sansyn-pc-extension/server/analysis.py

import json, re
from collections import Counter, defaultdict
from ai_client import chat_json, configured
from prompt_templates import build_system_prompt, build_task_prompt

ENGINE_VERSION='9.3.0'

ROLE_CONTRACT_VERSION='1.1'
ROLE_CONTRACT={
  'productStrategy': {'responsibility':'判断商品卖什么、为什么值得参考、下一款继承什么','outputs':['商品定位','核心产品资产','核心卖点结构','目标使用场景','值得继承的产品逻辑','需要调整的方向','商品与同类差异点','下一款策略假设']},
  'conversion': {'responsibility':'判断用户为什么点击、继续看并愿意购买','outputs':['标题表达','主图任务分工','主图核心信息','详情页信息链','卖点承接','信任证明','页面信息缺口','标题/主图/详情优化动作']},
  'consumerInsight': {'responsibility':'从评论和买家秀提炼真实使用体验并转成产品需求','outputs':['稳定正向体验','体验差异','高频顾虑','购前关注','售后问题','消费者语言','真实使用场景','买家秀验证','下一款必须解决的用户需求']},
  'productSupply': {'responsibility':'判断产品怎么改、哪些能落地、哪些必须验证','outputs':['材质','工艺','支数/克重','尺寸','套件','SKU结构','功能参数','规格适配','产品调整项','生产风险','打样验证项','首批验证项','不可直接下结论的产品假设']},
  'visualDesign': {'responsibility':'拆解视觉资产并转成下一款设计和生图方向','outputs':['风格','颜色','花型','元素','材质视觉','构图','拍摄方式','场景','光线','画面氛围','视觉锤','值得继承的视觉资产','下一款视觉变化方向','AI生图参考要求']},
  'executiveDecision': {'responsibility':'把专业判断压缩成老板可直接决策的结论','outputs':['为什么值得做','为什么能卖','最强资产','最大机会','最需要关注的问题','下一款第一优先级','建议保留什么','建议升级什么','探索方向','执行优先顺序']},
  'editorReview': {'responsibility':'对外发布前统一审核、压缩、纠错和去重','outputs':['事实一致性','证据支持','经营数据边界','产品对象与卖点匹配','图片分类','章节一致性','重复删除','措辞统一','字数压缩','最终结构']}
}

SYS='''你是“三笙电商商品开品分析引擎”，面向两类读者：老板/决策者与专业商品开发用户。你的工作不是展示AI分析过程，而是把真实商品证据压缩成可决策、可复用、可开品的专业报告。

角色协作规则（内容生产链）：
1. 商品策略、电商转化、消费者洞察、产品/供应链、视觉设计分别只判断自己的职责范围；可并行，但不得互相越权。
2. 商业决策/老板角色只综合前序角色的证据，不重新发明事实；输出为什么值得做、为什么能卖、第一优先级和执行顺序。
3. 总编审核角色是发布前质量门，不是新的观点来源；必须删除重复、纠正错配、标记假设并保留证据边界。
4. 每个角色的结论、风险、动作和假设都必须能回溯到 ACTUAL_EVIDENCE；没有证据时标记 hypothesis 或 unknown。

最终阅读逻辑：
1. 老板先看：为什么能卖、哪些资产值得继承、下一款优先做什么。
2. 专业用户再看：标题怎么表达、主图在卖什么、详情证明什么、SKU怎么选、消费者实际感受到什么。
3. 图片是主信息载体。用户必须能从图片旁的短标签直接看懂商品信息，不依赖长段解释。

硬规则：
1. 只使用 ACTUAL_EVIDENCE 和程序统计；缺失即未采集，不用常识补齐。
2. 禁止把内部推理标签输出给用户：不得出现[数据事实][分析判断][设计建议][证据不足]。
3. 禁止审核式好坏评分；优先使用“核心判断、值得继承、建议强化、专业动作、下一款方向”。
4. 同一结论默认只出现一次。不要在专业意见、卡片说明、建议中重复同一句信息。
5. 图片信息提炼优先于图片解释：从图片可核验文字/商品/场景中提炼 visualSignals（2-4个短标签），再给1条 nextAction。
6. 主图默认每张只输出：imageRole + visualSignals + nextAction。businessMeaning/evidenceDetail仅供内部或折叠层使用。
7. 详情页按4-6个信息任务组输出，每组：代表图 + visualSignals + nextAction；不得逐张写长文。
8. SKU图片数量不等于SKU数量。SKU重点回答选择维度与顺序；先解决适配，再解决审美。
9. 评论先聚合成稳定体验/体验差异/购前关注，再进入新品方案；买家秀只是证据。
10. 新品方案分来源等级：证据驱动优化 / 反馈驱动升级 / 探索性方向。探索性方向必须明确标记，不得伪装成已有事实。
11. 只要有商品基础证据且主图或详情任一存在，至少输出1套新品方案；证据较完整可输出2-3套。
12. 图片原图完整展示优先于铺满容器；报告渲染必须使用 contain，不以裁切换取整齐。
13. 默认可见层不使用小卡片堆叠思维：每屏只突出1个主结论、1-2张大图或一组大规格图片。
14. 图片上的文字/卖点是核心证据，visualSignals必须优先提取具体词，如规格、材质、权益、功能，而不是抽象评价。
15. 不输出私有思维过程。严格输出 JSON，不要 Markdown 或 HTML。
16. 没有真实内容的可选字段直接省略，不得输出空字符串、空数组、空对象或“暂无/未采集/待补充”等占位文案。'''

STYLE_RULES='''V9.3.0 输出压缩与可见性规则：
- reportSummary.verdict / professionalOpinion：<=55字。
- 老板速览 headline：<=28字；note：<=28字。
- 标题 currentExpression/reinforce：短词组，不写完整分析段落。
- 单张主图 imageRole<=10字；visualSignals 2-4项，每项<=12字；nextAction<=22字。
- 详情组 type<=8字；visualSignals 2-4项，每项<=12字；nextAction<=22字。
- SKU professionalOpinion<=45字；recommendedAction<=30字。
- 评论 insight<=35字；默认不再附第二段 implication。
- 新品方案 whyThisPlan<=45字；productAction<=35字；pageAction<=35字；validation<=25字。
- 禁止通过Renderer用省略号硬截断来制造短文；模型必须直接生成短文本。
- 详细构图/OCR/颜色/材质推理放 evidenceDetail，默认报告不展示。
- 同一事实不在同一章节重复三次。
- 没有真实内容的字段直接省略；禁止为了补结构输出“暂无、未采集、待补充、无”。
- 空字符串、空数组和空对象不得进入用户可见结论。数值0和明确的采集状态属于有效内容，可保留。
- evidenceIds 只用于追溯；用户可见层应由渲染器显示“来源类型+真实内容摘要”，不得让用户只看到字符编号。
'''

FINAL_REPORT_PROMPT='''任务G｜V9.3.0 七角色总编报告编排器。你只能压缩和映射前序分析，不重新发明事实。

输入中的 roleOutputs 来自七个职责角色。编排要求：商品策略进入“为什么能卖/值得继承”；电商转化进入标题、主图、详情；消费者洞察进入真实体验；产品/供应链进入SKU、参数与验证；视觉设计进入视觉资产与下一款视觉变化；老板决策压缩为首屏；总编审核负责删除角色间重复和冲突。角色输出不是额外证据，所有结论仍必须保留 evidenceIds。

最终顺序：老板速览 → 下一款方向 → 标题 → 主图 → 详情 → SKU → 消费者反馈。商品价值不要单独重复成一整章，吸收到老板速览。

严格输出 JSON：
{
  "reportSummary":{"title":"","verdict":"<=45字","coverImageId":"IMG_MAIN_0001"},
  "ownerOverview":{"chapterTitle":"老板速览","cards":[
    {"label":"为什么能卖","headline":"<=18字","note":"<=22字","evidenceIds":[]},
    {"label":"值得继承","headline":"<=18字","note":"<=22字","evidenceIds":[]},
    {"label":"下一款优先","headline":"<=18字","note":"<=22字","evidenceIds":[]}
  ]},
  "titleAnalysis":{"professionalOpinion":"<=45字","originalTitle":"","currentExpression":[{"value":"<=10字","evidenceIds":[]}],"reinforce":[{"topic":"<=10字","evidenceIds":[]}],"recommendedActions":[{"action":"<=22字","evidenceIds":[]}],"recommendedTitles":[]},
  "visualCommerce":{"professionalOpinion":"<=45字","items":[{"imageEvidenceId":"IMG_MAIN_0001","imageRole":"<=8字","visualSignals":["<=10字"],"businessMeaning":"<=22字","nextAction":"<=18字","evidenceIds":[]}],"evidenceDetail":[]},
  "detailCommerce":{"professionalOpinion":"<=45字","contentGroups":[{"type":"品牌服务|套件组成|材质品质|产品功能|使用便利|花型选择|用户验证|其他","representativeImageId":"IMG_DETAIL_0001","visualSignals":["<=10字"],"businessMeaning":"<=22字","nextAction":"<=18字","imageEvidenceIds":[],"evidenceIds":[]}],"pageSequence":[]},
  "skuAnalysis":{"professionalOpinion":"<=35字","selectionDimensions":[{"name":"<=8字","values":[],"evidenceIds":[]}],"recommendedOrder":[],"representativeImageIds":[],"recommendedAction":"<=24字","structuredStatus":"confirmed|missing","structuredCount":null,"imageCount":0},
  "customerExperience":{"professionalOpinion":"<=45字","consumerInsights":{"experienceSignals":[{"type":"稳定体验|体验差异|购前关注|对新品的启示","insight":"<=28字","evidenceIds":[]}],"productImplications":[]},"buyerEvidence":{}},
  "expressionContinuity":{"coreClaims":[{"claim":"<=24字","mainStatus":"expressed|partial|missing","detailStatus":"proven|partial|missing","reviewStatus":"validated|partial|missing","nextAction":"<=24字","evidenceIds":[]}]},
  "newProductPlans":{"professionalOpinion":"<=45字","plans":[{"sourceType":"证据驱动优化|反馈驱动升级|探索性方向","type":"现有基础优化|强化体验区分|强化设计选择|其他","name":"<=18字","whyThisPlan":"<=32字","visualReferenceImageIds":[],"productAction":"<=28字","pageAction":"<=28字","validation":"<=20字","evidenceIds":[]}]},
  "productExperience":{"parameterFacts":[],"gates":[]},
  "validationLoop":{"rows":[]}
}

硬规则：
- ownerOverview.cards 必须正好3项，且分别回答“为什么能卖 / 值得继承 / 下一款优先”。
- visualCommerce.items 中每张图必须优先给 visualSignals；不要写长 message。
- detailCommerce.contentGroups 证据允许时输出4-6组。
- newProductPlans.plans>=1；第三类创新若不是直接证据结论必须标 sourceType=探索性方向。
- productExperience.parameterFacts>=2、productExperience.gates>=2、validationLoop.rows>=3，继续用于工程证据校验，但默认报告不展示。
- 禁止用户可见字段出现[数据事实][分析判断][设计建议][证据不足]。
- 只返回有真实内容的可选字段；不得用空字符串、空数组、空对象或“暂无/未采集/待补充”填充版面。
- 七角色输出必须能归入“核心判断—关键发现—执行动作—风险/验证”；证据编号仅作机器追溯，不作为面向用户的说明文字。
'''

FINAL_REPORT_REPAIR_PROMPT='''任务G-R｜修复 V9.3.0 最终报告 JSON。只修复结构缺失和证据引用，不扩写正文。
必须保留短文本与图片信息优先规则。
强制最小结构：reportSummary.title/verdict非空；ownerOverview.cards>=3；productExperience.parameterFacts>=2；productExperience.gates>=2；newProductPlans.plans>=1；validationLoop.rows>=3。
如果 customerExperience 有评论证据，优先保留 experienceSignals；无评论时允许为空。
所有 evidenceIds 只能来自 ACTUAL_EVIDENCE。不要输出 Markdown、解释或包装层。
可选字段没有真实内容时直接省略，不得填入“暂无、未采集、待补充”或空容器。
''' 

TOPICS={
 '颜值/配色':['好看','漂亮','颜色','色彩','高级','颜值','款式','花色','图案','搭配'],
 '面料/触感':['柔软','舒服','舒适','亲肤','手感','面料','纯棉','全棉','透气','粗糙','刮'],
 '厚薄/季节':['薄','厚','厚实','轻薄','冬天','夏天','季节'],
 '色差':['色差','偏黄','偏白','偏暗','颜色不一样','实物和图片'],
 '做工/破损':['做工','线头','走线','缝','破','开线','油污','质量'],
 '尺寸/适配':['尺寸','大小','合适','床笠','床单','被套','罩不住','紧张'],
 '起球/掉毛':['起球','掉毛','粘毛','浮毛'],
 '缩水/掉色':['缩水','掉色','褪色','变形'],
 '气味':['味道','气味','异味'],
 '物流/服务':['物流','发货','客服','包装','快递','售后'],
 '价格/价值感':['划算','性价比','价格','便宜','值','贵']
 ,'尺码/上身':['尺码','码数','偏大','偏小','合身','显瘦','上身','版型']
 ,'口味/复购':['口味','味道','好吃','难吃','甜','咸','回购','复购']
 ,'成分/肤感':['成分','肤感','吸收','刺激','过敏','敏感','闷痘']
 ,'配置/性能':['配置','兼容','性能','速度','续航','发热','卡顿','连接']
}

def _txt(v): return '' if v is None else str(v)
def _arr(v): return v if isinstance(v,list) else []
def _uniq(xs):
    out=[]; seen=set()
    for x in xs or []:
        k=_txt(x).strip()
        if k and k not in seen: seen.add(k); out.append(x)
    return out

def _question_like(c):
    c=_txt(c).strip()
    return bool(re.search(r'问大家|更多回答|查看全部问答|大家问|^问[：:\s]|请问|会不会|有没有|多少价|(?:起球|缩水|掉色|尺寸).*[？?]',c))

