{ "meta": { "name": "Physical ↔ Cyber Atlas", "version": "0.2.0", "updated": "2026-11-03", "thesis": "每个物理系统都有一个赛博影子。两者之间真正缺的不是模型，而是可被核查的链接证据。", "unit": "链接点（Link Point）：物理侧的一个事实，如何被赛博侧表示，用什么证据证明它没错，最后由谁据此做决定。", "evidence_grades": { "A": "有公开标准或可复现实验支撑", "B": "公开工程实践，结论随场景与设备变化", "C": "行业自述或框架假设，尚未独立验证" }, "domains": { "physical-ai": "机器人、制造、具身智能", "longevity": "身体、健康寿命、生物医学证据", "cross": "两个域共用的链接机制" }, "layers": { "capture": "采集", "represent": "表示", "verify": "验证", "decide": "决策", "deploy": "运营" } }, "entries": [ { "id": "PA-001", "title": "尺度标定：像素世界 → 米制世界", "domain": "physical-ai", "layer": "represent", "physical": "场地的绝对尺寸、相机位姿、地面高程", "cyber": "具备 metric scale 的 USD / GLB 场景与相机内外参", "gap": "单目重建只给相对尺度；比例偏差不会被画面发现，却会让抓取高度与导航距离整体偏移", "evidence": "B", "evidence_note": "需要标定物或已知尺寸参照；精度随设备与场景变化，不存在固定承诺值", "buyer": "机器人团队的场景 / 仿真负责人", "decision": "这份场景资产能否直接用于训练或验收", "product": "Twin 场景交付的尺度核验项" }, { "id": "PA-002", "title": "传感器外参：谁在哪个坐标系", "domain": "physical-ai", "layer": "capture", "physical": "相机、激光雷达、机械臂基座之间的真实相对位姿", "cyber": "统一的 TF 树与标定参数文件", "gap": "外参错一个轴，感知与规划就各自活在不同的世界里，而单看任何一路都正常", "evidence": "B", "evidence_note": "标定方法成熟，但重装、碰撞、温度变化后需要复标；复标周期常被低估", "buyer": "现场集成与调试工程师", "decision": "这次部署的感知结果是否可信", "product": "现场标定核验清单" }, { "id": "PA-003", "title": "动态实体：人、叉车与临时堆放", "domain": "physical-ai", "layer": "represent", "physical": "场景中会移动的人、车辆与临时物体", "cyber": "带行为分布与出现频率的动态实体层", "gap": "静态场景再准，也训练不出对突发横穿的策略；动态层缺失会让真机首周暴露问题", "evidence": "B", "evidence_note": "动态建模依赖现场观察时长；短时拍摄容易低估低频高风险事件", "buyer": "仿真与安全评估负责人", "decision": "这个场景的风险覆盖是否足够", "product": "动态场景补采建议" }, { "id": "PA-004", "title": "任务协议：意图 → 可执行规格", "domain": "physical-ai", "layer": "decide", "physical": "客户口中的任务：把料箱从 A 搬到 B，别碰人", "cyber": "可判定的任务规格：成功条件、约束、失败判据、验收抽样", "gap": "自然语言任务无法自动判定成败，导致验收阶段各说各话", "evidence": "B", "evidence_note": "任务规格化在工业自动化中常见；跨场景复用仍需人工改写", "buyer": "客户方项目负责人", "decision": "这次交付是否算完成", "product": "任务规格工作坊模板" }, { "id": "PA-005", "title": "Sim-to-Real 差距：仿真里过了，现场不过", "domain": "physical-ai", "layer": "verify", "physical": "真机的摩擦、光照、反光、延迟与磨损", "cyber": "仿真参数与域随机化范围", "gap": "差距无法被一次性消除，只能被测量和记账；不记账就等于没看见", "evidence": "B", "evidence_note": "域随机化是公开实践，但参数选择高度依赖任务；没有通用最优解", "buyer": "算法与测试负责人", "decision": "仿真结果能否作为放行依据", "product": "Sim-to-Real 差距台账" }, { "id": "PA-006", "title": "失败分类学：同一种坏，说法不同", "domain": "physical-ai", "layer": "verify", "physical": "卡住、抓空、碰撞、超时、误识别", "cyber": "统一的失败码与根因标签", "gap": "不同团队对同一失败的描述不一致，导致修复无法积累，只会在不同项目重复踩", "evidence": "B", "evidence_note": "失败分类在制造业与软件业均有成熟范式，跨域映射仍需本地化", "buyer": "质量与运维负责人", "decision": "这次问题属于新问题还是旧问题复发", "product": "失败码表 