02月31日 金额 20'])['dates'], [])
+
+    def test_amount_order(self):
+        self.assertEqual(extract(['票价 ￥1058.00', '支付总额 2116'])['amounts'][0], '2116.00')
+
+
+class MatchingTests(unittest.TestCase):
+    def test_original_many_to_many_fixture(self):
+        state = workspace([train('i1', '王天行'), train('i2', '王超文')], [
+            material('p1', 'payment', ['1058.00', '2116.00'], '12306订单 中国铁路 支付总额 2116', ['2026-06-24']),
+            material('p2', 'payment', '2116.00', '12306消费 中国铁路', ['2026-06-22'])])
+        auto_match(state)
+        self.assertEqual(len(state['matches']), 1)
+        match = state['matches'][0]
+        self.assertEqual((len(match['invoices']), len(match['payments']), match['score'], match['category']), (2, 2, 100, '交通'))
+
+    def test_shared_travel_payment_evidence(self):
+        state = workspace([train('i1', '张三'), train('i2', '李四')], [material('p1', 'payment', '2116.00', '12306 支付成功', ['2026-06-24'])])
+        auto_match(state)
+        self.assertEqual(state['matches'][0]['score'], 95)
+
+    def test_unique_amount(self):
+        state = workspace([material('i', 'invoice', '100.00')], [material('p', 'payment', '100.00')])
+        auto_match(state)
+        self.assertEqual(state['matches'][0]['score'], 75)
+
+    def test_ambiguous_stays_unmatched(self):
+        state = workspace([material('i', 'invoice', '100.00', dates=['2026-08-01'])], [material('p1', 'payment', '100.00', dates=['2026-08-01']), material('p2', 'payment', '100.00', dates=['2026-08-01'])])
+        self.assertFalse(auto_match(state)['matches'])
+
+    def test_high_score_keeps_original_tie_rule(self):
+        state = workspace([material('i', 'invoice', '100.00', dates=['2026-08-01'], merchants=['测试公司'])], [material('p1', 'payment', '100.00', dates=['2026-08-01'], merchants=['测试公司']), material('p2', 'payment', '100.00', dates=['2026-08-01'], merchants=['测试公司'])])
+        self.assertEqual(len(auto_match(state)['matches']), 1)
+
+    def test_split_payment(self):
+        state = workspace([material('i', 'invoice', '100.00')], [material('p1', 'payment', '30.00'), material('p2', 'payment', '70.00')])
+        self.assertEqual(len(auto_match(state)['matches'][0]['payments']), 2)
+
+    def test_ambiguous_split_payment(self):
+        state = workspace([material('i', 'invoice', '100.00')], [material('p1', 'payment', '30.00'), material('p2', 'payment', '70.00'), material('p3', 'payment', '40.00'), material('p4', 'payment', '60.00')])
+        self.assertFalse(auto_match(state)['matches'])
+
+    def test_filename_does_not_match(self):
+        state = workspace([material('same', 'invoice', '100.00')], [material('same', 'payment', '90.00')])
+        self.assertFalse(auto_match(state)['matches'])
+
+    def test_decimal_combinations(self):
+        options = combinations([('first', Decimal('.10')), ('second', Decimal('.20'))], Decimal('.30'))
+        self.assertEqual(options, [['first', 'second']])
+
+    def test_original_categories(self):
+        for content, expected in [('海棠酒店 住宿 房费 客房', '住宿'), ('餐厅 美团外卖 咖啡 饮品', '餐饮'), ('办公用品 打印纸 墨盒 文具', '办公用品'), ('京东商城 采购家具设备', '采购'), ('客户招待 商务宴请', '招待')]:
+            self.assertEqual(category([material('i', 'invoice', '100.00', content)]), expected)
+
+    def test_summary_duplicate(self):
+        self.assertEqual(payment_total([material('order', 'payment', '100.00', '订单详情 支付总额 100.00'), material('proof', 'payment', '100.00', '支付成功 实付金额 100.00')]), Decimal('100.00'))
+
+    def test_invoice_total_labeled(self):
+        self.assertEqual(invoice_amount(material('i', 'invoice', ['200.00', '100.00'], '发票金额 100.00')), Decimal('100.00'))
+
+    def test_enrich_preserves_group_identity(self):
+        state = workspace([material('i', 'invoice', '100.00')], [material('p', 'payment', '100.00')])
+        enrich(auto_match(state))
+        self.assertTrue(state['matches'][0]['invoices'][0]['matched'])
+        self.assertEqual(state['matches'][0]['expenseAmount'], 100)
+
+
+if __name__ == '__main__':
+    unittest.main()

native-engine/tests/test_exports.py

@@ -0,0 +1,95 @@
+import copy
+import tempfile
+import unittest
+import zipfile
+from pathlib import Path
+
+from test_domain import material, workspace
+from domain import auto_match, enrich