BAD_TITLES = ('用户评价', '累计评价', '商品评价', '问大家', '参数信息', '规格参数', '图文详情', '商品详情')
BAD_PARAM_CODE = re.compile(r'(?:^\d+$|window\.|console\.|function\b|frontend_data|performance\.|[{}\[\]"\'=;<>])', re.I)
NEGATIVE_REVIEW_KEYS = ('不够', '一般', '不会复购', '偏小', '偏大', '缩水', '掉色', '起球', '掉毛', '异味', '破损', '开线', '色差', '不合身', '难吃', '变质', '漏液', '过敏', '刺激', '不兼容', '卡顿', '发热', '续航差')
IMAGE_GROUPS = ('main', 'detail', 'sku', 'buyerShow')
BAD_IMAGE_URL = re.compile(r'(?:avatar|sns_logo|credit|logo|icon|sprite|shopmanager|/tps/|/tfs/TB1[^/]+-\d{1,4}-\d{1,4}\.png|tps-\d{1,3}-\d{1,3}|defaultRate)', re.I)
BAD_PAGE_IMAGE_URL = re.compile(r'(?:[-_]rate(?:\.|_|-)|rate_livephoto|tbbala|/tps/|/tfs/TB1[^/]+-\d{1,4}-\d{1,4}\.png)', re.I)

def _clean_product_title(v):
    value=re.sub(r'\s+',' ',_txt(v)).strip()
    suffix=r'(?:[-_｜|·\s]*)?(?:(?:www\.)?(?:tmall|taobao)\.com)?(?:天猫(?:网)?|淘宝网?)'
    previous=None
    while value and value!=previous:
        previous=value
        value=re.sub(suffix+r'\s*$','',value,flags=re.I).rstrip('-_｜|· ')
    return value

def _valid_title(v):
    value=_clean_product_title(v)
    return value if 4<=len(value)<=260 and not any(x in value for x in BAD_TITLES) and not re.search(r'登录|验证码|访问受限',value) else ''

def _valid_param(name,value):
    name=re.sub(r'\s+',' ',_txt(name)).strip(); value=re.sub(r'\s+',' ',_txt(value)).strip()
    if not name or not value or len(name)>60 or len(value)>220:return False
    if BAD_PARAM_CODE.search(name) or re.search(r'(?:console\.log|window\.|performance\.now|frontend_data)',value,re.I):return False
    return bool(re.search(r'[\u4e00-\u9fffA-Za-z]',name))

def _image_url(value):
    if isinstance(value,dict):
        value=next((_txt(value.get(k)) for k in ('url','picUrl','imageUrl','src','bigUrl','thumbnail') if value.get(k)), '')
    value=_txt(value).strip().replace('&amp;','&')
    if value.startswith('//'):value='https:'+value
    return value if value.startswith(('http://','https://','data:image/')) and not BAD_IMAGE_URL.search(value) else ''

def _clean_image_urls(values):
    return _uniq([url for url in (_image_url(x) for x in values or []) if url])

def _normalize_image_groups(raw, reviews):
    source=raw.get('images',{}) if isinstance(raw.get('images'),dict) else {}
    meta=raw.get('meta',{}) if isinstance(raw.get('meta'),dict) else {}
    strict=(meta.get('imageClassificationVersion')=='strict-v1' and isinstance(source.get('provenance'),list))
    review_links={};legacy_review_links={}
    for review_index,review in enumerate(reviews,1):
        cleaned=_clean_image_urls(review.get('images',[]));review['images']=cleaned
        for url in cleaned:
            review_links.setdefault(url,[]).append(review_index)
            if review.get('source')=='dom_visible':legacy_review_links.setdefault(url,[]).append(review_index)
    incoming=[]
    for record in source.get('provenance',[]) or []:
        if not isinstance(record,dict):continue
        url=_image_url(record.get('url'));group=record.get('group');origin=_txt(record.get('source')).strip()
        if url and group in IMAGE_GROUPS and origin:incoming.append({'url':url,'group':group,'source':origin})
    groups={key:[] for key in IMAGE_GROUPS};provenance=[];assigned=set()
    buyer_candidates=[]
    if strict:
        buyer_candidates=[x['url'] for x in incoming if x['group']=='buyerShow' and x['source']=='review_record' and x['url'] in review_links]
    else:
        buyer_candidates=list(legacy_review_links)
    for url in _uniq(buyer_candidates):
        if url in assigned:continue
        assigned.add(url);groups['buyerShow'].append(url)
        link_index=review_links if strict else legacy_review_links
        for review_index in link_index.get(url,[]):provenance.append({'url':url,'group':'buyerShow','source':'review_record','reviewIndex':review_index,'reviewEvidenceId':f'REV_{review_index:06d}'})
    if strict:
        allowed_sources={'sku':{'dom_sku_selector','structured_sku_selector'},'main':{'meta_og_image','dom_main_gallery','structured_main_gallery'},'detail':{'dom_product_description','structured_product_description'}}
        for group in ('sku','main','detail'):
            for record in incoming:
                if record['group']!=group or record['source'] not in allowed_sources[group] or record['url'] in assigned:continue
                if group in ('main','detail','sku') and BAD_PAGE_IMAGE_URL.search(record['url']):continue
                assigned.add(record['url']);groups[group].append(record['url']);provenance.append(record)
    else:
        # V9.2.3 legacy/standalone compatibility:
        # Older and standalone collectors already separate main/detail/sku arrays but do not emit strict-v1 provenance.
        # Do NOT discard these verified-by-collector groups merely because provenance metadata is absent.
        legacy_sources={'main':'legacy_group_main','detail':'legacy_group_detail','sku':'legacy_group_sku'}
        for group in ('main','detail','sku'):
            for raw_url in source.get(group,[]) or []:
                url=_image_url(raw_url)
                if not url or url in assigned or BAD_PAGE_IMAGE_URL.search(url):continue
                assigned.add(url);groups[group].append(url)
                provenance.append({'url':url,'group':group,'source':legacy_sources[group]})
    source_values=[]
    for key in IMAGE_GROUPS:
        if isinstance(source.get(key),list):source_values.extend(source.get(key))
    original_count=len(set(_clean_image_urls(source_values)))
    all_images=_uniq(sum((groups[k] for k in IMAGE_GROUPS),[]))
    policy='strict-v1' if strict else 'legacy-review-only'
    previous_excluded=(source.get('classification',{}) or {}).get('unverifiedExcludedCount',0)
    try:previous_excluded=max(0,int(previous_excluded))
    except:previous_excluded=0
    return {**groups,'all':all_images,'provenance':provenance,'classification':{'version':policy,'classifiedCount':len(all_images),'unverifiedExcludedCount':max(previous_excluded,original_count-len(all_images),0),'groupsAreDisjoint':True}}

def _normalize_questions(values):
    out=[];seen=set()
    def add(q,a='',source='unknown'):
        q=re.sub(r'^(?:问大家|大家问)\s*[·\d]*\s*','',_txt(q)).strip(' ，,。')
        q=re.sub(r'\s*(?:更多回答|查看全部问答)\s*$','',q).strip(' ，,。')
        if len(q)<3:return
        key=re.sub(r'[\s，,。！？?]','',q)
        if key in seen:return
        seen.add(key);out.append({'question':q[:1800],'answer':_txt(a).strip()[:1800],'source':source})
    for x in values or []:
        if isinstance(x,str):x={'question':x}
        if not isinstance(x,dict):continue
        q=_txt(x.get('question') or x.get('title') or x.get('ask') or x.get('content')).strip()
        a=_txt(x.get('answer') or x.get('reply')).strip(); source=x.get('source') or 'questions_input'
        chunks=re.findall(r'(?:^|\s)问[：:\s]+(.+?)(?=\s*更多回答|\s*查看全部问答|\s+问[：:\s]|$)',q)
        if chunks:
            for chunk in chunks:add(chunk,'',source)
        else:add(q,a,source)
    return out

def _normalize_question(x, source='unknown'):
    if isinstance(x,str): return {'question':x.strip(),'answer':'','source':source}
    if not isinstance(x,dict): return None
    q=_txt(x.get('question') or x.get('title') or x.get('ask') or x.get('content')).strip()
    a=_txt(x.get('answer') or x.get('reply')).strip()
    if len(q)<3:return None
    return {'question':q[:1800],'answer':a[:1800],'source':x.get('source') or source}

def normalize(raw):
    r=dict(raw or {})
    reviews=[]; questions=_normalize_questions(r.get('questions',[]) or []); seen=set(); qseen={re.sub(r'\s+','',x['question'])[:260] for x in questions}
    for x in r.get('reviews',[]) or []:
        if isinstance(x,str): x={'content':x}
        c=_txt(x.get('content') or x.get('text')).strip(); append=_txt(x.get('appendContent') or x.get('append')).strip()
        if re.fullmatch(r'\d{4}年\d{1,2}月\d{1,2}日已购[：:].+',c):continue
        if re.search(r'该用户(?:未填写评价内容|没有填写评价|觉得商品非常好)',c):
            if append:c=append;append=''
            else:continue
        if len(c)<3 or c in ('此用户没有填写评价','系统默认好评'): continue
        if _question_like(c):
            qq=_normalize_question({'question':c,'source':x.get('source') or 'misclassified_review'},'misclassified_review')
            if qq:
                k=re.sub(r'\s+','',qq['question'])[:260]
                if k not in qseen:qseen.add(k);questions.append(qq)
            continue
        key=re.sub(r'\s+','',c)[:260]
        if key in seen: continue
        seen.add(key)
        reviews.append({'content':c[:2200],'appendContent':append[:1200],
                        'rating':x.get('rating') or x.get('score') or x.get('star'),'sku':_txt(x.get('sku') or x.get('skuInfo')).strip()[:260],
                        'date':_txt(x.get('date') or x.get('time')).strip()[:100],'images':_clean_image_urls(x.get('images',[]) or [])[:24],'likes':x.get('likes'),'source':x.get('source') or ''})
    r['reviews']=reviews; r['questions']=questions[:5000]
    attrs=[];attr_seen=set()
    for x in r.get('attributes',[]) or []:
        if not isinstance(x,dict):continue
        name=x.get('name');value=x.get('value')
        if not _valid_param(name,value):continue
        key=(re.sub(r'\s+','',_txt(name)),re.sub(r'\s+','',_txt(value)))
        if key in attr_seen:continue
        attr_seen.add(key);attrs.append({'name':_txt(name).strip(),'value':_txt(value).strip()})
    r['attributes']=attrs[:500]
    product=dict(r.get('product') or {})
    product['title']=_valid_title(product.get('title'))
    r['product']=product
    r['images']=_normalize_image_groups(r,reviews)
    return r

def _attribute_rows(raw):
    return [x for x in raw.get('attributes',[]) if isinstance(x,dict) and _valid_param(x.get('name'),x.get('value'))]

def _find_attr(attrs, words):
    for i,x in enumerate(attrs,1):
        name=_txt(x.get('name'))
        if any(word in name for word in words):return x,f'ATTR_{i:04d}'
    return None,''

CATEGORY_TEMPLATES={
  'apparel':{
    'label':'服饰','keywords':('服饰','女装','男装','童装','连衣裙','衬衫','裤子','外套','内衣','鞋','尺码','版型'),
    'dimensions':(
      ('尺码与适配',('尺码','尺寸','适用身高','适用体重','码数')),
      ('版型与轮廓',('版型','衣长','裤型','领型','袖长','腰型')),
      ('面料与成分',('面料','材质','成分','里料','含量')),
      ('做工与细节',('工艺','做工','缝制','拉链','纽扣')),
      ('颜色与款式',('颜色','色号','款式','图案','风格')),
      ('洗护与标准',('洗护','洗涤','执行标准','安全类别','等级')),
    )},
  'food':{
    'label':'食品','keywords':('食品','零食','饮料','咖啡','茶','牛奶','饼干','坚果','冲饮','口味','配料','保质期','净含量'),
    'dimensions':(
      ('规格与净含量',('规格','净含量','重量','数量','包装规格')),
      ('口味与食用体验',('口味','风味','甜度','食用方法')),
      ('配料与营养',('配料','成分','营养','能量','蛋白质')),
      ('保质期与储存',('保质期','生产日期','储存','保存')),
      ('产地与合规',('产地','生产许可证','执行标准','厂名','厂址')),
      ('包装与运输',('包装','包装方式','运输','破损')),
    )},
  'beauty':{
    'label':'美妆','keywords':('美妆','护肤','彩妆','面霜','精华','乳液','面膜','口红','粉底','防晒','肤质','功效'),
    'dimensions':(
      ('功效宣称',('功效','功能','效果','宣称')),
      ('适用肤质与人群',('肤质','适用人群','适用部位','肌肤')),
      ('成分与配方',('成分','配方','主要成分','净含量')),
      ('使用方式',('使用方法','使用步骤','用法','频次')),
      ('敏感与风险',('敏感','注意事项','禁忌','过敏')),
      ('备案与保质期',('备案','批准文号','执行标准','保质期','生产日期')),
    )},
  'digital':{
    'label':'数码','keywords':('数码','手机','电脑','平板','耳机','相机','键盘','鼠标','充电器','处理器','内存','存储','兼容'),
    'dimensions':(
      ('核心配置',('型号','处理器','内存','存储','容量','功率','分辨率')),
      ('兼容性',('兼容','适用型号','系统','接口','协议')),
      ('性能与续航',('性能','续航','电池','速度','刷新率','传输')),
      ('版本与规格',('版本','配置','颜色','尺寸','重量')),
      ('使用问题',('使用说明','连接','安装','故障')),
      ('质保与售后',('质保','保修','售后','认证','执行标准')),
    )},
  'home_textile':{
    'label':'家纺','keywords':('家纺','床品','四件套','三件套','被套','床单','床笠','枕套','被芯','枕芯','毛毯'),
    'dimensions':(
      ('结构与规格',('件数','套件组成','尺寸','规格','床型')),
      ('材质与触感',('面料','材质','成分','支数','密度','克重')),
      ('颜色与花型',('颜色','花型','图案','风格')),
      ('工艺与做工',('工艺','织造','印染','缝制')),
      ('洗后表现',('洗护','洗涤','缩水','掉色','起球')),
      ('场景适配',('适用场景','适用人群','季节')),
    )},
}

def detect_category(raw):
    product=raw.get('product') or {}; attrs=_attribute_rows(raw)
    corpus=' '.join([_txt(product.get('category')),_txt(product.get('title'))]+[f"{x.get('name')} {x.get('value')}" for x in attrs]+[_txt(x.get('group'))+' '+_txt(x.get('name')) for x in raw.get('sku',[]) if isinstance(x,dict)])
    scores={key:sum(1 for word in spec['keywords'] if word and word in corpus) for key,spec in CATEGORY_TEMPLATES.items()}
    key=max(scores,key=scores.get) if scores and max(scores.values()) else 'generic'
    if key=='generic':
        return {'key':'generic','label':_txt(product.get('category')).strip() or '通用商品','confidence':0.35,'dimensions':()}
    total=max(1,sum(scores.values()))
    spec=CATEGORY_TEMPLATES[key]
    return {'key':key,'label':spec['label'],'confidence':round(min(0.98,0.55+scores[key]/total*0.4),2),'dimensions':spec['dimensions']}

def _category_parameter_facts(raw, limit=12):
    attrs=_attribute_rows(raw); route=detect_category(raw); used=set(); facts=[]
    dimensions=route.get('dimensions') or ()
    if not dimensions:
        dimensions=tuple((_txt(x.get('name')).strip(),(_txt(x.get('name')).strip(),)) for x in attrs[:8])
    for label,aliases in dimensions:
        matches=[];refs=[]
        for i,item in enumerate(attrs,1):
            if i in used:continue
            name=_txt(item.get('name')).strip()
            if any(alias and alias in name for alias in aliases):
                used.add(i);matches.append(item);refs.append(f'ATTR_{i:04d}')
        if not matches:continue
        value='；'.join(f"{x.get('name')}：{x.get('value')}" for x in matches)
        facts.append({'label':label,'value':concise_fact(value,240),'productMeaning':'当前类目需要核对的页面产品定义','evidenceIds':refs})
        if len(facts)>=limit:return facts
    for i,item in enumerate(attrs,1):
        if i in used:continue
        facts.append({'label':_txt(item.get('name')),'value':concise_fact(item.get('value'),180),'productMeaning':'页面采集参数，开发新品前与实物或资料核对','evidenceIds':[f'ATTR_{i:04d}']})
        if len(facts)>=limit:break
    return facts

def _category_gates(raw, issue=None):
    issue=issue or _review_issue(raw); route=detect_category(raw); attrs=_attribute_rows(raw); gates=[]
    if issue:
        gates.append({'priority':'P0','label':f"评论问题：{issue['label']}",'action':issue['action'],'businessReason':concise_fact(issue['text'],72),'evidenceIds':[issue['evidenceId']]})
    for fact in _category_parameter_facts(raw,6):
        gates.append({'priority':'P0' if len(gates)<3 else 'P1','label':f"核对{fact['label']}",'action':f"核对样品、SKU、参数和详情页中的{fact['label']}是否一致",'businessReason':f"{route['label']}商品的关键选择信息必须可追溯",'evidenceIds':fact['evidenceIds']})
    return gates[:7]

def _category_validation_rows(raw):
    route=detect_category(raw); facts=_category_parameter_facts(raw,8); refs=_uniq(sum((_arr(x.get('evidenceIds')) for x in facts),[])); issue=_review_issue(raw)
    issue_refs=[issue['evidenceId']] if issue else []
    return [
      {'stage':'商品定义','owner':'商品企划 / 商品运营','hypothesis':'','metrics':f"锁定{route['label']}商品的关键参数、SKU和目标场景",'successMeaning':'商品定义与已采参数逐项对应','nextAction':'通过后发起打样；不一致则退回定义','evidenceIds':refs},
      {'stage':'样品与资料','owner':'供应链 / 品控','hypothesis':'','metrics':'核对实物、标签、包装、合规资料和页面宣称','successMeaning':'页面信息与真实商品一致','nextAction':'通过后锁版；不一致则暂停对应宣称','evidenceIds':refs},
      {'stage':'场景验证','owner':'商品企划 / 品控','hypothesis':'','metrics':'按当前类目关键使用场景验证规格、功能和体验','successMeaning':'验证结果有记录并可回溯到SKU','nextAction':'通过后进入页面制作；不通过则调整产品','evidenceIds':refs+issue_refs},
      {'stage':'页面验收','owner':'视觉与内容','hypothesis':'','metrics':'逐项核对主图、详情图、SKU、参数和营销表达','successMeaning':'图片来源清晰且没有结构、颜色或功能冲突','nextAction':'通过后上架；不通过则重新采集或修改页面','evidenceIds':refs},
      {'stage':'首批反馈','owner':'客服与数据','hypothesis':'','metrics':'按SKU记录咨询、评论、退款售后原话与问题批次','successMeaning':'反馈可以归因到具体规格和问题类型','nextAction':'保留通过SKU；集中问题返回供应链整改','evidenceIds':issue_refs},
    ]

def _product_label(raw):
    product=raw.get('product',{}); title=_valid_title(product.get('title'))
    if title:return title
    route=detect_category(raw); attrs=_attribute_rows(raw); parts=[]
    for words in [('品牌',),('品类','类别','商品类型'),('型号','款式','口味','色号')]:
        x,_=_find_attr(attrs,words)
        if x and _txt(x.get('value')).strip() not in parts:parts.append(_txt(x.get('value')).strip())
    return ' '.join(parts[:3]) or route['label'] or '当前商品'

def _review_ref(raw, predicate):
    for i,x in enumerate(raw.get('reviews',[]) or [],1):
        text=_txt(x.get('content'))+' '+_txt(x.get('appendContent'))
        if predicate(text):return text.strip(),f'REV_{i:06d}'
    return '',''

def _buyer_evidence(raw, limit=24):
    images=raw.get('images',{}) or {}; buyer=images.get('buyerShow',[]) or []; index_by_url={url:i for i,url in enumerate(buyer,1)}
    out=[];seen=set()
    for review_index,review in enumerate(raw.get('reviews',[]) or [],1):
        for url in review.get('images',[]) or []:
            image_index=index_by_url.get(url)
            if not image_index or url in seen:continue
            seen.add(url);out.append({'imageEvidenceId':f'IMG_BUYERSHOW_{image_index:04d}','reviewEvidenceId':f'REV_{review_index:06d}','quote':concise_fact(review.get('content'),72),'sku':_txt(review.get('sku')).strip(),'date':_txt(review.get('date')).strip(),'sourceLabel':'评论原图'})
            if len(out)>=limit:return out
    return out