v0" }, { "id": "PA-007", "title": "验收证据链：结论从哪来", "domain": "physical-ai", "layer": "decide", "physical": "一次真实运行及其现场条件", "cyber": "需求 → 运行 → 事件 → 失败 → 证据 → 缺口 → 验收结论", "gap": "结论往往只存在于会议记录里，无法回溯当时到底跑了什么", "evidence": "B", "evidence_note": "证据链本身是方法而非产品；落地需要与客户既有验收流程对齐", "buyer": "采购与验收方", "decision": "Ready / Needs Work / Blocked", "product": "Twin 验收证据包" }, { "id": "PA-008", "title": "换型追溯：产线一变，结论作废", "domain": "physical-ai", "layer": "deploy", "physical": "产线布局、物料、节拍、人员的变更", "cyber": "场景版本与变更影响面", "gap": "换型后沿用旧验收结论，是工业 AI 最常见的隐性事故来源", "evidence": "B", "evidence_note": "变更管理在制造体系中有标准做法；与 AI 模型的耦合影响尚无通用标准", "buyer": "产线负责人", "decision": "换型后要不要重新验证", "product": "变更影响评估表" }, { "id": "PA-009", "title": "现场数据血缘：这段视频能不能用", "domain": "physical-ai", "layer": "capture", "physical": "拍摄时间、地点、设备、授权范围", "cyber": "带来源与权限标记的数据集", "gap": "数据一旦脱离来源就无法证明可用，客户也不敢把产线数据交出去", "evidence": "B", "evidence_note": "数据血缘在数据工程中成熟；工业现场的授权与脱敏实践差异很大", "buyer": "客户的 IT / 法务 / 数据 owner", "decision": "这批数据能否进入处理流程", "product": "数据交接单" }, { "id": "PA-010", "title": "安全包络：什么情况下必须停", "domain": "physical-ai", "layer": "decide", "physical": "人机距离、速度上限、急停可达性", "cyber": "可计算的安全约束与降级策略", "gap": "安全常在最后才被加进来，导致策略与硬件都要返工", "evidence": "A", "evidence_note": "工业机器人与人机协作有公开安全标准可引用；具体条文需按行业与地区核对", "buyer": "安全与合规负责人", "decision": "这套方案能否进入现场", "product": "安全包络检查表" }, { "id": "PA-011", "title": "实时性预算：每一毫秒花在哪", "domain": "physical-ai", "layer": "deploy", "physical": "感知、决策、执行的端到端时延", "cyber": "各环节的时延预算与实测分布", "gap": "平均值达标不代表安全，尾部延迟才决定会不会撞上", "evidence": "B", "evidence_note": "时延测量是成熟工程，但现场网络与调度会改变分布", "buyer": "系统架构师", "decision": "当前配置能否满足节拍与安全要求", "product": "时延预算表" }, { "id": "PA-012", "title": "多机协同：一致的世界观", "domain": "physical-ai", "layer": "deploy", "physical": "多台设备的相对位置与时间同步", "cyber": "共享地图、时钟与任务分配状态", "gap": "每台设备各自正确，合起来却会互相挡住；一致性没有单机指标", "evidence": "B", "evidence_note": "多机调度有公开研究基础；实际可靠性依赖现场网络与时间同步质量", "buyer": "系统集成商", "decision": "这套多机方案能否上线", "product": "多机一致性检查项" }, { "id": "LG-001", "title": "生物标志物 → claim：一个指标能说明什么", "domain": "longevity", "layer": "represent", "physical": "一次采血、一次测量得到的数值", "cyber": "带人群、方法与不确定性的 claim 陈述", "gap": "指标变化常被直接翻译成健康结论，跳过了人群、剂量与时间尺度", "evidence": "A", "evidence_note": "循证医学有成熟的证据分级体系可引用；跨 claim 的自动映射仍是开放问题", "buyer": "品牌方的科学 / 