+from exports import export_expense, export_ppt, export_travel
+from PIL import Image
+from openpyxl import load_workbook
+from pptx import Presentation
+
+
+class ExportTests(unittest.TestCase):
+    def setUp(self):
+        self.temporary = tempfile.TemporaryDirectory()
+        self.root = Path(self.temporary.name)
+        image = self.root / 'receipt.png'
+        Image.new('RGB', (400, 200), 'white').save(image)
+        self.state = enrich(auto_match(workspace([material('i', 'invoice', '100.00')], [material('p', 'payment', '100.00')])))
+        for item in self.state['invoices'] + self.state['payments']:
+            item['previewPath'] = str(image)
+        self.template = Path(__file__).parents[1] / 'personal-expense-template.xlsx'
+
+    def tearDown(self):
+        self.temporary.cleanup()
+
+    def test_simple_ppt_with_photo_placeholder(self):
+        destination = self.root / 'test.pptx'
+        export_ppt(self.state, destination, True)
+        deck = Presentation(destination)
+        self.assertEqual(len(deck.slides), 3)
+        self.assertEqual(deck.slide_width, 1280 * 12700)
+        self.assertTrue(any('实物照片' in shape.text for shape in deck.slides[2].shapes if shape.has_text_frame))
+
+    def test_travel_category_excludes_photo(self):
+        self.state['matches'][0]['category'] = '交通'
+        destination = self.root / 'test.pptx'
+        export_ppt(self.state, destination, False)
+        self.assertEqual(len(Presentation(destination).slides), 2)
+
+    def test_complex_payment_pagination(self):
+        self.state['matches'][0]['payments'] *= 5
+        destination = self.root / 'test.pptx'
+        export_ppt(self.state, destination, True)
+        self.assertEqual(len(Presentation(destination).slides), 5)
+
+    def test_expense_template_values_and_formulas(self):
+        destination = self.root / 'expense.xlsx'
+        match = self.state['matches'][0]
+        export_expense(self.state, destination, self.template, {match['id']: '=不是公式'}, ['经办人'])
+        workbook = load_workbook(destination)
+        self.assertEqual(workbook.sheetnames, ['个人报销单'])
+        sheet = workbook.active
+        self.assertEqual(sheet['C9'].value, '=不是公式')
+        self.assertEqual(sheet['C9'].data_type, 's')
+        self.assertEqual(sheet['H9'].value, 100)
+        self.assertEqual(sheet['G9'].value, 1)
+        self.assertEqual(sheet['A30'].value, '经办人：')
+        self.assertEqual(sheet['D30'].value, '')
+        original = load_workbook(self.template)['1个人报销单']
+        formulas = {cell.coordinate: cell.value for row in original for cell in row if cell.data_type == 'f'}
+        self.assertTrue(formulas)
+        for coordinate, value in formulas.items():
+            self.assertEqual(sheet[coordinate].value, value)
+        self.assertEqual(str(sheet.print_area).split('!')[-1], str(original.print_area).split('!')[-1])
+
+    def test_expense_pagination(self):
+        self.state['matches'] = [copy.deepcopy(self.state['matches'][0]) for _ in range(14)]
+        destination = self.root / 'expense.xlsx'
+        export_expense(self.state, destination, self.template, {}, [])
+        workbook = load_workbook(destination)
+        self.assertEqual(workbook.sheetnames, ['个人报销单-1', '个人报销单-2'])
+        self.assertEqual(workbook.worksheets[1]['H9'].value, 100)
+        self.assertEqual(workbook.worksheets[1]['H10'].value, '')
+        self.assertEqual(workbook.worksheets[1]['A30'].value, '部门长：')
+
+    def test_travel_only_verified_invoices(self):
+        travel = self.state['invoices'][0]['ocr']['travel']
+        travel.update(type='train', travelerName='张三', departure='北京南', destination='上海虹桥', departureTime='2026-06-24 15:00', transportNumber='G21')
+        destination = self.root / 'travel.xlsx'
+        export_travel(self.state, destination)
+        sheet = load_workbook(destination).active
+        self.assertEqual(sheet.max_row, 2)
+        self.assertEqual(sheet['C2'].value, '张三')
+        self.assertEqual(sheet['G2'].value, 'G21')
+
+    def test_no_travel_raises(self):
+        with self.assertRaises(ValueError):
+            export_travel(self.state, self.root / 'travel.xlsx')
+
+
+if __name__ == '__main__':
+    unittest.main()