REVIEW_OUTCOME_RULES=(
  ('花型与空间呈现',('好看','漂亮','可爱','高级','颜值','搭配儿童房','风格也搭','氛围感','少女心'),'明确反馈花型好看、可爱或能搭配房间'),
  ('触感与睡感',('柔软','舒服','舒适','亲肤','滑滑','光滑','手感很好','手感特别'),'明确反馈柔软、舒服、亲肤或光滑'),
  ('儿童接受度',('孩子很喜欢','孩子特别喜欢','宝宝超喜欢','宝宝很喜欢','女儿很喜欢'),'明确反馈孩子、宝宝或女儿喜欢'),
  ('质量与做工',('做工也很好','做工细节都很棒','做工好精致','质量真的很好','质量特别好','质量非常好','品质真不错','质感很好'),'明确反馈质量或做工表现好'),
  ('复购意愿',('再买一套','下次还会来','下次再回购','认准你家'),'明确表达再次购买、回购或认准店铺'),
)

REVIEW_ISSUE_RULES=(
  ('size_fit','尺寸与床型适配',r'罩不住|铺不满|盖不住|垂边不够|床单.{0,10}(?:短|小|窄|不够)|(?:尺寸|大小).{0,8}(?:偏小|偏大|不合适|不符)|被套.{0,8}(?:偏小|偏大|不合适|不符)','按评论对应 SKU 在实际床垫铺装，记录床单垂边、床笠包裹和被套适配后再锁版'),
  ('wash_stability','首洗缩水或掉色',r'缩水|掉色|褪色|洗后.{0,8}(?:变形|发硬|变小|小了)','按标示洗护完成首洗，对比洗前洗后尺寸、颜色和表面状态'),
  ('pilling_shedding','起球或掉毛',r'起球|掉毛|粘毛|浮毛','完成面料与成品洗后起球、掉毛检查，不通过则调整面料或后整理'),
  ('odor','气味',r'异味|刺鼻|味道.{0,8}(?:大|重|难闻)','检查面料、印染、填充与包装来源，成品散味验收通过后再入仓'),
  ('sewing_breakage','缝制或破损',r'破损|开线|脱线|线头.{0,8}(?:多|严重)|走线.{0,8}(?:歪|差)|缝.{0,8}(?:开|坏)','复核裁剪、缝制、锁边和出货外观，问题批次返修后复检'),
  ('material_mismatch','材质信息不一致',r'不是纯棉|不是全棉|材质.{0,8}(?:不符|不一样)|面料.{0,8}(?:不符|不一样)|货不对板','核对供应资料、水洗标、吊牌、样品和页面材质表述，一处不一致即暂停宣称'),
  ('packaging_transfer','包装污染或转移',r'包装.{0,10}(?:染|脏|掉色)|油墨|印字.{0,8}(?:沾|染)|包装袋.{0,8}(?:沾|染)','做包装接触与运输模拟，检查油墨、颜色、气味和污渍转移'),
  ('pillow_filling','枕芯或填充支撑',r'枕头.{0,8}(?:太高|太低|太硬|太软|塌)|枕芯.{0,8}(?:太高|太低|太硬|太软|塌)|填充.{0,8}(?:少|不均|结团)','核对填充重量、均匀度和支撑感，按套件 BOM 留样验收'),
  ('color_difference','实物色差',r'色差|实物.{0,8}(?:偏黄|偏白|偏暗|颜色不一样)|图片.{0,8}(?:不一样|差很多)','自然光核对实物色号与页面图片，页面补充真实色差说明'),
)

def _review_issue(raw):
    reviews=raw.get('reviews',[]) or []
    route=detect_category(raw)
    dynamic_rules={
      'apparel':(
        ('apparel_size','尺码或版型适配',r'尺码|码数|偏大|偏小|不合身|显胖|版型.{0,8}(?:不符|奇怪)','按评论对应SKU复核尺码表、版型数据和真人试穿效果'),
        ('apparel_workmanship','服饰做工问题',r'开线|脱线|线头|拉链|纽扣|破损','复核对应批次的缝制、辅料和出货外观'),
        ('apparel_color','服饰图实色差',r'色差|实物.{0,8}颜色|图片.{0,8}不一样','在统一自然光下核对实物色号与页面图片'),
      ),
      'food':(
        ('food_taste','口味反馈分化',r'难吃|太甜|太咸|口味.{0,8}(?:怪|淡|重)','按口味SKU统计反馈并复核配方与页面描述'),
        ('food_package','食品包装或运输问题',r'漏液|胀包|破包|碎了|包装.{0,8}(?:破|坏)','复核包装密封、内衬和运输模拟结果'),
        ('food_freshness','新鲜度或保质风险',r'变质|过期|不新鲜|临期|发霉','核对生产批次、保质期、储存和出库规则'),
      ),
      'beauty':(
        ('beauty_sensitive','敏感或刺激反馈',r'过敏|刺激|刺痛|泛红|闷痘|爆痘','按肤质和使用方式复核对应SKU与成分风险提示'),
        ('beauty_claim','功效体验不一致',r'没效果|无效|不吸收|搓泥|拔干|油腻','拆分功效宣称、使用方法与真实反馈，降低过强承诺'),
        ('beauty_package','美妆包装问题',r'漏液|泵头|瓶口|包装.{0,8}(?:坏|破)','复核包材、密封和运输跌落测试'),
      ),
      'digital':(
        ('digital_compat','兼容或连接问题',r'不兼容|连不上|连接.{0,8}(?:失败|断)|识别不了','按设备型号、系统和接口复核兼容列表'),
        ('digital_performance','性能或续航问题',r'卡顿|发热|死机|续航.{0,8}(?:差|短)|掉电|速度慢','按配置版本复测性能、温度、续航和稳定性'),
        ('digital_quality','数码质量问题',r'失灵|故障|坏了|杂音|断触|黑屏','定位批次和故障场景，复核出厂测试与售后记录'),
      ),
    }
    for key,label,pattern,action in dynamic_rules.get(route['key'],()):
        rx=re.compile(pattern,re.I)
        for i,review in enumerate(reviews,1):
            if not isinstance(review,dict):continue
            text=(' '.join(x for x in (_txt(review.get('content')).strip(),_txt(review.get('appendContent')).strip()) if x)).strip()
            if text and rx.search(text):
                return {'key':key,'label':label,'action':action,'text':text,'evidenceId':f'REV_{i:06d}','sku':_txt(review.get('sku')).strip()}
    for key,label,pattern,action in REVIEW_ISSUE_RULES:
        rx=re.compile(pattern,re.I)
        for i,review in enumerate(reviews,1):
            if not isinstance(review,dict):continue
            text=(' '.join(x for x in (_txt(review.get('content')).strip(),_txt(review.get('appendContent')).strip()) if x)).strip()
            if text and rx.search(text):
                return {'key':key,'label':label,'action':action,'text':text,'evidenceId':f'REV_{i:06d}','sku':_txt(review.get('sku')).strip()}
    for i,review in enumerate(reviews,1):
        if not isinstance(review,dict):continue
        text=(' '.join(x for x in (_txt(review.get('content')).strip(),_txt(review.get('appendContent')).strip()) if x)).strip()
        if text and any(k in text for k in NEGATIVE_REVIEW_KEYS):
            return {'key':'other','label':'评论明确问题','action':'按评论原文定位对应 SKU 和批次，复核实物后再决定是否修版','text':text,'evidenceId':f'REV_{i:06d}','sku':_txt(review.get('sku')).strip()}
    return {}

HOME_TEXTILE_FACT_GROUPS=(
  ('适用场景 / 人群',('适用场景','适用人群','适用对象'),'用于锁定使用场景与商品表达'),
  ('套件 BOM',('套件组成','套件件数','件数','组合形式','套件种类'),'用于逐件核对套装构成与包装清单'),
  ('材质 / 成分',('成分含量','床单面料材质','被面材质','被里材质','枕套材质','面料材质','棉种类','材质'),'用于核对供应资料、水洗标与页面宣称'),
  ('支数 / 密度 / 克重',('面料支数','纱线支数','织物密度','面料密度','克重'),'用于锁定面料规格，不等同于检测结果'),
  ('床型 / 成品尺寸',('适用床尺寸','床品尺寸','被套尺寸','床单尺寸','床笠尺寸','枕套尺寸','尺寸'),'用于床型铺装和 SKU 适配校验'),
  ('工艺 / 结构',('床品工艺','织造工艺','印染工艺','面料工艺','绗缝工艺','填充工艺','工艺','款式'),'用于样品结构与生产工艺核对'),
  ('等级 / 标准',('产品等级','安全类别','执行标准','质量等级'),'用于核对吊牌、检测资料与页面口径'),
  ('洗护说明',('洗涤说明','洗护说明','洗涤方式','护理方法'),'用于首洗验证和售后说明'),
)

def _home_textile_parameter_facts(raw, limit=10):
    attrs=_attribute_rows(raw);used=set();facts=[]
    for label,words,meaning in HOME_TEXTILE_FACT_GROUPS:
        matches=[];refs=[]
        for i,item in enumerate(attrs,1):
            if i in used:continue
            name=_txt(item.get('name')).strip()
            if any(word in name for word in words):
                used.add(i);matches.append(item);refs.append(f'ATTR_{i:04d}')
        if not matches:continue
        value='；'.join((f"{x.get('name')}：{x.get('value')}" if len(matches)>1 else _txt(x.get('value'))) for x in matches)
        facts.append({'label':label,'value':concise_fact(value,220),'productMeaning':meaning,'evidenceIds':refs})
        if len(facts)>=limit:return facts
    for i,item in enumerate(attrs,1):
        if i in used:continue
        facts.append({'label':_txt(item.get('name')),'value':concise_fact(item.get('value'),180),'productMeaning':'页面采集参数，批量上架前与实物核对','evidenceIds':[f'ATTR_{i:04d}']})
        if len(facts)>=limit:break
    return facts

def _home_textile_gates(raw, issue=None):
    attrs=_attribute_rows(raw);issue=issue or _review_issue(raw)
    material,material_id=_find_attr(attrs,('成分含量','床单面料材质','被面材质','被里材质','材质','棉种类'))
    pieces,pieces_id=_find_attr(attrs,('套件组成','套件件数','件数','组合形式'))
    size,size_id=_find_attr(attrs,('适用床尺寸','床品尺寸','被套尺寸','床单尺寸','尺寸'))
    wash,wash_id=_find_attr(attrs,('洗涤说明','洗护说明','洗涤方式','护理方法'))
    grade,grade_id=_find_attr(attrs,('产品等级','安全类别','执行标准'))
    gates=[]
    if pieces:gates.append({'priority':'P0','label':'套件 BOM 一致','action':'逐件核对样品、包装清单、SKU 名称和详情页的品类、件数与尺寸','businessReason':'套件构成必须与页面参数一致','evidenceIds':[pieces_id]})
    if material or grade:gates.append({'priority':'P0','label':'材质与标签一致','action':'核对供应资料、水洗标、吊牌、样品和页面的成分、等级与执行标准','businessReason':'页面宣称必须能回溯到真实资料','evidenceIds':[x for x in (material_id,grade_id) if x]})
    if size:gates.append({'priority':'P0','label':'床型与尺寸实测','action':'按页面覆盖床型和床垫高度逐一铺装，记录床单垂边、床笠包裹与被套适配','businessReason':'每个 SKU 通过实床铺装后才能锁版','evidenceIds':[size_id]})
    gates.append({'priority':'P0','label':'首洗前后验收','action':'按标示洗护完成首洗，记录各部件尺寸、颜色、起球掉毛和外观变化','businessReason':'结果须符合页面声明及企业验收标准','evidenceIds':[wash_id] if wash_id else []})
    gates.append({'priority':'P1','label':'缝制与填充验收','action':'检查走线、锁边、拉链、线头、污渍；含枕芯或被芯时同步检查填充重量与均匀度','businessReason':'成品外观与套件完整性通过后再入仓','evidenceIds':[pieces_id] if pieces_id else []})
    gates.append({'priority':'P1','label':'气味与包装转移','action':'完成密封、运输模拟和开箱检查，记录气味、油墨、颜色与污渍转移','businessReason':'包装与商品接触后不得污染成品','evidenceIds':[]})
    if issue:gates.insert(0,{'priority':'P0','label':f"评论问题：{issue['label']}",'action':issue['action'],'businessReason':concise_fact(issue['text'],72),'evidenceIds':[issue['evidenceId']]})
    return gates[:7]

def _home_textile_validation_rows(raw):
    attrs=_attribute_rows(raw);all_refs=[f'ATTR_{i:04d}' for i in range(1,min(len(attrs),8)+1)]
    issue=_review_issue(raw);issue_refs=[issue['evidenceId']] if issue else []
    return [
      {'stage':'商品定义','owner':'商品企划 / 商品运营','hypothesis':'','metrics':'锁定目标场景、套件 BOM、材质工艺、床型尺寸与 SKU','successMeaning':'商品定义表与页面已采参数逐项对应','nextAction':'通过后发起打样；不一致则退回定义','evidenceIds':all_refs},
      {'stage':'资料与标签','owner':'供应链 / 品控','hypothesis':'','metrics':'核对供应资料、水洗标、吊牌、包装和样品','successMeaning':'成分、等级、标准与洗护口径一致','nextAction':'通过后确认样品；不一致则暂停宣称','evidenceIds':all_refs},
      {'stage':'实床与首洗','owner':'品控 / 供应链','hypothesis':'','metrics':'按 SKU 实床铺装；首洗前后记录尺寸、掉色、起球掉毛与外观','successMeaning':'符合页面声明及企业验收标准，记录可追溯','nextAction':'通过后锁版；不通过则调整尺寸、面料或工艺','evidenceIds':all_refs+issue_refs},
      {'stage':'成品与包装','owner':'品控 / 仓配','hypothesis':'','metrics':'检查缝制、填充、污渍、气味及包装接触转移','successMeaning':'成品与包装验收项全部通过','nextAction':'通过后入仓；问题批次返修复检','evidenceIds':issue_refs},
      {'stage':'页面验收','owner':'视觉与内容','hypothesis':'','metrics':'逐项核对主图、详情、SKU 表、尺寸图、材质与洗护说明','successMeaning':'页面图片来源清晰，且与样品、包装、标签一致','nextAction':'通过后上架；不通过则停止投放','evidenceIds':all_refs},
      {'stage':'首批反馈','owner':'客服与数据','hypothesis':'','metrics':'按 SKU 记录咨询、评论、退款售后原话与问题批次','successMeaning':'反馈均可回溯到 SKU、订单和问题类型','nextAction':'保留通过 SKU；集中问题返回供应链整改','evidenceIds':[]},
    ]