法规负责人", "decision": "这句话能不能对外说", "product": "Longevity Atlas claim screen" }, { "id": "LG-002", "title": "剂量-反应：多少才算有用", "domain": "longevity", "layer": "verify", "physical": "真实摄入量与暴露时间", "cyber": "剂量区间与效应曲线", "gap": "研究里的剂量和产品里的剂量常常不是一个数量级", "evidence": "A", "evidence_note": "剂量-反应是药理学与营养学的基本方法；具体物质差异极大", "buyer": "产品研发负责人", "decision": "配方剂量是否有依据", "product": "剂量对照表" }, { "id": "LG-003", "title": "研究质量分级：这篇能不能信", "domain": "longevity", "layer": "verify", "physical": "一项研究的真实执行质量", "cyber": "样本量、对照、盲法、预注册、利益冲突的结构化评分", "gap": "标题相似的论文，可信度可能差一个数量级", "evidence": "A", "evidence_note": "偏倚风险评估工具有公开范式；评分仍需人工判断", "buyer": "科学顾问与内容团队", "decision": "这条证据能否进入结论", "product": "证据分级卡" }, { "id": "LG-004", "title": "人群适用性：对谁有效", "domain": "longevity", "layer": "decide", "physical": "研究人群与目标用户的身体条件差异", "cyber": "可迁移性判断与外推边界", "gap": "在运动员身上有效，不等于对 55 岁久坐人群有效", "evidence": "B", "evidence_note": "外推性判断依赖临床经验；自动判断容易过度自信", "buyer": "品牌方与临床顾问", "decision": "这条 claim 能否用于目标人群", "product": "适用人群标注" }, { "id": "LG-005", "title": "混杂因素：相关不是因果", "domain": "longevity", "layer": "verify", "physical": "收入、运动、睡眠、用药等共同变化的因素", "cyber": "调整后的效应估计与敏感性分析", "gap": "观察性研究里的相关性被当成因果，是长寿营销最常见的错误", "evidence": "A", "evidence_note": "因果推断方法公开成熟；数据质量决定结论上限", "buyer": "科学审阅人", "decision": "这条结论能说到多强", "product": "因果强度标注" }, { "id": "LG-006", "title": "终点效度：测的东西重要吗", "domain": "longevity", "layer": "represent", "physical": "真实的健康寿命与功能状态", "cyber": "替代终点：某指标、某评分、某生物年龄", "gap": "替代终点改善不等于活得更好，中间可能隔着很长的因果链", "evidence": "B", "evidence_note": "替代终点在临床研究中被广泛讨论；不同终点的效度差异很大", "buyer": "研究与产品负责人", "decision": "这个结果值不值得对外讲", "product": "终点效度说明" }, { "id": "LG-007", "title": "补充剂身份与纯度：瓶子里是什么", "domain": "longevity", "layer": "capture", "physical": "原料来源、批次、杂质与标示量", "cyber": "可追溯的批号与检测报告", "gap": "同名成分不同来源，实际效果与安全性可能完全不同", "evidence": "B", "evidence_note": "第三方检测是公开实践；报告覆盖率与抽检比例因品类而异", "buyer": "供应链与质量负责人", "decision": "这个批次能否放行", "product": "成分溯源卡" }, { "id": "LG-008", "title": "纵向追踪：一次测量不算数", "domain": "longevity", "layer": "deploy", "physical": "同一个人随时间变化的真实状态", "cyber": "带测量误差的个体时间序列", "gap": "个体波动常大于干预效应，单点对比极易得出错误结论", "evidence": "B", "evidence_note": "个体监测依赖测量一致性与足够采样；消费级设备误差需单独评估", "buyer": "用户与健康顾问", "decision": "这次干预有没有用", "product": "纵向基线模板" }, { "id": "LG-009", "title": "相互作用风险：叠加会不会出事", "domain": "longevity", "layer": "decide", "physical": "多种成分与既有用药的共同作用", "cyber": "相互作用规则与风险等级", "gap": "单品安全不等于组合安全，而组合正是用户的实际用法", "evidence": "B", "evidence_note": "药物相互作用数据库存在；保健品组合的证据覆盖显著不足", "buyer": "品牌方与执业医师", "decision": "这个组合要不要提示风险", "product": "组合风险标注" }, { "id": "LG-010", "title": "可复现性：别人做不出来怎么办", "domain": "longevity", "layer": "verify", "physical": "独立团队重复同一干预", "cyber": "复现结果与原始结果的差异记录", "gap": "单一研究被当作定论，复现失败却很少被写进宣传", "evidence": "A", "evidence_note": "复现性是科学方法的基本要求；营养与运动领域的复现率本身是公开议题", "buyer": "科学顾问", "decision": "这条证据的置信度上限", "product": "复现状态标记" }, { "id": "CX-001", "title": "证据链结构：两个域共用的骨架", "domain": "cross", "layer": "verify", "physical": "一次真实发生的事件", "cyber": "需求 → 运行 → 事件 → 证据 → 缺口 → 结论", "gap": "机器人验收和健康 claim 看似无关，缺的是同一套可核查结构", "evidence": "B", "evidence_note": "该结构为本数据库的组织假设，尚未在两类客户中同时验证", "buyer": "质量、法规、验收负责人", "decision": "结论是否可被第三方复核", "product": "Atlas 核心 schema" }, { "id": "CX-002", "title": "数据血缘与版本：结论属于哪一版", "domain": "cross", "layer": "capture", "physical": "数据产生时的真实条件", "cyber": "来源、时间、处理链、版本号", "gap": "模型与数据都在变，结论却不带版本，导致无法判断何时失效", "evidence": "B", "evidence_note": "数据版本化在工程界成熟；跨组织共享仍缺统一约定", "buyer": "数据 owner 与审计方", "decision": "这个结论现在还有效吗", "product": "版本化交付规范" }, { "id": "CX-003", "title": "不确定性量化：别只给一个数", "domain": "cross", "layer": "represent", "physical": "真实世界的随机性与测量误差", "cyber": "区间、分布与置信水平", "gap": "单点估计看起来干净，却让人误以为确定", "evidence": "B", "evidence_note": "不确定性量化方法成熟；在商业沟通中常被省略", "buyer": "决策者", "decision": "这个结果够不够稳", "product": "区间输出规范" }, { "id": "CX-004", "title": "基准套件：拿什么比", "domain": "cross", "layer": "verify", "physical": "真实任务的分布", "cyber": "固定任务集与评分规则", "gap": "没有共同基准，任何“更好”都无法比较", "evidence": "B", "evidence_note": "机器人领域已有公开基准；与本数据库的链接点尚未对齐", "buyer": "技术评估方", "decision": "这个方案是否值得进入下一轮", "product": "链接点基准集" }, { "id": "CX-005", "title": "决策阈值：什么时候算通过", "domain": "cross", "layer": "decide", "physical": "真实风险的容忍度", "cyber": "可执行的通过 / 不通过规则", "gap": "阈值往往在项目后期由情绪决定，而不是由风险与成本决定", "evidence": "B", "evidence_note": "阈值设定需要客户参与；本库只提供结构，不预设数值", "buyer": "客户决策人", "decision": "Ready / Needs Work / Blocked", "product": "阈值工作坊" }, { "id": "CX-006", "title": "模型漂移：世界变了，模型没变", "domain": "cross", "layer": "deploy", "physical": "环境、人群、物料、流程的持续变化", "cyber": "漂移监测与再验证触发条件", "gap": "上线即终点的心态，让很多系统在半年后悄悄失效", "evidence": "B", "evidence_note": "漂移监测在机器学习运维中成熟；触发再验证的阈值需按场景设定", "buyer": "运维与质量负责人", "decision": "什么时候必须重新验证", "product": "漂移触发清单" }, { "id": "CX-007", "title": "合成数据合法性：能不能拿来训练", "domain": "cross", "layer": "capture", "physical": "真实数据的采集授权边界", "cyber": "合成数据与真实数据的配比与用途声明", "gap": "合成数据好用，但用来证明真实世界结论时必须有边界说明", "evidence": "B", "evidence_note": "