def _review_outcomes(raw):
    reviews=raw.get('reviews',[]) or [];out=[]
    for label,keys,result in REVIEW_OUTCOME_RULES:
        matched=[]
        for i,x in enumerate(reviews,1):
            if not isinstance(x,dict):continue
            text=_txt(x.get('content'))+' '+_txt(x.get('appendContent'))
            if any(k in text for k in keys):matched.append(f'REV_{i:06d}')
        if matched:out.append({'label':label,'value':f'{len(matched)} 条评论{result}','note':f'评论结果 · {len(matched)}/{len(reviews)}','evidenceIds':matched})
    return out

def _find_outcome(outcomes, label):
    for item in outcomes or []:
        if isinstance(item,dict) and item.get('label')==label:return item
    return {}

def _generic_selling_points(raw, base=None, issue=None):
    base=base or baseline(raw); issue=issue or _review_issue(raw); route=detect_category(raw)
    points=[]
    for fact in _category_parameter_facts(raw,5):
        value=concise_fact(fact.get('value'),52)
        if not value:continue
        points.append({
          'slogan':f"{fact.get('label')}：{value}",
          'consumerValue':f"让消费者直接确认{route['label']}商品的{fact.get('label')}",
          'evidenceSource':'页面参数',
          'productAction':f"主图或详情页用实物、参数或资料证明{fact.get('label')}",
          'evidenceIds':fact.get('evidenceIds') or []
        })
    outcomes=_review_outcomes(raw)
    for outcome in outcomes[:2]:
        refs=_arr(outcome.get('evidenceIds'))
        if refs:points.append({'slogan':concise_fact(outcome.get('value'),46),'consumerValue':'把评论已验证体验转成页面可理解的利益表达','evidenceSource':'评论聚合','productAction':'保留对应产品体验，并用实拍或细节证据承接','evidenceIds':refs[:8]})
    if issue:
        points.append({'slogan':f"优先修正{issue['label']}",'consumerValue':'避免下一款重复当前已出现的问题','evidenceSource':'评论明确问题','productAction':issue['action'],'evidenceIds':[issue['evidenceId']]})
    unique=[];seen=set()
    for point in points:
        key=point.get('slogan')
        if key and key not in seen:seen.add(key);unique.append(point)
    return unique[:6]

def _derive_next_selling_points(raw, base=None, issue=None):
    if detect_category(raw)['key']!='home_textile':
        return _generic_selling_points(raw,base,issue)
    base=base or baseline(raw); issue=issue or _review_issue(raw)
    attrs=_attribute_rows(raw); outcomes=_review_outcomes(raw)
    title=_txt((raw.get('product') or {}).get('title')).strip()
    sku_text=' '.join(_txt(x.get('name') or x.get('text') or x.get('value')) for x in raw.get('sku',[]) if isinstance(x,dict))[:3000]
    attr_text=' '.join(f"{x.get('name')}:{x.get('value')}" for x in attrs)[:3000]
    corpus=' '.join([title,sku_text,attr_text])
    size,size_id=_find_attr(attrs,('适用床尺寸','床品尺寸','被套尺寸','床单尺寸','床笠尺寸','尺寸'))
    material,material_id=_find_attr(attrs,('成分含量','床单面料材质','被面材质','被里材质','枕套材质','面料材质','材质','棉种类'))
    craft,craft_id=_find_attr(attrs,('床品工艺','织造工艺','面料密度','面料支数','工艺'))
    pattern,pattern_id=_find_attr(attrs,('主图案类型','风格','颜色分类','款式'))
    pieces,pieces_id=_find_attr(attrs,('件数','套件组成','套件件数','组合形式'))
    touch=_find_outcome(outcomes,'触感与睡感')
    design=_find_outcome(outcomes,'花型与空间呈现')
    size_outcome=_find_outcome(outcomes,'儿童接受度') if '儿童' in corpus else {}
    points=[]
    def add(slogan, consumer, evidence, action, refs):
        refs=[x for x in _uniq(refs) if x]
        if not refs:return
        points.append({'slogan':slogan,'consumerValue':consumer,'evidenceSource':evidence,'productAction':action,'evidenceIds':refs})
    has_dorm_scene=bool(re.search(r'宿舍|大学|大学生',corpus))
    has_small_bed=bool(re.search(r'0\.?9|0.9|1\.?2|1.2|90|120',corpus))
    scene='宿舍床专用' if has_dorm_scene else ('小床专用' if has_small_bed else '目标床型专用')
    if size:
        size_value=_txt(size.get('value'))
        if has_dorm_scene and re.search(r'0\.?9|0.9|90',size_value+corpus):
            add('0.9m 宿舍床专用，不用猜尺寸','消费者一眼知道这套床品适配哪张宿舍床','页面尺寸参数、目标场景与 SKU 信息', '首图拍 0.9m 宿舍床效果；详情页单独列被套、床单/床笠、枕套尺寸', [size_id,pieces_id])
        elif has_small_bed:
            add('小床尺寸单独讲清，不让用户猜','消费者一眼知道这套床品适配哪张床','页面尺寸参数与 SKU 信息', '首图拍对应床型效果；详情页单独列被套、床单/床笠、枕套尺寸', [size_id,pieces_id])
        else:
            add(f'{scene}，尺寸直接看清','减少买错床型和规格的理解成本','页面尺寸参数', '按主销床型分别拍铺设边界和套件尺寸图', [size_id,pieces_id])
    if re.search(r'床笠',corpus) and re.search(r'床单',corpus):
        add('床笠款 / 床单款分开买，不混规格','消费者能判断自己要包床垫款还是普通床单款','SKU 与参数中同时出现床笠、床单', '分标题、分 SKU、分主图、分详情页说明两种结构', [x for x in (size_id,pieces_id) if x])
    elif re.search(r'床笠',corpus):
        add('床笠款包住床垫，铺上不跑位','强调包裹、防滑和宿舍床垫适配','页面 SKU/参数出现床笠', '拍四角包裹、床垫厚度、洗后回弹', [size_id,pieces_id])
    if material:
        mat=_txt(material.get('value'))
        add(('100%棉基础面料，贴身不将就' if '100' in mat or '棉' in mat else '基础面料信息讲清楚'),'消费者知道材质承诺来自页面参数','页面材质/成分参数', '材质页同时展示水洗标、吊牌和面料近景', [material_id]+_arr(touch.get('evidenceIds'))[:3])
    if touch.get('evidenceIds'):
        add('全棉要柔软，不要偏硬刮肤','把“全棉”转成可感知的触感利益','评论聚合中有柔软、亲肤或舒服反馈', '打样增加触感验收、柔软后整理和首洗后手感复检', _arr(touch.get('evidenceIds'))[:6]+([material_id] if material_id else []))
    if issue and issue.get('key')=='wash_stability':
        add('首洗后尺寸稳定，洗完还能铺得平','消费者买床品要省心耐用','评论明确出现缩水或掉色反馈', '每批做首洗尺寸、颜色和外观记录，页面展示洗前洗后结果', [issue.get('evidenceId'),size_id])
    elif issue and issue.get('key')=='size_fit':
        add('铺上合适，边界不尴尬','消费者最怕买回去罩不住、铺不满','评论明确出现尺寸适配问题', '按 SKU 实床铺设并记录床单垂边、床笠包裹和被套适配', [issue.get('evidenceId'),size_id])
    if design.get('evidenceIds') or pattern:
        add(('耐看花色，宿舍日常好打理' if has_dorm_scene else '耐看花色，日常好打理'),'花色不是只讲好看，要讲耐看、耐脏、适合日常','评论花型反馈或页面花型参数', '保留已验证花型方向；主图补自然光、局部纹样和整套陈列', _arr(design.get('evidenceIds'))[:5]+([pattern_id] if pattern_id else []))
    if craft:
        add('工艺参数讲明白，品质感有出处','支数、密度、印染或织造信息不只放参数表','页面工艺/支数/密度参数', '详情页用面料近景、工艺说明和水洗标承接参数', [craft_id])
    return points[:6]

def _apply_selling_points(exp, raw, base=None):
    if not isinstance(exp,dict):return exp
    base=base or baseline(raw); issue=_review_issue(raw); points=_derive_next_selling_points(raw,base,issue)
    plans=exp.setdefault('newProductPlans',{})
    plan_list=plans.setdefault('plans',[])
    if plan_list and isinstance(plan_list[0],dict):
        existing=_arr(plan_list[0].get('sellingPoints'))
        plan_list[0]['sellingPoints']=points[:5] or existing
        if points:
            kept=[x for x in _arr(plan_list[0].get('specs')) if not (isinstance(x,dict) and x.get('label')=='下一款明确卖点')]
            plan_list[0]['specs']=[{'label':'下一款明确卖点','value':' / '.join(x['slogan'] for x in points[:3])}]+kept
    if points:
        overview=exp.setdefault('ownerOverview',{})
        cards=[x for x in overview.setdefault('cards',[]) if not (isinstance(x,dict) and x.get('label')=='下一款明确卖点')]
        cards.append({'label':'下一款明确卖点','headline':points[0]['slogan'],'bullets':[x['slogan'] for x in points[1:3]],'tone':'good','evidenceIds':points[0].get('evidenceIds',[])})
        overview['cards']=cards
    exp['contentGovernance']={
        'role':'专业电商报告输出总监',
        'structure':'商品产品分析报告',
        'sourceRule':'页面参数、主图/详情/SKU/评论图、评论统计、问答和监测数据分来源呈现',
        'evidenceRule':'每个结论必须引用 evidenceIds；缺失证据不补写',
        'visualRule':'报告渲染采用时尚网页设计美学：大图完整展示、清晰层级、短结论、证据卡片化'
    }
    return exp

def _generic_platform_overview(raw, base=None):
    base=base or baseline(raw); route=detect_category(raw); attrs=_attribute_rows(raw); sales=raw.get('sales') or {}; issue=_review_issue(raw)
    price=_txt(sales.get('currentPrice')).strip(); sold=_txt(sales.get('sold') or sales.get('cumulativeSales')).strip(); facts=[]
    facts.append({'label':'商品是什么','value':_product_label(raw),'note':f"自动识别：{route['label']}",'evidenceIds':['P_TITLE'] if _valid_title((raw.get('product') or {}).get('title')) else []})
    if price or sold:facts.append({'label':'价格与销量','value':' · '.join(x for x in (f'¥{price}' if price else '',f'公开销量 {sold}' if sold else '') if x),'note':'采集时页面公开值','evidenceIds':[x for x in ('S_CURRENTPRICE' if price else '','S_SOLD' if sold else '') if x]})
    if attrs:facts.append({'label':'动态参数','value':f'已采集 {len(attrs)} 项','note':'；'.join(f"{x.get('name')}：{x.get('value')}" for x in attrs[:3]),'evidenceIds':[f'ATTR_{i:04d}' for i in range(1,min(3,len(attrs))+1)]})
    outcomes=_review_outcomes(raw); actions=[]
    if outcomes:actions.append({'label':'保留','value':'保留评论已验证的实际体验','note':outcomes[0].get('value'),'evidenceIds':outcomes[0].get('evidenceIds') or []})
    if issue:actions.append({'label':'修正','value':issue['action'],'note':issue['label'],'evidenceIds':[issue['evidenceId']]})
    refs=[f'ATTR_{i:04d}' for i in range(1,min(3,len(attrs))+1)]
    if refs:actions.append({'label':'下一步','value':f"按{route['label']}关键维度核对样品、SKU和页面一致性",'note':'当前证据支持商品拆解与新品开发','evidenceIds':refs})
    public=_txt(base.get('publicReviewCount')).strip();sample=f"已采有效评论 {base.get('reviewCount',0)}"+(f" / 页面公开 {public}" if public else '')
    sample+='；评论采集已确认完整' if base.get('reviewCollectionComplete') else '；评论采集未确认完整'
    return {'chapterTitle':'商品分析结论','verdict':'先保留已确认信息，再处理评论问题和页面缺口','facts':facts,'outcomes':outcomes[:5],'actions':actions,'checks':_category_gates(raw,issue)[:3],'questions':[{'label':'购买问答原文','value':concise_fact(x.get('question'),64),'evidenceIds':[f'QA_{i:05d}']} for i,x in enumerate(raw.get('questions',[])[:2],1) if isinstance(x,dict)],'sampleNote':sample,'decisionBoundary':{'supported':'商品拆解、样品验证、规格校准、页面修正和新品方案','excluded':'生产规模、利润、转化、市场容量','required':'完整评论、流量转化、退款售后、成本毛利'}}

def build_platform_overview(raw, base=None):
    if detect_category(raw)['key']!='home_textile':
        return _generic_platform_overview(raw,base)
    base=base or baseline(raw);attrs=_attribute_rows(raw);sales=raw.get('sales',{}) or {};topics=base.get('topics',[]) or []
    label=_product_label(raw);price=_txt(sales.get('currentPrice')).strip();sold=_txt(sales.get('sold') or sales.get('cumulativeSales')).strip()
    material,material_id=_find_attr(attrs,('床单面料材质','被面材质','材质','成分'))
    thread,thread_id=_find_attr(attrs,('面料支数',));pieces,pieces_id=_find_attr(attrs,('件数','套件组成'))
    size_attr,size_id=_find_attr(attrs,('适用床尺寸','床品尺寸','尺寸'));audience,audience_id=_find_attr(attrs,('适用人群',))
    craft,craft_id=_find_attr(attrs,('床品工艺','织造工艺','工艺'));grade,grade_id=_find_attr(attrs,('产品等级','安全类别','执行标准'))
    promotions=[x for x in raw.get('promotions',[]) or [] if isinstance(x,dict) and _txt(x.get('text')).strip()]
    issue=_review_issue(raw);neg_text=issue.get('text','');neg_id=issue.get('evidenceId','')
    facts=[]
    product_parts=[];product_refs=[]
    for x,eid in ((material,material_id),(thread,thread_id),(pieces,pieces_id),(craft,craft_id)):
        if x:product_parts.append(f"{x.get('name')}：{x.get('value')}");product_refs.append(eid)
    facts.append({'label':'商品是什么','value':label,'note':'；'.join(product_parts[:3]),'evidenceIds':product_refs[:3]})
    if audience:facts.append({'label':'页面适用人群','value':audience.get('value'),'note':'页面参数原文','evidenceIds':[audience_id]})
    if price or sold:
        facts.append({'label':'价格与销量','value':' · '.join(x for x in (f'¥{price}' if price else '',f'公开销量 {sold}' if sold else '') if x),'note':'采集时页面公开值','evidenceIds':[x for x in ('S_CURRENTPRICE' if price else '','S_SOLD' if sold else '') if x]})
    if promotions:facts.append({'label':'页面活动','value':concise_fact(promotions[0].get('text'),72),'note':'活动以采集时页面展示为准','evidenceIds':['PROMO_0001']})
    outcomes=_review_outcomes(raw)
    checks=[]
    if size_attr:checks.append({'priority':'运营校验','label':'核对床型与套件尺寸','value':size_attr.get('value'),'evidenceIds':[size_id]})
    if neg_id:checks.append({'priority':'明确反馈','label':f"重点核对{issue['label']}",'value':concise_fact(neg_text,72),'evidenceIds':[neg_id]})
    if grade:checks.append({'priority':'页面参数','label':grade.get('name'),'value':grade.get('value'),'evidenceIds':[grade_id]})
    questions=[]
    for i,q in enumerate(raw.get('questions',[])[:3],1):
        text=concise_fact(q.get('question'),64) if isinstance(q,dict) else ''
        if text:questions.append({'label':'购买问答原文','value':text,'evidenceIds':[f'QA_{i:05d}']})
    outcome_map={x.get('label'):x for x in outcomes}
    appearance=outcome_map.get('花型与空间呈现',{});touch=outcome_map.get('触感与睡感',{})
    positive_refs=_uniq((appearance.get('evidenceIds') or [])+(touch.get('evidenceIds') or []))
    actions=[]
    if positive_refs:actions.append({'label':'保留','value':'保留评论已验证的花型呈现与触感表达','note':'按本次有效评论统计','evidenceIds':positive_refs})
    if issue:actions.append({'label':'修正','value':issue['action'],'note':f"评论明确出现{issue['label']}",'evidenceIds':[neg_id]})
    actions.append({'label':'下一步','value':'进入样品验证；同步核对成本、家纺验收项与页面一致性','note':'当前证据支持样品和页面校准','evidenceIds':[x for x in (size_id,neg_id) if x]})
    verdict='保留已验证体验，优先核对尺寸、成本和页面一致性'
    public=_txt(base.get('publicReviewCount')).strip()
    sample=f"已采有效评论 {base.get('reviewCount',0)}"+(f" / 页面公开 {public}" if public else '')
    sample+='；评论采集已确认完整' if base.get('reviewCollectionComplete') else '；评论采集未确认完整'
    return {'chapterTitle':'商品分析结论','verdict':verdict,'facts':facts[:4],'outcomes':outcomes[:5],'actions':actions,'checks':checks[:3],'questions':questions[:2],'sampleNote':sample,
            'decisionBoundary':{'supported':'商品拆解、样品验证、规格校准、页面信息修正','excluded':'生产规模、利润、转化、市场容量','required':'完整评论、流量转化、退款售后、成本毛利'}}