合成数据在感知训练中常见；作为验收证据的接受度仍在形成中", "buyer": "客户与合规方", "decision": "这份证据能否用于放行", "product": "合成数据使用声明" }, { "id": "CX-008", "title": "责任归属：出错时谁签字", "domain": "cross", "layer": "decide", "physical": "事故的真实后果", "cyber": "责任矩阵与签核记录", "gap": "模型、集成商、客户三方都能说自己没错，因为没有人被事先指派", "evidence": "C", "evidence_note": "责任划分依赖合同与行业惯例；本条目为框架假设", "buyer": "法务与项目负责人", "decision": "风险由谁承担", "product": "责任矩阵模板" }, { "id": "PA-013", "title": "光照与反光：换个角度就像换了个物体", "domain": "physical-ai", "layer": "represent", "physical": "现场光照强度、色温、金属与塑料表面的镜面反射", "cyber": "材质 PBR 参数与光照配置", "gap": "反光在仿真里被简化，真机上却会让视觉模型直接失效；同一物体换个角度就像换了一个", "evidence": "B", "evidence_note": "PBR 是公开标准；现场光照测量在项目里常被跳过", "buyer": "感知算法负责人", "decision": "这套视觉方案在现场光线下还能不能用", "product": "光照条件清单" }, { "id": "PA-014", "title": "末端执行器：夹爪不是一个抽象的点", "domain": "physical-ai", "layer": "represent", "physical": "夹爪行程、夹持力、指面材质与磨损", "cyber": "末端执行器的几何与接触模型", "gap": "仿真把夹爪当刚体点，真机是柔性接触；抓取策略在两者之间不通用", "evidence": "B", "evidence_note": "接触建模方法公开；柔性体与磨损的准确建模成本高", "buyer": "集成与调试工程师", "decision": "这个抓取策略能否直接迁移到现场", "product": "末端建模清单" }, { "id": "PA-015", "title": "力与触觉：看不见的那部分反馈", "domain": "physical-ai", "layer": "capture", "physical": "接触力、滑移、材料形变", "cyber": "力 / 触觉通道与时序数据", "gap": "视觉方案在接触瞬间没有信息，插孔、拧紧、软物抓取都会失手", "evidence": "B", "evidence_note": "触觉传感在装配场景有成熟应用；成本与标定仍是门槛", "buyer": "工艺工程师", "decision": "这类装配任务能否只靠视觉", "product": "触觉需求评估" }, { "id": "PA-016", "title": "时间同步：谁的时间才是真的", "domain": "physical-ai", "layer": "capture", "physical": "多传感器的真实采样时刻", "cyber": "统一时钟与时间戳对齐", "gap": "毫秒级错位在静态标定里看不出来，在高速运动时直接产生伪影", "evidence": "B", "evidence_note": "时间同步有公开工程方案；现场网络抖动会让标称精度失效", "buyer": "数采与标定工程师", "decision": "这批多传感器数据能否融合", "product": "时间同步检查项" }, { "id": "PA-017", "title": "遮挡与盲区：没看见不等于不存在", "domain": "physical-ai", "layer": "represent", "physical": "货架背后、设备阴影里、相机视野之外", "cyber": "盲区地图与置信度", "gap": "模型在没看见的地方会自信地猜，而不是承认不知道", "evidence": "B", "evidence_note": "占据栅格与置信度地图有公开方法；工程落地依赖传感器布置", "buyer": "安全与感知负责人", "decision": "这个部署的盲区风险是否可接受", "product": "盲区标注规范" }, { "id": "PA-018", "title": "标注一致性：两个人标的是不是同一件事", "domain": "physical-ai", "layer": "capture", "physical": "真实物体的边界与类别", "cyber": "标注规范、争议样本与一致性指标", "gap": "标注分歧被平均掉，模型学到的是折中而不是事实", "evidence": "B", "evidence_note": "一致性指标是公开方法；争议样本的裁决机制更影响结果", "buyer": "数据负责人", "decision": "这批标注能否用于训练和评估", "product": "标注规范模板" }, { "id": "PA-019", "title": "材质与惯量：仿真为什么发飘", "domain": "physical-ai", "layer": "represent", "physical": "质量、摩擦系数、质心位置", "cyber": "刚体参数与接触参数", "gap": "参数靠猜，仿真结论就只是参数的回声", "evidence": "B", "evidence_note": "参数辨识有公开方法；辨识所需的激励动作常在现场无法执行", "buyer": "仿真工程师", "decision": "仿真结果里有多少来自真实物理", "product": "参数标定表" }, { "id": "PA-020", "title": "运动学链：图纸和实机差多少", "domain": "physical-ai", "layer": "represent", "physical": "连杆长度、关节限位、装配误差", "cyber": "URDF 