def _core_solution_ok(exp):
    if not isinstance(exp,dict):return False
    checks=(
      bool(_txt((exp.get('reportSummary') or {}).get('title')).strip()),
      bool(_txt((exp.get('reportSummary') or {}).get('verdict')).strip()),
      isinstance((exp.get('ownerOverview') or {}).get('cards'),list) and len((exp.get('ownerOverview') or {}).get('cards'))>=3,
      isinstance((exp.get('productExperience') or {}).get('parameterFacts'),list) and len((exp.get('productExperience') or {}).get('parameterFacts'))>=2,
      isinstance((exp.get('productExperience') or {}).get('gates'),list) and len((exp.get('productExperience') or {}).get('gates'))>=2,
      isinstance((exp.get('newProductPlans') or {}).get('plans'),list) and len((exp.get('newProductPlans') or {}).get('plans'))>=1,
      isinstance((exp.get('validationLoop') or {}).get('rows'),list) and len((exp.get('validationLoop') or {}).get('rows'))>=3,
    )
    return all(checks)

def core_solution_diagnostics(exp):
    exp=exp if isinstance(exp,dict) else {}
    specs=(
      ('reportSummary.title',bool(_txt((exp.get('reportSummary') or {}).get('title')).strip()),1,1),
      ('reportSummary.verdict',bool(_txt((exp.get('reportSummary') or {}).get('verdict')).strip()),1,1),
      ('ownerOverview.cards',isinstance((exp.get('ownerOverview') or {}).get('cards'),list),len((exp.get('ownerOverview') or {}).get('cards') or []),3),
      ('productExperience.parameterFacts',isinstance((exp.get('productExperience') or {}).get('parameterFacts'),list),len((exp.get('productExperience') or {}).get('parameterFacts') or []),2),
      ('productExperience.gates',isinstance((exp.get('productExperience') or {}).get('gates'),list),len((exp.get('productExperience') or {}).get('gates') or []),2),
      ('newProductPlans.plans',isinstance((exp.get('newProductPlans') or {}).get('plans'),list),len((exp.get('newProductPlans') or {}).get('plans') or []),1),
      ('validationLoop.rows',isinstance((exp.get('validationLoop') or {}).get('rows'),list),len((exp.get('validationLoop') or {}).get('rows') or []),3),
    )
    checks=[]
    for field,right_type,actual,minimum in specs:
        ok=bool(right_type and actual>=minimum)
        checks.append({'field':field,'actual':actual,'minimum':minimum,'ok':ok})
    return {'ok':all(x['ok'] for x in checks),'checks':checks,'missing':[x['field'] for x in checks if not x['ok']]}

def repair_core_solution_with_evidence(draft, raw, base=None):
    """Fill only missing core report structure from deterministic, evidence-backed facts."""
    draft=dict(draft or {})
    evidence=build_fallback_experience(raw,base)
    valid_ids={x.get('id') for x in evidence_ledger(normalize(raw)) if x.get('id')}
    scalar_rules=(('reportSummary','title'),('reportSummary','verdict'))
    list_rules=(
      ('ownerOverview','cards',3),
      ('productExperience','parameterFacts',2),('productExperience','gates',2),
      ('newProductPlans','plans',1),('validationLoop','rows',3),
    )
    for section,key in scalar_rules:
        current=dict(draft.get(section) or {})
        source=evidence.get(section) or {}
        if not _txt(current.get(key)).strip():current[key]=source.get(key,'')
        draft[section]=current
    for section,key,minimum in list_rules:
        current=dict(draft.get(section) or {})
        source=evidence.get(section) or {}
        values=_filter_evidence_items(current.get(key),valid_ids)
        source_values=_filter_evidence_items(source.get(key),valid_ids)
        if section=='ownerOverview' and key=='cards' and len(source_values)<minimum:
            for fact in _filter_evidence_items((evidence.get('productExperience') or {}).get('parameterFacts'),valid_ids):
                source_values.append({'label':fact.get('label') or '商品参数','headline':fact.get('value') or fact.get('note') or '已采集','bullets':[fact.get('note')] if fact.get('note') else [],'tone':'normal','evidenceIds':fact.get('evidenceIds') or []})
        for item in source_values:
            if len(values)>=minimum:break
            if isinstance(item,dict) and item not in values:values.append(item)
        current[key]=values;draft[section]=current
    return draft

def build_generic_fallback_experience(raw, base=None):
    raw=normalize(raw);base=base or baseline(raw);route=detect_category(raw);attrs=_attribute_rows(raw);sales=raw.get('sales') or {};issue=_review_issue(raw)
    label=_product_label(raw);review_count=base.get('reviewCount',0);question_count=base.get('questionCount',0);sold=_txt(sales.get('sold')).strip();price=_txt(sales.get('currentPrice')).strip()
    refs=[f'ATTR_{i:04d}' for i in range(1,min(len(attrs),8)+1)];facts=_category_parameter_facts(raw);gates=_category_gates(raw,issue);points=_generic_selling_points(raw,base,issue)
    main_ids=[f'IMG_MAIN_{i:04d}' for i,_ in enumerate((raw.get('images') or {}).get('main',[])[:8],1)]
    visual=[]
    roles=('商品全貌与第一印象','核心卖点与使用场景','规格、款式或选择信息','材质、成分或配置证据','细节、功能或实际效果','服务、合规或补充说明')
    for i,eid in enumerate(main_ids):
        role=roles[min(i,len(roles)-1)]
        visual.append({'label':f'{i+1:02d} / 已核验主图','headline':role,'explanation':'','imageEvidenceId':eid,'frameInfo':'按原图完整展示，识别主体、场景、配色、文字与信息层级','persuasionTask':role,'weakness':'','action':'确保本图承担独立购买确认任务，并与参数和SKU一致','evidenceIds':[eid]})
    owner=[
      {'label':'商品定义','headline':label,'bullets':[f"自动识别为{route['label']}"],'tone':'normal','evidenceIds':['P_TITLE'] if _valid_title((raw.get('product') or {}).get('title')) else refs[:1]},
      {'label':'公开销售事实','headline':' · '.join(x for x in (f'¥{price}' if price else '',f'销量 {sold}' if sold else '') if x) or '价格或销量未采集','bullets':[],'tone':'normal','evidenceIds':[x for x in ('S_CURRENTPRICE' if price else '','S_SOLD' if sold else '') if x]},
      {'label':'当前用户信号','headline':f'有效评论 {review_count} 条','bullets':[x.get('topic') for x in (base.get('topics') or [])[:3]],'tone':'good','evidenceIds':[f'REV_{i:06d}' for i in range(1,min(review_count,3)+1)]},
      {'label':'当前明确问题','headline':issue.get('label') if issue else '评论未形成明确问题','bullets':[concise_fact(issue.get('text'),58)] if issue else [],'tone':'risk' if issue else 'normal','evidenceIds':[issue.get('evidenceId')] if issue else refs[:1]},
    ]
    stages=[]
    for i,topic in enumerate((base.get('topics') or [])[:3],1):
        examples=topic.get('examples') or [];evidence=[]
        for j,review in enumerate(raw.get('reviews',[]) or [],1):
            text=_txt(review.get('content'))
            if any(k in text for k in TOPICS.get(topic.get('topic'),[])):evidence.append(f'REV_{j:06d}')
        if evidence:stages.append({'key':f'signal{i}','stageNo':f'{i:02d}','label':topic.get('topic'),'headline':f"{topic.get('mentions')} 条评论提及",'userExpectation':'','positiveConfirmation':concise_fact(examples[0],90) if examples else '', 'friction':'','businessImpact':'将该信号对应到具体SKU和页面证据','quotes':[{'type':'评论','text':concise_fact(examples[0],90),'evidenceId':evidence[0]}] if examples else [],'evidenceIds':evidence[:8]})
    if len(stages)<2 and refs:
        for i,fact in enumerate(facts[:2],len(stages)+1):stages.append({'key':f'fact{i}','stageNo':f'{i:02d}','label':fact.get('label'),'headline':fact.get('value'),'userExpectation':'看清商品关键信息','positiveConfirmation':'页面参数已采集','friction':'','businessImpact':'与SKU和详情页保持一致','quotes':[],'evidenceIds':fact.get('evidenceIds') or []})
    plan_specs=[{'label':x.get('label'),'value':x.get('value')} for x in facts[:6]]
    exp={
      'reportSummary':{'title':f'{label}产品分析报告','subtitle':f'{route["label"]} · 有效评论 {review_count} 条 · 参数 {len(attrs)} 项','verdict':'保留有证据的商品价值，先修正明确问题，再形成差异化新品','verdictDetail':'页面事实、图片、评论和问答分来源呈现。','monitorSummary':''},
      'platformOverview':_generic_platform_overview(raw,base),
      'ownerOverview':{'chapterTitle':f'{route["label"]}商品事实、用户信号与下一款动作','chapterDeck':'','thesis':'','sampleSummary':f'有效评论 {review_count} 条，购买问答 {question_count} 条','cards':owner},
      'visualCommerce':{'chapterTitle':'主图逐张承担不同购买确认任务','chapterDeck':'','items':visual},
      'customerExperience':{'chapterTitle':'当前评论只形成用户信号，不扩大为普遍结论','chapterDeck':'','stats':[{'label':'有效评论','value':str(review_count)},{'label':'购买问答','value':str(question_count)},{'label':'有效参数','value':str(len(attrs))}],'journeyTitle':'评论体验、问题与购买顾虑分开呈现','journeyDeck':'','stages':stages,'buyerEvidence':{'title':'评论原图与对应评价','items':_buyer_evidence(raw)},'signals':{'title':'当前用户信号','positive':[],'questions':[],'risks':[{'label':issue.get('label'),'countOrNote':'明确评论文本','evidenceIds':[issue.get('evidenceId')]}] if issue else []}},
      'productExperience':{'chapterTitle':f'{route["label"]}动态产品定义','chapterDeck':'','parameterFacts':facts,'thesis':'','thesisDetail':'','gates':gates},
      'newProductPlans':{'chapterTitle':'稳健优化、设计升级与差异探索','chapterDeck':'','plans':[{'type':'稳健优化款','name':f'{label}优化款','positioning':'保留已确认价值，优先修正评论问题与页面信息冲突','sellingPoints':points,'specs':plan_specs,'manufacturingGate':{'title':'产品验证','content':'按当前类目关键参数核对样品、SKU、标签、包装和页面'},'visualCommerce':{'title':'主图与详情','content':'首图确认商品全貌，后续图片依次证明规格、材质/成分/配置、使用效果与服务'},'evidenceIds':_uniq((refs[:5])+([issue.get('evidenceId')] if issue else []))}], 'commercialSummary':{'title':'从当前商品保留什么','thesis':'只复用有证据的体验和信息结构','keep':points[0].get('slogan') if points else '页面已采参数','fix':issue.get('label') if issue else '页面与商品一致性','prove':'样品、SKU、参数、图片和评论可以相互核对'}},
      'validationLoop':{'chapterTitle':f'{route["label"]}新品验证闭环','chapterDeck':'','rows':_category_validation_rows(raw)}
    }
    return _apply_selling_points(exp,raw,base)