或等效运动学模型", "gap": "图纸参数与实机装配存在偏差，碰撞检测因此失真", "evidence": "B", "evidence_note": "运动学校准方法成熟；装配误差需实测而非取标称值", "buyer": "设备商与集成商", "decision": "这套运动学模型能否用于规划", "product": "模型核验项" }, { "id": "PA-021", "title": "可通行语义：地图上空的不等于能走", "domain": "physical-ai", "layer": "represent", "physical": "地面材质、坡度、门槛、临时障碍", "cyber": "可通行区域与代价地图", "gap": "几何上空的，物理上过不去；语义层缺失导致规划出不可执行的路径", "evidence": "B", "evidence_note": "语义地图有公开研究基础；代价函数需按设备能力标定", "buyer": "导航算法负责人", "decision": "这条路径在现场能否执行", "product": "可通行性标注" }, { "id": "PA-022", "title": "仓位与库存：世界是活的", "domain": "physical-ai", "layer": "capture", "physical": "货物位置、数量、SKU 变化", "cyber": "库存状态与场景实例", "gap": "场景按某一时刻建模，几小时后货就挪了；验收用的场景可能已经不存在", "evidence": "B", "evidence_note": "库存系统可提供状态，但与仿真场景的同步常缺失", "buyer": "仓储运营负责人", "decision": "这次测试结果是否代表现状", "product": "场景时效声明" }, { "id": "PA-023", "title": "人体姿态与工效：人也是系统的一部分", "domain": "physical-ai", "layer": "represent", "physical": "工人的身高、动作幅度、搬运姿势", "cyber": "人体模型与可达域", "gap": "只算机器人的可达域，现场就会互相挡路，或把人逼到工效边界之外", "evidence": "B", "evidence_note": "人因工程有公开标准；个体差异需要现场抽样", "buyer": "工艺与安全负责人", "decision": "这个人机工位能否成立", "product": "人体工效检查表" }, { "id": "PA-024", "title": "节拍模型：单机快不等于整线快", "domain": "physical-ai", "layer": "verify", "physical": "真实循环时间与等待", "cyber": "节拍仿真与瓶颈定位", "gap": "单机时间达标不等于整线节拍达标，排队与等待常被忽略", "evidence": "B", "evidence_note": "离散事件仿真方法成熟；输入分布需现场实测", "buyer": "产线规划负责人", "decision": "这套方案能否满足产能", "product": "节拍评估表" }, { "id": "PA-025", "title": "坐标系约定：Z 朝上还是 Y 朝上", "domain": "physical-ai", "layer": "represent", "physical": "真实世界的重力方向", "cyber": "坐标系约定与重力轴", "gap": "不同工具链默认不同，导入后整个场景躺倒或翻面，而文件本身没有报错", "evidence": "A", "evidence_note": "主流格式的坐标约定是公开的；跨工具链仍需显式约定", "buyer": "3D 与仿真工程师", "decision": "资产能否直接导入下游", "product": "导入检查清单" }, { "id": "PA-026", "title": "单位与精度声明：数字背后是什么", "domain": "physical-ai", "layer": "represent", "physical": "米、毫米、英寸与真实公差", "cyber": "单位元数据与有效位数", "gap": "单位错了不会报错，精度虚标也不会被自动发现", "evidence": "B", "evidence_note": "单位元数据可强制声明；精度必须实测", "buyer": "交付双方的技术接口人", "decision": "这份资产能否被下游直接使用", "product": "资产规格卡" }, { "id": "PA-027", "title": "感知精度：mAP 