def build_fallback_experience(raw, base=None):
    if detect_category(raw)['key']!='home_textile':
        return build_generic_fallback_experience(raw,base)
    raw=normalize(raw);base=base or baseline(raw);attrs=_attribute_rows(raw);sales=raw.get('sales',{});label=_product_label(raw)
    sold=_txt(sales.get('sold') or sales.get('cumulativeSales')).strip();price=_txt(sales.get('currentPrice')).strip()
    topics=base.get('topics',[]) or []; top=topics[:3];review_outcomes=_review_outcomes(raw)
    outcome_counts={x.get('label'):len(x.get('evidenceIds') or []) for x in review_outcomes}
    topic_cards=[]
    for t in top:
        topic_cards.append({'label':t.get('topic'),'headline':f"{t.get('mentions',0)} 条评论提及",'explanation':'','bullets':[],'tone':'good','evidenceIds':[]})
    issue=_review_issue(raw);neg_text=issue.get('text','');neg_id=issue.get('evidenceId','')
    size_attr,size_attr_id=_find_attr(attrs,('适用床尺寸','床品尺寸','尺寸'))
    material,material_id=_find_attr(attrs,('床单面料材质','被面材质','材质','成分'))
    thread,thread_id=_find_attr(attrs,('面料支数',))
    pieces,pieces_id=_find_attr(attrs,('件数','套件组成'))
    craft,craft_id=_find_attr(attrs,('床品工艺','织造工艺','工艺'))
    grade,grade_id=_find_attr(attrs,('产品等级','安全类别','执行标准'))
    attr_refs=[x for x in (material_id,thread_id,pieces_id,size_attr_id,craft_id,grade_id) if x]
    top_text='、'.join(f"{x.get('label')} {len(x.get('evidenceIds') or [])}/{base.get('reviewCount',0)}" for x in review_outcomes[:3]) or '当前评论未形成可统计结果'
    review_count=base.get('reviewCount',0);question_count=base.get('questionCount',0)
    customer_results=[]
    if outcome_counts.get('触感与睡感'):customer_results.append(f"{outcome_counts['触感与睡感']} 条确认触感")
    if outcome_counts.get('花型与空间呈现'):customer_results.append(f"{outcome_counts['花型与空间呈现']} 条确认花型")
    sales_bits=[]
    if sold:sales_bits.append(f'公开销量 {sold}')
    if price:sales_bits.append(f'采集价 ¥{price}')
    sales_bits.extend([f'有效评论 {review_count} 条',f'购买问答 {question_count} 条'])
    owner_cards=[
      {'label':'销售事实','headline':' · '.join(sales_bits[:2]) or '页面未展示销量与价格','bullets':sales_bits[2:],'tone':'normal','evidenceIds':[x for x in ('S_SOLD' if sold else '', 'S_CURRENTPRICE' if price else '') if x]},
      {'label':'评论已验证体验','headline':top_text,'bullets':['按本次实际评论文本统计'],'tone':'good','evidenceIds':[]},
      {'label':'当前明确问题','headline':(f"1 条评论明确出现{issue['label']}" if issue else '现有评论未出现规则可识别的明确问题'),'bullets':([concise_fact(neg_text,52)] if neg_text else []),'tone':'risk' if neg_id else 'normal','evidenceIds':[neg_id] if neg_id else []},
      {'label':'参数与信任','headline':f'已采 {len(attrs)} 项有效参数','bullets':[f"{x.get('name')}：{x.get('value')}" for x in attrs[:2]],'tone':'normal','evidenceIds':attr_refs[:3]},
      {'label':'下一款怎么开','headline':('保留已验证体验，先修正'+issue['label'] if issue else '保留已验证体验，进入家纺样品验收'),'bullets':['BOM、样品、标签、包装与页面口径必须一致'],'tone':'good','evidenceIds':([neg_id] if neg_id else [])+attr_refs[:2]},
    ]
    promotions=[x for x in raw.get('promotions',[]) or [] if isinstance(x,dict) and _txt(x.get('text')).strip()]
    if promotions:
        owner_cards.insert(1,{'label':'页面活动','headline':concise_fact(promotions[0].get('text'),46),'bullets':['报告只记录采集时页面展示内容'],'tone':'normal','evidenceIds':['PROMO_0001']})
    main_ids=[]
    for i,url in enumerate((raw.get('images',{}) or {}).get('main',[]) or [],1):
        if re.search(r'(?:avatar|sns_logo|shopmanager|tps-\d{1,3}-\d{1,3}|logo)',_txt(url),re.I):continue
        main_ids.append(f'IMG_MAIN_{i:04d}')
        if len(main_ids)>=3:break
    visual_items=[]
    visual_defs=[
      ('01 / 已核验主图','首图：看清商品全貌','主体、套件形态、花型面积与页面文字必须同屏核对','保留原图比例，优先检查商品是否完整、文案是否遮挡主体'),
      ('02 / 已核验主图','辅图：补充花型与搭配','观察配色、纹样密度、床品铺陈方式和儿童房场景信息','把可见花型、配色和套件组成拆成可复用拍摄清单'),
      ('03 / 已核验主图','辅图：承接规格信息','核对床型、件数、尺寸和页面参数是否一致','图上出现的规格表达必须与 SKU、参数表和详情页保持一致')
    ]
    for eid,defs in zip(main_ids,visual_defs):
        item_label,headline,task,action=defs
        visual_items.append({'label':item_label,'headline':headline,'explanation':'','imageEvidenceId':eid,'frameInfo':'按原图完整展示，重点看主体、花型、配色、场景、文案层级','persuasionTask':task,'weakness':'','action':action,'evidenceIds':[eid]})
    detail_id=''
    for i,url in enumerate((raw.get('images',{}) or {}).get('detail',[]) or [],1):
        if re.search(r'(?:avatar|sns_logo|shopmanager|tps-\d{1,3}-\d{1,3}|logo)',_txt(url),re.I):continue
        detail_id=f'IMG_DETAIL_{i:04d}';break
    if detail_id:
        visual_items.append({'label':'04 / 已核验详情图','headline':'商品详情页原图','explanation':'','imageEvidenceId':detail_id,'frameInfo':'完整展示','persuasionTask':'来源：商品详情描述容器','weakness':'','action':'图片文字与页面参数表逐项核对','evidenceIds':[detail_id]})
    positive=[]
    for outcome in review_outcomes[:5]:
        count=len(outcome.get('evidenceIds') or [])
        positive.append({'label':f"{outcome.get('label')}：{count}/{review_count} 条验证",'countOrNote':str(count),'evidenceIds':outcome.get('evidenceIds') or []})
    first_positive=(top[0] if top else {'topic':'有效评论','mentions':review_count})
    quote1,qid1=_review_ref(raw,lambda text:any(k in text for k in ('好看','漂亮','颜色','图案','花色')))
    quote2,qid2=_review_ref(raw,lambda text:any(k in text for k in ('舒服','柔软','亲肤','手感')))
    stages=[
      {'key':'see','stageNo':'01','label':'看到商品','headline':f"{first_positive.get('topic')}是当前最强评论信号",'userExpectation':'看清商品与花型','positiveConfirmation':f"{first_positive.get('mentions',0)} 条有效评论提及{first_positive.get('topic')}",'friction':'','businessImpact':'首图完整展示商品与花型','quotes':[{'type':'评论','text':quote1,'evidenceId':qid1}] if quote1 else [],'evidenceIds':[qid1] if qid1 else []},
      {'key':'choose','stageNo':'02','label':'选择规格','headline':'规格必须对应实际床型与套件 BOM','userExpectation':'一次看清床型、件数和各部件尺寸','positiveConfirmation':_txt(size_attr.get('value')) if size_attr else '','friction':(f"1 条评论明确出现{issue['label']}" if issue else ''),'businessImpact':'页面、样品与包装统一规格口径','quotes':[{'type':'评论','text':neg_text,'evidenceId':neg_id}] if neg_id else [],'evidenceIds':[x for x in (size_attr_id,neg_id) if x]},
      {'key':'use','stageNo':'03','label':'实际使用','headline':'触感反馈已形成明确正向信号','userExpectation':'柔软、亲肤、睡感舒适','positiveConfirmation':next((f"{x.get('mentions')} 条有效评论提及{x.get('topic')}" for x in topics if x.get('topic')=='面料/触感'),''),'friction':'','businessImpact':'材质页参数与评论原话并列证明','quotes':[{'type':'评论','text':quote2,'evidenceId':qid2}] if quote2 else [],'evidenceIds':[qid2] if qid2 else []},
    ]
    param_facts=_home_textile_parameter_facts(raw)
    material_value=_txt(material.get('value')) if material else ''
    size_value=_txt(size_attr.get('value')) if size_attr else ''
    plan_specs=[]
    if material_value:plan_specs.append({'label':'材质','value':material_value})
    if thread:plan_specs.append({'label':'面料支数','value':thread.get('value')})
    if size_value:plan_specs.append({'label':'SKU','value':size_value})
    if craft:plan_specs.append({'label':'工艺','value':craft.get('value')})
    return _apply_selling_points({
      'reportSummary':{'title':f'{label}产品分析报告','subtitle':'；'.join(sales_bits),'verdict':'保留已验证体验，优先核对尺寸、成本和页面一致性','verdictDetail':'只使用已采页面参数与实际评论。','monitorSummary':''},
      'platformOverview':build_platform_overview(raw,base),
      'ownerOverview':{'chapterTitle':f"{top_text}，{issue['label']+'需先修正' if issue else '进入家纺样品验收'}",'chapterDeck':'','thesis':'','sampleSummary':f'有效评论 {review_count} 条，购买问答 {question_count} 条','cards':owner_cards[:6]},
      'visualCommerce':{'chapterTitle':'完整展示商品，再依次解释花型、材质与规格','chapterDeck':'','items':visual_items},
      'customerExperience':{'chapterTitle':'，'.join(customer_results)+(f"；1 条明确出现{issue['label']}" if issue else ''),'chapterDeck':'','stats':[{'label':'有效评论','value':str(review_count),'note':''},{'label':'购买问答','value':str(question_count),'note':''},{'label':'有效参数','value':str(len(attrs)),'note':''}], 'journeyTitle':'从看到商品到首洗：只呈现评论已验证结果','journeyDeck':'','stages':stages,'buyerEvidence':{'title':'评论原图与对应评价','items':_buyer_evidence(raw)},'signals':{'title':'评论结果与购买前关注点','positive':positive,'questions':[{'label':f'采集到 {question_count} 条购买问答','countOrNote':str(question_count),'evidenceIds':[]} ] if question_count else [],'risks':[{'label':concise_fact(neg_text,44),'countOrNote':'1 条明确文本','evidenceIds':[neg_id]}] if neg_id else []}},
      'productExperience':{'chapterTitle':'家纺开品底板：BOM、材质、尺寸与首洗逐项过门','chapterDeck':'','parameterFacts':param_facts,'thesis':'','thesisDetail':'','gates':_home_textile_gates(raw,issue)},
      'newProductPlans':{'chapterTitle':('只做一个方向：保留已验证体验，先修正'+issue['label'] if issue else '只做一个方向：保留已验证体验，先完成家纺验收'),'chapterDeck':'','plans':[{'type':'主销款','name':f'{label}规格校准款','positioning':('沿用现有评论确认的花型与触感表达，优先修正'+issue['label'] if issue else '沿用现有评论确认的体验表达，先完成家纺样品验收'),'sellingPoints':_derive_next_selling_points(raw,base,issue),'specs':plan_specs,'manufacturingGate':{'title':'制造闸门','content':'核对套件 BOM、材质标签与各床型铺装；完成首洗、缝制填充、气味包装验收'},'visualCommerce':{'title':'首图与商详','content':'首图完整展示商品；后续图依次说明花型材质、套件清单、床型尺寸和洗护证据'},'evidenceIds':([neg_id] if neg_id else [])+attr_refs}], 'commercialSummary':{'title':'从当前商品保留什么','thesis':'只复用评论已确认的体验','keep':top_text,'fix':(issue['label'] if issue else '家纺样品与页面一致性'),'prove':'BOM、样品、标签、包装、页面五处一致'}},
      'validationLoop':{'chapterTitle':'六道家纺验收门决定上架、返修或停止','chapterDeck':'','rows':_home_textile_validation_rows(raw)}
    }, raw, base)

def concise_fact(value,limit=48):
    text=re.sub(r'\s+',' ',_txt(value)).strip()
    return text if len(text)<=limit else text[:limit].rstrip('，。； ')+'…'

REPORT_LANGUAGE_REPLACEMENTS=(
  ('客户为什么购买','评论已验证体验'),('用户为什么购买','评论已验证体验'),('用户为什么买','评论已验证体验'),
  ('成交原因','现有证据'),('决定成交','影响页面理解'),('推动选择','提供选择信息'),
  ('提升转化','验证页面效果'),('降低退货','验证售后结果'),('省心型大众爆款','当前高销量商品'),('爆款','当前商品'),
  ('建议立项验证','进入样品验证'),('有条件立项验证','先补齐验证项'),('暂不建议立项','暂停新增样品动作'),
  ('立项判断','产品分析结论'),('立项','样品验证'),('不成立','需修正'),('成立','已确认'),('诊断报告','产品分析报告'),('商品诊断','商品分析'),
  ('经营结论','产品分析结论'),('经营判断','产品分析结论'),('经营总览','产品摘要'),('老板决策总览','产品摘要'),
  ('经营与开品报告','产品分析报告'),('商品经营结论','商品分析结论'),('商品经营','商品产品'),
  ('建议进入样品验证','进入样品验证'),('再做量产决策','再安排生产测算'),('不做量产决策','不输出生产规模判断'),
  ('量产前','批量上架前'),('量产判断','生产规模判断'),('量产规模','生产规模'),('量产测算','生产测算'),
  ('有条件样品验证','样品校准分析'),
)

def _sanitize_report_language(value):
    if isinstance(value,dict):return {k:_sanitize_report_language(v) for k,v in value.items()}
    if isinstance(value,list):return [_sanitize_report_language(x) for x in value]
    if not isinstance(value,str):return value
    for source,target in REPORT_LANGUAGE_REPLACEMENTS:value=value.replace(source,target)
    return value

def _separate_qa_signals(exp):
    customer=exp.get('customerExperience') or {};signals=customer.get('signals') or {}
    questions=list(signals.get('questions') or []);risks=[]
    for item in signals.get('risks') or []:
        if not isinstance(item,dict):continue
        refs=[x for x in item.get('evidenceIds',[]) or [] if isinstance(x,str)]
        if refs and all(x.startswith('QA_') for x in refs):
            moved=dict(item);moved['label']='购买前关注：'+_txt(item.get('label')).removeprefix('购买前关注：')
            questions.append(moved)
        else:risks.append(item)
    signals['questions']=questions;signals['risks']=risks;customer['signals']=signals
    for stage in customer.get('stages') or []:
        if not isinstance(stage,dict):continue
        refs=[x for x in stage.get('evidenceIds',[]) or [] if isinstance(x,str)]
        quotes=[x for x in stage.get('quotes',[]) or [] if isinstance(x,dict)]
        only_questions=bool(refs) and all(x.startswith('QA_') for x in refs)
        if only_questions or (quotes and all(_txt(x.get('type')) in ('问答','购买问答') for x in quotes)):
            if stage.get('friction') and not stage.get('userExpectation'):stage['userExpectation']=stage['friction']
            stage['friction']=''
    return exp

def _filter_evidence_items(values, valid_ids, image=False):
    out=[]
    for item in values or []:
        if not isinstance(item,dict):continue
        refs=[x for x in item.get('evidenceIds',[]) or [] if x in valid_ids]
        image_id=item.get('imageEvidenceId') if image else ''
        if image_id in valid_ids and image_id not in refs:refs.append(image_id)
        if not refs:continue
        clean=dict(item);clean['evidenceIds']=refs;out.append(clean)
    return out

def _evidence_backed_solution(exp, valid_ids):
    exp=dict(exp or {})
    owner=exp.get('ownerOverview') or {};owner['cards']=_filter_evidence_items(owner.get('cards'),valid_ids);exp['ownerOverview']=owner
    visual=exp.get('visualCommerce') or {};visual['items']=_filter_evidence_items(visual.get('items'),valid_ids,True);exp['visualCommerce']=visual
    customer=exp.get('customerExperience') or {};customer['stages']=_filter_evidence_items(customer.get('stages'),valid_ids)
    signals=customer.get('signals') or {}
    for key in ('positive','questions','risks'):signals[key]=_filter_evidence_items(signals.get(key),valid_ids)
    customer['signals']=signals
    ci=customer.get('consumerInsights') or {}
    ci['experienceSignals']=_filter_evidence_items(ci.get('experienceSignals'),valid_ids)
    ci['productImplications']=_filter_evidence_items(ci.get('productImplications'),valid_ids)
    customer['consumerInsights']=ci;exp['customerExperience']=customer
    product=exp.get('productExperience') or {}
    product['parameterFacts']=_filter_evidence_items(product.get('parameterFacts'),valid_ids)
    product['gates']=_filter_evidence_items(product.get('gates'),valid_ids);exp['productExperience']=product
    plans=exp.get('newProductPlans') or {};plans['plans']=_filter_evidence_items(plans.get('plans'),valid_ids)[:3];exp['newProductPlans']=plans
    validation=exp.get('validationLoop') or {};rows=_filter_evidence_items(validation.get('rows'),valid_ids)
    validation['rows']=[x for x in rows if _txt(x.get('owner')).strip() and _txt(x.get('owner')).strip()!='项目负责人'];exp['validationLoop']=validation
    return exp

def ensure_experience_solution(analysis):
    raw=normalize(analysis.get('facts',{}) or {})
    analysis['facts']=raw
    analysis['baseline']=baseline(raw)
    analysis['evidenceLedger']=evidence_ledger(raw)
    was_fallback=bool((analysis.get('meta') or {}).get('fallbackReport'))
    analysis.setdefault('meta',{})['engineVersion']=ENGINE_VERSION
    route=detect_category(raw)
    analysis['meta']['industryTemplate']=route['key']
    analysis['meta']['categoryLabel']=route['label']
    analysis['meta']['categoryConfidence']=route['confidence']
    analysis['meta']['analysisDimensions']=route['dimensions']
    analysis['meta']['imageSourcePolicy']=(raw.get('images',{}).get('classification',{}) or {}).get('version') or 'legacy-review-only'
    valid_ids={x.get('id') for x in analysis['evidenceLedger'] if x.get('id')}
    exp=_evidence_backed_solution(analysis.get('experienceSolution') or {},valid_ids)
    if not _arr((exp.get('newProductPlans') or {}).get('plans')) and _arr((analysis.get('launchPlans') or {}).get('plans')):
        exp.setdefault('newProductPlans',{})['plans']=_filter_evidence_items((analysis.get('launchPlans') or {}).get('plans'),valid_ids)[:3]
        if (analysis.get('launchPlans') or {}).get('professionalOpinion'):
            exp['newProductPlans']['professionalOpinion']=(analysis.get('launchPlans') or {}).get('professionalOpinion')
    exp=_separate_qa_signals(exp)
    exp=_sanitize_report_language(_apply_selling_points(exp,raw,analysis['baseline']))
    analysis['experienceSolution']=exp
    analysis['meta'].pop('fallbackReport',None)
    analysis['meta']['fallbackReport']=False
    if was_fallback:
        analysis['meta']['reportReady']=False
        analysis['meta']['reportBlockedReason']='历史分析标记为兜底结果，本版本拒绝重新渲染为正式报告。'
        return analysis
    validation=core_solution_diagnostics(exp)
    analysis['meta']['reportValidation']=validation
    if not validation['ok']:
        # 审核是质量标记，不再阻断报告输出；缺失项在报告中显式提示，供人工复核。
        analysis['meta']['reportReady']=True
        analysis['meta']['reportAuditPassed']=False
        analysis['meta']['reportQuality']='needs_review'
        analysis['meta']['reportNotice']='审核未通过，报告仍已生成。请优先补齐：'+', '.join(validation['missing'])
        return analysis
    analysis['meta']['reportReady']=True
    analysis['meta']['reportAuditPassed']=True
    analysis['meta']['reportQuality']='passed'
    analysis['meta'].pop('reportBlockedReason',None)
    customer=exp.setdefault('customerExperience',{})
    customer['stats']=[{'label':'有效评论','value':str(analysis['baseline'].get('reviewCount',0))},{'label':'购买问答','value':str(analysis['baseline'].get('questionCount',0))},{'label':'有效参数','value':str(analysis['baseline'].get('attributeCount',0))}]
    customer['buyerEvidence']={'title':'评论原图与对应评价','items':_buyer_evidence(raw)}
    exp.pop('shopperDecision',None)
    plans=exp.setdefault('newProductPlans',{})
    summary=exp.setdefault('reportSummary',{})
    summary['title']=_clean_product_title(re.sub(r'(床品|商品)商品经营',r'\1产品',_txt(summary.get('title'))))
    analysis['experienceSolution']=exp
    return analysis