解释不了现场", "domain": "physical-ai", "layer": "verify", "physical": "现场的目标尺寸、距离与遮挡", "cyber": "分类、检测、分割指标", "gap": "数据集指标好，不代表工作距离上的漏检率可接受", "evidence": "A", "evidence_note": "指标定义公开；与业务风险之间的映射需按场景建立", "buyer": "感知负责人", "decision": "这个模型能否上现场", "product": "分场景精度报告" }, { "id": "PA-028", "title": "定位漂移：跑两小时之后", "domain": "physical-ai", "layer": "verify", "physical": "长时间运行后的真实位置", "cyber": "里程计与重定位误差", "gap": "短测很准，连续运行后地图和现实对不上，且不会主动报警", "evidence": "B", "evidence_note": "漂移评估方法公开；阈值需按任务精度要求设定", "buyer": "运维负责人", "decision": "是否需要重定位机制", "product": "漂移测试方案" }, { "id": "PA-029", "title": "抓取成功率：一个数字藏住一场灾难", "domain": "physical-ai", "layer": "verify", "physical": "不同物体、姿态、光照下的真实成功率", "cyber": "分层统计与置信区间", "gap": "总体成功率很好，但某个 SKU 可能几乎全错", "evidence": "A", "evidence_note": "分层统计是标准做法；分层维度需与客户业务对齐", "buyer": "客户验收方", "decision": "这个指标是否足以放行", "product": "分层验收表" }, { "id": "PA-030", "title": "长尾覆盖：测试越跑越顺的原因", "domain": "physical-ai", "layer": "verify", "physical": "低频但高风险的现场事件", "cyber": "场景覆盖率与未覆盖清单", "gap": "测试集越来越顺，是因为难例从来没被采到，而不是系统变强了", "evidence": "B", "evidence_note": "覆盖率度量有公开方法；长尾事件的发现依赖现场观察", "buyer": "测试负责人", "decision": "覆盖是否足够进入下一阶段", "product": "长尾清单" }, { "id": "PA-031", "title": "回归测试：修好一个，弄坏两个", "domain": "physical-ai", "layer": "verify", "physical": "每次改版后的真实表现", "cyber": "固定回归集与基线指标", "gap": "没有回归集，改动的副作用要等到客户现场才被发现", "evidence": "B", "evidence_note": "回归测试在软件工程中成熟；机器人场景的成本主要在数据维护", "buyer": "研发负责人", "decision": "这个版本能否发布", "product": "回归集规范" }, { "id": "PA-032", "title": "训练测试污染：指标为什么虚高", "domain": "physical-ai", "layer": "verify", "physical": "真实的新场景", "cyber": "训练集与评估集的隔离记录", "gap": "同一场地、同一批数据既训练又评估，指标必然好看", "evidence": "A", "evidence_note": "数据隔离是公开的基本要求；实践中常因数据少而被打破", "buyer": "技术评估方", "decision": "这个指标可不可信", "product": "数据隔离检查" }, { "id": "PA-033", "title": "验收抽样：抽多少、抽哪些、谁在场", "domain": "physical-ai", "layer": "decide", "physical": "真实运行中的样本分布", "cyber": "抽样规则与判定阈值", "gap": "抽样方案常在验收当天临时决定，结果无法复核", "evidence": "B", "evidence_note": "抽样方法有公开统计基础；样本量取决于可接受的漏判风险", "buyer": "采购与验收方", "decision": "这次抽检能否代表整体", "product": "抽样方案模板" }, { "id": "PA-034", "title": "行业标准差异：一套标准套不了所有现场", "domain": "physical-ai", "layer": "decide", "physical": "汽车、3C、仓储各自的现场约束", "cyber": "可配置的通过标准", "gap": "标准过严卖不动，过松则出事；差异不在条文，而在风险容忍度", "evidence": "B", "evidence_note": "各行业有公开规范可引用；跨行业映射需逐条核对", "buyer": "解决方案负责人", "decision": "这个客户该用哪套标准", "product": "行业标准映射表" }, { "id": "PA-035", "title": "ROI 