def evidence_ledger(raw):
    ev=[]
    def add(eid,etype,label,value,source='page',meta=None):
        if value in (None,'',[],{}): return
        ev.append({'id':eid,'type':etype,'label':label,'value':value,'source':source,'meta':meta or {}})
    p=raw.get('product',{}); s=raw.get('sales',{})
    for k,label in [('title','商品标题'),('shop','店铺'),('brand','品牌'),('category','类目'),('itemId','Item ID'),('skuId','SKU ID')]: add('P_'+k.upper(),'product',label,p.get(k),'page')
    for k,label in [('currentPrice','当前采集价格'),('sold','销量公开值'),('cumulativeSales','累计销量'),('monthlySales','月销量'),('ranking','榜单/排名')]: add('S_'+k.upper(),'sales',label,s.get(k),'page')
    for i,x in enumerate(raw.get('monitoring',[])[:30],1): add(f'MON_{i:03d}','monitoring','商品监测数据',x,'monitoring')
    for i,x in enumerate(raw.get('sku',[])[:500],1): add(f'SKU_{i:04d}','sku','SKU',x,'page')
    for i,x in enumerate(raw.get('attributes',[])[:500],1): add(f'ATTR_{i:04d}','attribute','页面属性',x,'page_claim')
    for i,x in enumerate(raw.get('promotions',[])[:300],1): add(f'PROMO_{i:04d}','promotion','促销/服务',x,'page_claim')
    for i,x in enumerate(raw.get('reviews',[]),1): add(f'REV_{i:06d}','review','消费者评价',x,'consumer')
    for i,x in enumerate(raw.get('questions',[])[:2000],1): add(f'QA_{i:05d}','question','购买问答',x,'consumer')
    images=raw.get('images',{}) or {};provenance=images.get('provenance',[]) or []
    for group in IMAGE_GROUPS:
        for i,u in enumerate(images.get(group,[])[:1000],1):
            records=[x for x in provenance if isinstance(x,dict) and x.get('group')==group and x.get('url')==u]
            source=records[0].get('source') if records else ''
            meta={'group':group,'source':source,'classification':(images.get('classification',{}) or {}).get('version')}
            if group=='buyerShow':
                meta['reviewEvidenceIds']=_uniq([x.get('reviewEvidenceId') for x in records if x.get('reviewEvidenceId')])
            add(f'IMG_{group.upper()}_{i:04d}','image',group,u,'verified_image',meta)
    return ev

def baseline(raw):
    rev=raw.get('reviews',[]); buckets={'good':0,'mid':0,'bad':0,'unknown':0}; topic=Counter(); examples=defaultdict(list); sku_topic=defaultdict(Counter)
    for x in rev:
        try:s=float(x.get('rating'))
        except:s=0
        if s>=4:buckets['good']+=1
        elif s==3:buckets['mid']+=1
        elif s>0:buckets['bad']+=1
        else:buckets['unknown']+=1
        c=x['content']; sku=x.get('sku') or '未识别SKU'
        for t,keys in TOPICS.items():
            if any(k in c for k in keys):
                topic[t]+=1; sku_topic[sku][t]+=1
                if len(examples[t])<8: examples[t].append(c[:220])
    collection=raw.get('collection') or {}
    return {'reviewCount':len(rev),'questionCount':len(raw.get('questions',[])),'ratingBuckets':buckets,
            'topics':[{'topic':k,'mentions':v,'sampleRate':round(v/max(1,len(rev)),4),'examples':examples[k][:6]} for k,v in topic.most_common(24)],
            'skuTopicSignals':[{'sku':sku,'topics':[{'topic':k,'mentions':v} for k,v in cnt.most_common(8)]} for sku,cnt in sorted(sku_topic.items(),key=lambda kv:sum(kv[1].values()),reverse=True)[:30]],
            'imageCounts':{k:len(raw.get('images',{}).get(k,[]) or []) for k in ['main','detail','sku','buyerShow','all']},
            'skuCount':len(raw.get('sku',[]) or []),'attributeCount':len(raw.get('attributes',[]) or []),'promotionCount':len(raw.get('promotions',[]) or []),
            'reviewCollectionComplete':bool(collection.get('reviewCollectionComplete')),'reviewStopReason':collection.get('reviewStopReason') or '',
            'publicReviewCount':collection.get('publicReviewCount') or '', 'reviewPagesVisited':collection.get('reviewPagesVisited') or 0}

def compact_evidence(ledger, allowed_types=None, limit=400):
    xs=[x for x in ledger if not allowed_types or x['type'] in allowed_types]
    return xs[:limit]

def require_refs(obj, valid):
    if isinstance(obj,dict):
        if isinstance(obj.get('evidenceIds'),list):
            obj['evidenceIds']=[x for x in obj['evidenceIds'] if x in valid]
            if not obj['evidenceIds']:
                obj['insufficientEvidence']=True
                obj.setdefault('missingEvidence',['未找到可核验引用'])
                try: obj['confidence']=min(float(obj.get('confidence',0.4)),0.45)
                except: obj['confidence']=0.35
        for v in obj.values(): require_refs(v,valid)
    elif isinstance(obj,list):
        for v in obj: require_refs(v,valid)
    return obj

def call(prompt,payload,max_tokens,images=None):
    # 统一模板入口：业务任务只提供 task 文本和变量值，接入其他后端时可替换 registry。
    try:
        task = build_task_prompt(prompt, tuple(payload.keys()) if isinstance(payload, dict) else ('evidence',))
        user = task+'\n'+STYLE_RULES+'\nACTUAL_EVIDENCE:\n'+json.dumps(payload,ensure_ascii=False)
        return chat_json(build_system_prompt(SYS), user, images=images, max_tokens=max_tokens),None
    except Exception as e:return {},str(e)

def batch_review_analysis(raw, valid, progress=lambda *a:None, enabled=True):
    reviews=raw.get('reviews',[])
    if not enabled or not reviews: return {'batches':[],'aggregate':{}}, None
    BATCH=160
    summaries=[]; errors=[]
    total=max(1,(len(reviews)+BATCH-1)//BATCH)
    for bi,start in enumerate(range(0,len(reviews),BATCH),1):
        chunk=[]
        for j,r in enumerate(reviews[start:start+BATCH],start+1):
            chunk.append({'evidenceId':f'REV_{j:06d}','content':r.get('content','')[:520],'appendContent':r.get('appendContent','')[:260],'rating':r.get('rating'),'sku':r.get('sku'),'date':r.get('date')})
        prompt='''任务｜评论批次研究。不要做泛泛总结，只抽取能影响开品的消费者事实。输出：
{"batchFindings":[{"finding":"","type":"purchase_driver|satisfaction|pain|expectation_gap|prepurchase_concern|sku_signal|usage_scene","businessMeaning":"","evidenceIds":[],"confidence":0}],"representativeQuotes":[{"text":"原文","evidenceId":"REV_000001","role":""}]}
要求：每条 finding 最多引用 8 个真实评论ID；不要估算全量次数。'''
        z,e=call(prompt,{'batchIndex':bi,'reviews':chunk},5200)
        if e: errors.append(f'batch {bi}: {e}')
        summaries.append(require_refs(z,valid))
        pct=78+int(8*bi/total)
        progress('review',f'processing {bi}/{total}',pct)
    prompt='''任务｜全量评论商业聚合。输入包含：程序对全部评论的统计 + 每批模型结论。请形成能直接影响商品定义、页面表达与下一款方案的消费者洞察，不得做评论罗列。每一项必须回答‘这说明什么、对商品意味着什么、下一款如何处理’；不得把批次出现次数当全量次数，只有 programStats 中的次数可以写为数量。
输出：
{
"consumerThesis":{"statement":"","businessMeaning":"","evidenceIds":[],"confidence":0},
"purchaseDrivers":[{"driver":"","whatUserIsBuying":"","commercialValue":"","productAction":"","evidenceIds":[],"confidence":0}],
"satisfactionDrivers":[{"driver":"","commercialValue":"","keepAction":"","evidenceIds":[],"confidence":0}],
"painPoints":[{"problem":"","severity":"high|medium|low","businessImpact":"","productAction":"","evidenceIds":[],"confidence":0}],
"expectationGaps":[{"expected":"","actual":"","commercialRisk":"","productOpportunity":"","evidenceIds":[],"confidence":0}],
"prePurchaseConcerns":[{"concern":"","conversionImpact":"","pageAction":"","evidenceIds":[],"confidence":0}],
"skuSignals":[{"sku":"","signal":"","decision":"","evidenceIds":[],"confidence":0}],
"verbatimQuotes":[{"text":"原文","evidenceId":"REV_000001","role":""}]
}'''
    payload={'programStats':baseline(raw),'batchSummaries':summaries}
    z,e=call(prompt,payload,9000)
    if e: errors.append('aggregate: '+e)
    return {'batches':summaries,'aggregate':require_refs(z,valid)}, '; '.join(errors) if errors else None

def question_analysis(raw, valid, enabled=True):
    qs=raw.get('questions',[])
    if not enabled or not qs:return {},None
    payload=[]
    for i,q in enumerate(qs[:1200],1): payload.append({'evidenceId':f'QA_{i:05d}','question':q.get('question','')[:700],'answer':q.get('answer','')[:700]})
    prompt='''任务｜购买前疑虑研究。这里全部是问大家/购买问答，不是差评。回答用户在下单前担心什么、页面哪里没有解释清楚、哪些疑虑被真实买后评论验证。不得把提问本身当产品缺陷。
输出：{"prePurchaseThesis":{"statement":"","commercialMeaning":"","evidenceIds":[],"confidence":0},"concerns":[{"concern":"","whyItBlocksPurchase":"","isProductRiskOrInfoGap":"product_risk|information_gap|unknown","commercialAction":"","evidenceIds":[],"confidence":0}],"validatedByReviews":[{"questionConcern":"","reviewReality":"","decision":"","evidenceIds":[],"confidence":0}],"unansweredQuestions":[{"question":"","pageFix":"","evidenceIds":[],"confidence":0}]}
只引用真实 QA_/REV_ ID。'''
    z,e=call(prompt,{'questions':payload,'reviewStats':baseline(raw)},7200)
    return require_refs(z,valid),e


def _evidence_backed_launch_fallback(raw, valid_ids):
    """V9.2: 只用已有证据构造最小新品方向，避免证据充足时整章为空。"""
    title=_txt((raw.get('product') or {}).get('title')).strip() or '当前商品'
    refs=[]
    for eid in ('P_TITLE','S_CURRENTPRICE','S_SOLD'):
        if eid in valid_ids:refs.append(eid)
    attr_refs=sorted([x for x in valid_ids if x.startswith('ATTR_')])[:4]
    review_refs=sorted([x for x in valid_ids if x.startswith('REV_')])[:2]
    image_refs=sorted([x for x in valid_ids if x.startswith('IMG_MAIN_')])[:2]
    refs=_uniq(refs+attr_refs+review_refs+image_refs)
    if not refs:return {'strategyMode':'mixed','professionalOpinion':'','plans':[]}
    changes=[]
    if attr_refs:changes.append({'action':'优先核对并统一当前商品的规格、材质与页面口径','evidenceIds':attr_refs[:2]})
    if review_refs:changes.append({'action':'把消费者已验证体验与差异反馈转成下一款产品选择说明','evidenceIds':review_refs})
    return {
      'strategyMode':'mixed',
      'professionalOpinion':'保留当前商品已有证据支持的产品逻辑，下一款优先把规格选择、体验表达和页面证明做得更清楚。',
      'plans':[{
        'type':'现有基础优化','name':_clean_product_title(title)+' 优化款',
        'whyThisPlan':'基于当前商品参数、图片与消费者反馈做有限优化，不引入无证据的新材料或市场结论。',
        'visualReferenceImageIds':image_refs,'productChanges':changes,
        'skuStrategy':'先按真实结构化SKU与参数建立选择层级；缺失项待补采后再细化。',
        'mainImagePlan':['首图保持商品全貌与核心场景','后续图片依次解释规格、材质/配置与证明信息'],
        'detailPagePlan':['按主图提出的信息逐项解释和证明','消费者反馈只作为购后体验证据'],
        'validationItems':['样品、SKU、参数、图片与页面文字口径一致'],
        'evidenceIds':refs,'confidence':0.55
      }]
    }

def analyze(raw, progress=lambda *a:None):
    raw=dict(raw or {}); model_available=raw.pop('_model_available',None)
    model_configured=configured(); use_ai=model_configured if model_available is None else bool(model_available and model_configured)
    raw=normalize(raw); base=baseline(raw); ledger=evidence_ledger(raw); valid={x['id'] for x in ledger}
    route=detect_category(raw)
    out={'meta':{'modelConfigured':model_configured,'modelUsed':use_ai,'engineVersion':ENGINE_VERSION,'roleContractVersion':ROLE_CONTRACT_VERSION,'roleContract':ROLE_CONTRACT,'actualOnly':True,'dynamicReport':True,'industryTemplate':route['key'],'categoryLabel':route['label'],'categoryConfidence':route['confidence'],'analysisDimensions':route['dimensions']},'facts':raw,'baseline':base,'evidenceLedger':ledger,'modelErrors':{}}

    progress('evidence','processing',28)
    limitations=[]
    if not base['reviewCollectionComplete']:
        limitations.append('评论采集未确认到达平台可访问末页；当前报告使用本次实际获取的全部评论，不宣称为平台全量。')
    if base['questionCount']==0: limitations.append('未采集到问大家，无法判断下单前主要疑虑。')
    if not raw.get('sales',{}).get('sold') and not raw.get('sales',{}).get('cumulativeSales'): limitations.append('未采集到销量公开值，不能判断经营表现。')
    out['dataQuality']={'grade':'A' if base['reviewCollectionComplete'] and base['reviewCount']>=100 and base['imageCounts']['main']>=3 else 'B+' if base['reviewCount']>=100 else 'B' if base['reviewCount']>=30 else 'C',
      'reviewSample':base['reviewCount'],'reviewCollectionComplete':base['reviewCollectionComplete'],'reviewStopReason':base['reviewStopReason'],
      'publicReviewCount':base['publicReviewCount'],'questionSample':base['questionCount'],'images':base['imageCounts'],'skuCount':base['skuCount'],'attributeCount':base['attributeCount'],'promotionCount':base['promotionCount'],'knownLimitations':limitations}
    progress('evidence','done',32)

    # 先做报告策划：决定这款商品最需要讲清楚什么，而不是先套章节。
    progress('planner','processing',34)
    if use_ai:
        prompt='''任务0｜报告策划。先判断“这款商品最需要被讲清楚的产品问题是什么”，再决定报告内容和顺序。输出：
{
"coreStory":"一句具体产品分析主线",
"decisionQuestion":"运营者最需要看清的一个产品问题",
"primaryQuestions":["3-6个具体问题"],
"sectionPlan":[{"module":"executive|facts|positioning|fashion|main_visual|title_promo|detail|consumer|qa|pain|sku|competition|opportunity|plans|gates","title":"必须是针对该商品的结果型标题","priority":100,"reason":"为什么需要出现"}],
"deemphasize":[{"module":"","reason":"为什么当前证据下不重点展开"}]
}
要求：executive/facts/plans/gates必须出现；其他模块按实际证据和商业价值动态排序。标题禁止使用空泛的“消费者分析”“视觉分析”。'''
        z,e=call(prompt,{'baseline':base,'facts':compact_evidence(ledger,{'product','sales','attribute','promotion','sku'},260)},5000)
        if e: out['modelErrors']['planner']=e
        out['reportPlan']=require_refs(z,valid)
    else: out['reportPlan']={}
    progress('planner','done',38)

    progress('product','processing',40)
    if use_ai:
        prompt='''任务A｜商品价值结构。输出具体到产品定义与运营动作：
{"positioning":{"statement":"","decisionImpact":"","evidenceIds":[],"confidence":0},"valuePillars":[{"name":"","customerValue":"","businessMeaning":"","mustKeep":"","evidenceIds":[],"confidence":0}],"claimVsProof":[{"claim":"","status":"consumer_supported|page_claim_only|conflicted|not_enough_evidence","commercialRisk":"","evidenceIds":[],"confidence":0}],"positioningRisks":[{"risk":"","businessImpact":"","action":"","evidenceIds":[],"confidence":0}]}'''
        z,e=call(prompt,compact_evidence(ledger,{'product','sales','sku','attribute','promotion','review'},320),6200)
        if e:out['modelErrors']['product']=e
        out['productDecision']=require_refs(z,valid)
    else: out['productDecision']={}
    progress('product','done',48)

    progress('fashion','processing',49)
    fashion_imgs=(raw.get('images',{}).get('main',[])[:6]+raw.get('images',{}).get('detail',[])[:6])
    if use_ai and fashion_imgs:
        prompt='''任务A2｜时尚与设计资产判断。只分析实际图片与页面文字中的色彩、花型/纹样、材质视觉、风格语言、细节工艺和人群审美表达。禁止把当前商品风格说成市场趋势，因为没有外部趋势数据。
输出：{"fashionThesis":{"statement":"","commercialMeaning":"","evidenceIds":[],"confidence":0},"designAssets":[{"asset":"","observation":"","commercialValue":"","developmentAction":"","evidenceIds":[],"confidence":0}],"styleSignals":[{"signal":"","whatItMeansForProduct":"","evidenceIds":[],"confidence":0}],"fashionRisks":[{"risk":"","why":"","action":"","evidenceIds":[],"confidence":0}],"trendBoundary":"没有外部趋势/竞品数据时不能判断流行度与生命周期"}'''
        z,e=call(prompt,{'evidence':compact_evidence(ledger,{'product','sku','attribute','image'},280)},7200,fashion_imgs)
        if e:out['modelErrors']['fashion']=e
        out['fashionDecision']=require_refs(z,valid)
    else: out['fashionDecision']={}
    progress('fashion','done',50)

    progress('visual','processing',51)
    main_imgs=raw.get('images',{}).get('main',[])[:8]
    if use_ai and main_imgs:
        prompt='''任务B｜主图商业信息提炼。目标不是解释图片，而是让专业用户一眼看到“这张图在卖什么”。逐张提炼图片中可核验的商品信息标签，再给一个短动作。
输出：
{"visualThesis":{"professionalOpinion":"<=45字","evidenceIds":[]},"coreClaims":[],"imageRoles":[{"imageEvidenceId":"IMG_MAIN_0001","imageRole":"<=8字","visualSignals":["2-3个，每项<=10字"],"businessMeaning":"<=22字","nextAction":"<=18字","evidenceIds":[],"confidence":0}],"evidenceDetail":[{"imageEvidenceId":"IMG_MAIN_0001","composition":"","textInfo":"","productSubject":"","patternColor":"","evidenceIds":[]}]}
visualSignals必须来自图片可见文字、商品、场景或已确认参数，例如“40支新疆棉/六件套/免费仓储”，不能写抽象评价。
'''
        z,e=call(prompt,{'evidence':compact_evidence(ledger,{'product','sales','image','attribute'},220)},6800,main_imgs)
        if e:out['modelErrors']['visual']=e
        out['visualDecision']=require_refs(z,valid)
    else: out['visualDecision']={}
    progress('visual','done',57)

    progress('merchandising','processing',58)
    if use_ai:
        prompt='''任务C｜标题表达压缩。只回答三件事：当前已经表达什么、还应强化什么、建议怎么改。不要输出解释型长段落。
输出：
{"titleAnalysis":{"professionalOpinion":"<=45字","currentExpression":[{"value":"<=10字","evidenceIds":[]}],"reinforce":[{"topic":"<=10字","evidenceIds":[]}],"recommendedActions":[{"action":"<=22字","evidenceIds":[]}],"recommendedTitles":[]},"promotionAnalysis":{}}
'''
        z,e=call(prompt,{'evidence':compact_evidence(ledger,{'product','sales','promotion','sku','attribute'},300)},7200)
        if e:out['modelErrors']['merchandising']=e
        out['merchandisingDecision']=require_refs(z,valid)
    else: out['merchandisingDecision']={}
    progress('merchandising','done',64)

    progress('detail','processing',65)
    detail_imgs=raw.get('images',{}).get('detail',[])[:20]
    if use_ai and detail_imgs:
        prompt='''任务D｜详情页证明信息提炼。按4-6个信息任务组选代表图，不按图片流水账。每组只提炼2-4个图片信息标签和1条专业动作。
输出：
{"detailThesis":{"professionalOpinion":"<=45字","evidenceIds":[]},"contentGroups":[{"type":"品牌服务|套件组成|材质品质|产品功能|使用便利|花型选择|用户验证|其他","representativeImageId":"IMG_DETAIL_0001","visualSignals":["2-3项，每项<=10字"],"businessMeaning":"<=22字","nextAction":"<=18字","imageEvidenceIds":[],"evidenceIds":[]}],"continuityChecks":[{"claim":"<=24字","mainImageEvidenceIds":[],"detailEvidenceIds":[],"status":"closed_loop|detail_missing|main_missing|conflict|review_pending","nextAction":"<=24字","evidenceIds":[]}],"pageSequence":[],"evidenceDetail":[]}
'''
        z,e=call(prompt,{'visualDecision':out.get('visualDecision',{}),'evidence':compact_evidence(ledger,{'product','attribute','promotion','image'},360)},8200,detail_imgs)
        if e:out['modelErrors']['detail']=e
        out['detailDecision']=require_refs(z,valid)
    else: out['detailDecision']={}
    progress('detail','done',72)

    progress('review','processing',74)
    rb,e=batch_review_analysis(raw,valid,progress,enabled=use_ai)
    if e: out['modelErrors']['review']=e
    out['reviewBatchAnalysis']=rb.get('batches',[])
    out['consumerResearch']=rb.get('aggregate',{})
    progress('review','done',86)

    progress('qa','processing',87)
    qz,qe=question_analysis(raw,valid,enabled=use_ai)
    if qe: out['modelErrors']['qa']=qe
    out['questionResearch']=qz or {}
    progress('qa','done',89)

    progress('competition','processing',90)
    out['competitionDecision']={'status':'missing_competitor_dataset','statement':'当前只有单品证据，不输出相对竞品的价格、设计、材质或营销优劣。','commercialMeaning':'单品报告只分析当前商品的页面表达、参数定义和购后反馈。','requiredCompetitorSet':['同核心搜索词','同目标场景','同主销规格','同可比价格带'],'compareDimensions':['当前价/主销价','销量/榜单公开值','标题槽位','主图内容结构','材质/工艺宣称','SKU结构','促销服务','评论痛点','详情证明方式']}
    progress('competition','done',91)

    progress('strategy','processing',92)
    if use_ai:
        context={'reportPlan':out.get('reportPlan',{}),'dataQuality':out['dataQuality'],'baseline':base,'productDecision':out.get('productDecision',{}),'visualDecision':out.get('visualDecision',{}),'merchandisingDecision':out.get('merchandisingDecision',{}),'detailDecision':out.get('detailDecision',{}),'consumerResearch':out.get('consumerResearch',{}),'questionResearch':out.get('questionResearch',{}),'fashionDecision':out.get('fashionDecision',{}),'competitionDecision':out.get('competitionDecision',{}),'evidence':compact_evidence(ledger,{'product','sales','sku','attribute','promotion','review','question'},420)}
        prompt='''任务E｜老板决策摘要。不要写大段商业分析，只形成可供最终编排器使用的三个答案：为什么能卖、值得继承、下一款优先。
输出：
{"executiveDecision":{"verdict":"<=45字","oneSentence":"<=55字","ifOnlyOneAction":"<=35字","evidenceIds":[]},"whyItSells":[{"mechanism":"<=20字","evidenceIds":[]}],"mustKeep":[{"item":"<=20字","evidenceIds":[]}],"mustFix":[{"item":"<=20字","fixAction":"<=28字","evidenceIds":[]}],"decisionGates":[]}
'''
        z,e=call(prompt,context,9800)
        if e:out['modelErrors']['strategy']=e
        out['commercialDecision']=require_refs(z,valid)
    else: out['commercialDecision']={}
    progress('strategy','done',94)

    progress('plans','processing',95)
    if use_ai:
        context={'reportPlan':out.get('reportPlan',{}),'commercialDecision':out.get('commercialDecision',{}),'consumerResearch':out.get('consumerResearch',{}),'questionResearch':out.get('questionResearch',{}),'fashionDecision':out.get('fashionDecision',{}),'competitionDecision':out.get('competitionDecision',{}),'productDecision':out.get('productDecision',{}),'visualDecision':out.get('visualDecision',{}),'merchandisingDecision':out.get('merchandisingDecision',{}),'detailDecision':out.get('detailDecision',{}),'baseline':base}
        prompt='''任务F｜下一款方案。面向老板先做决策卡，再让专业用户展开细节。每套默认只保留：来源等级、为什么做、产品动作、页面动作、验证项。
至少1套，证据充分2-3套。
来源等级：
- 证据驱动优化：直接来自当前商品页面/参数/评论；
- 反馈驱动升级：直接来自消费者体验差异/购前关注；
- 探索性方向：设计创新或扩展想法，必须明确标记，不能当作已验证结论。
输出：
{"professionalOpinion":"<=45字","plans":[{"sourceType":"证据驱动优化|反馈驱动升级|探索性方向","type":"现有基础优化|强化体验区分|强化设计选择|其他","name":"<=18字","whyThisPlan":"<=32字","visualReferenceImageIds":[],"productAction":"<=28字","pageAction":"<=28字","validation":"<=20字","evidenceIds":[],"confidence":0}],"nextValidation":[]}
'''
        z,e=call(prompt,context,9800)
        if e:out['modelErrors']['plans']=e
        out['launchPlans']=require_refs(z,valid)
        if not _arr((out.get('launchPlans') or {}).get('plans')):
            out['launchPlans']=_evidence_backed_launch_fallback(raw,valid)
    else: out['launchPlans']={}
    progress('plans','done',98)

    # 最终编辑层：把前面的事实与分析统一收敛为“全方位产品体验解决方案”。
    progress('editor','processing',98)
    if use_ai:
        context={
          'monitoring':raw.get('monitoring',[]), 'baseline':base, 'categoryRoute':route,
          'productDecision':out.get('productDecision',{}), 'fashionDecision':out.get('fashionDecision',{}),
          'visualDecision':out.get('visualDecision',{}), 'merchandisingDecision':out.get('merchandisingDecision',{}),
          'detailDecision':out.get('detailDecision',{}), 'consumerResearch':out.get('consumerResearch',{}),
          'questionResearch':out.get('questionResearch',{}), 'commercialDecision':out.get('commercialDecision',{}),
          'launchPlans':out.get('launchPlans',{}),
          'roleOutputs':{
            'productStrategy':out.get('productDecision',{}),
            'conversion':{'title':out.get('merchandisingDecision',{}),'visual':out.get('visualDecision',{}),'detail':out.get('detailDecision',{})},
            'consumerInsight':{'reviews':out.get('consumerResearch',{}),'questions':out.get('questionResearch',{})},
            'productSupply':{'product':out.get('productDecision',{}),'baseline':base},
            'visualDesign':out.get('fashionDecision',{}),
            'executiveDecision':out.get('commercialDecision',{}),
            'editorReview':{'scope':ROLE_CONTRACT['editorReview']['outputs']}
          },
          'evidence':compact_evidence(ledger,{'monitoring','product','sales','sku','attribute','promotion','review','question','image'},650)
        }
        prompt=FINAL_REPORT_PROMPT
        imgs=(raw.get('images',{}).get('main',[])[:8]+raw.get('images',{}).get('sku',[])[:8]+raw.get('images',{}).get('detail',[])[:16]+raw.get('images',{}).get('buyerShow',[])[:8])
        z,e=call(prompt,context,12000,imgs)
        if e: out['modelErrors']['experienceSolution']=e
        out['experienceSolution']=require_refs(z,valid)
    else:
        out['experienceSolution']={}
    # 最终编排模型不可用或被异常图片拒绝时，仍需用真实页面、参数、评论和问答生成完整本地分析，不能只输出最小结构。
    final_draft=out.get('experienceSolution') or {}
    if not isinstance(final_draft,dict) or not _txt((final_draft.get('reportSummary') or {}).get('title')).strip():
        out['experienceSolution']=build_fallback_experience(raw,base)
        out.setdefault('modelErrors',{})['experienceSolutionFallback']='最终编排未返回可用内容，已按真实采集证据生成本地完整分析。'

    # 将分角色产物显式挂载到结果，供报告样式、审计和后续导出复用；不复制或丢弃既有分析字段。
    out['roleOutputs']={
      'productStrategy':out.get('productDecision',{}),
      'conversion':{'title':out.get('merchandisingDecision',{}),'visual':out.get('visualDecision',{}),'detail':out.get('detailDecision',{})},
      'consumerInsight':{'reviews':out.get('consumerResearch',{}),'questions':out.get('questionResearch',{})},
      'productSupply':{'product':out.get('productDecision',{}),'sku':out.get('skuDecision',{}),'baseline':base},
      'visualDesign':out.get('fashionDecision',{}),
      'executiveDecision':out.get('commercialDecision',{}),
      'editorReview':{'status':'pending','scope':ROLE_CONTRACT['editorReview']['outputs']}
    }
    result=ensure_experience_solution(out)
    final_exp=result.get('experienceSolution') or {}
    role_outputs=result.setdefault('roleOutputs',{})
    # 以总编后的最终结构补齐角色视图，确保角色层与对外报告完全一致。
    role_outputs['conversion']={
      'title':final_exp.get('titleAnalysis') or role_outputs.get('conversion',{}).get('title',{}),
      'visual':final_exp.get('visualCommerce') or role_outputs.get('conversion',{}).get('visual',{}),
      'detail':final_exp.get('detailCommerce') or role_outputs.get('conversion',{}).get('detail',{})
    }
    role_outputs['consumerInsight']=final_exp.get('customerExperience') or role_outputs.get('consumerInsight',{})
    role_outputs['productSupply']={
      'sku':final_exp.get('skuAnalysis') or {},
      'product':final_exp.get('productExperience') or role_outputs.get('productSupply',{}).get('product',{}),
      'baseline':base
    }
    role_outputs['visualDesign']=role_outputs.get('visualDesign') or {}
    role_outputs['executiveDecision']=final_exp.get('ownerOverview') or role_outputs.get('executiveDecision',{})
    if use_ai and not result['meta'].get('reportReady'):
        first_validation=result['meta'].get('reportValidation') or core_solution_diagnostics(result.get('experienceSolution'))
        progress('editor','processing repair',99)
        repair_payload={
          'failedChecks':first_validation.get('checks',[]),
          'draft':result.get('experienceSolution') or {},
          'evidence':compact_evidence(ledger,{'monitoring','product','sales','sku','attribute','promotion','review','question','image'},650),
        }
        repaired,repair_error=call(FINAL_REPORT_REPAIR_PROMPT,repair_payload,12000,imgs)
        if repair_error:out['modelErrors']['experienceSolutionRepair']=repair_error
        if repaired:
            out['experienceSolution']=require_refs(repaired,valid)
            result=ensure_experience_solution(out)
        result['meta']['modelRepairAttempted']=True
        result['meta']['initialReportValidation']=first_validation
    if use_ai and not result['meta'].get('reportReady'):
        before=result['meta'].get('reportValidation') or core_solution_diagnostics(result.get('experienceSolution'))
        out['experienceSolution']=repair_core_solution_with_evidence(result.get('experienceSolution'),raw,base)
        out['meta'].update(result.get('meta') or {})
        out['meta']['evidenceStructuralRepairApplied']=True
        out['meta']['preStructuralRepairValidation']=before
        result=ensure_experience_solution(out)
    result.setdefault('roleOutputs',{})['editorReview']={
      'status':'passed' if result['meta'].get('reportReady') else 'needs_repair',
      'validation':result['meta'].get('reportValidation') or {},
      'scope':ROLE_CONTRACT['editorReview']['outputs']
    }
    result.setdefault('meta',{})['roleContractVersion']=ROLE_CONTRACT_VERSION
    result.setdefault('meta',{})['roleContract']=ROLE_CONTRACT
    progress('editor','done' if result['meta'].get('reportReady') else 'failed',99)
    return result