归因：收益到底是谁带来的", "domain": "physical-ai", "layer": "decide", "physical": "实际节省的人工、良率与停机时间", "cyber": "收益归因模型", "gap": "客户倾向于把整体效率提升都算给机器人，真正贡献无法拆开", "evidence": "B", "evidence_note": "归因需要对照基线；现场很少保留干净的基线数据", "buyer": "客户财务与运营", "decision": "这笔投资值不值得继续", "product": "ROI 测算表" }, { "id": "PA-036", "title": "模型与数据归属：训练出来的东西是谁的", "domain": "physical-ai", "layer": "decide", "physical": "客户场地的真实数据", "cyber": "训练出的模型、权重与衍生数据", "gap": "合同常只写交付物，不写模型归属，二次合作时才发现卡住", "evidence": "C", "evidence_note": "归属由合同决定；本条为常见争议点的框架归纳", "buyer": "客户法务与采购", "decision": "这个模型能否复用到其他客户", "product": "归属条款清单" }, { "id": "PA-037", "title": "上线灰度：一次全量就没有退路", "domain": "physical-ai", "layer": "deploy", "physical": "真实现场的渐进放量", "cyber": "灰度策略与回滚条件", "gap": "直接全量上线，出问题时既没有对照，也没有台阶下", "evidence": "B", "evidence_note": "灰度发布在软件工程中成熟；物理现场的并行运行成本更高", "buyer": "现场运维负责人", "decision": "什么时候扩大使用范围", "product": "灰度上线计划" }, { "id": "PA-038", "title": "告警阈值：天天误报等于没有告警", "domain": "physical-ai", "layer": "deploy", "physical": "真实故障发生前的征兆", "cyber": "告警规则与阈值", "gap": "阈值拍脑袋，要么长期误报被忽略，要么真出事不响", "evidence": "B", "evidence_note": "阈值需用历史数据标定；故障前的有效征兆往往稀缺", "buyer": "运维负责人", "decision": "这条告警要不要处理", "product": "告警阈值基线" }, { "id": "PA-039", "title": "人员培训与接管：什么时候该人来", "domain": "physical-ai", "layer": "deploy", "physical": "现场人员的实际操作能力", "cyber": "操作手册与接管流程", "gap": "系统越自动，人越不知道异常时该怎么接管", "evidence": "B", "evidence_note": "人机交接有公开研究；现场演练是有效性的关键", "buyer": "客户现场主管", "decision": "现场人员是否具备接管能力", "product": "培训与接管清单" }, { "id": "PA-040", "title": "网络与边缘算力：演示环境不是工厂", "domain": "physical-ai", "layer": "deploy", "physical": "现场网络覆盖、延迟与断网", "cyber": "边缘推理部署与带宽预算", "gap": "演示时一切正常，工厂网络一抖就全线停摆", "evidence": "B", "evidence_note": "边缘部署方案成熟；现场网络质量必须实测", "buyer": "IT 与系统架构师", "decision": "这套部署能否在客户网络下运行", "product": "部署环境评估" }, { "id": "LG-011", "title": "检测方法学：同一个指标，两个数", "domain": "longevity", "layer": "capture", "physical": "采血时间、体位、空腹状态与检测平台", "cyber": "方法学标注与可比性说明", "gap": "同一指标换平台就不可直接比较，趋势被误读成变化", "evidence": "A", "evidence_note": "方法学差异有公开讨论；跨平台换算需厂商提供对照数据", "buyer": "检测机构与品牌方", "decision": "两次结果能否直接比较", "product": "检测方法卡" }, { "id": "LG-012", "title": "可穿戴精度：界面只给一个数", "domain": "longevity", "layer": "capture", "physical": "真实的睡眠、心率与活动量", "cyber": "设备输出与误差分布", "gap": "消费级设备在特定人群与动作下误差显著放大，但界面不显示误差", "evidence": "B", "evidence_note": "设备精度有公开评测；不同型号与佩戴方式差异很大", "buyer": "产品与数据负责人", "decision": "能否用这个数据下结论", "product": "设备误差说明" }, { "id": "LG-013", "title": "自报偏差：问卷里的饮食不是真的饮食", "domain": "longevity", "layer": "capture", "physical": "真实摄入与活动量", "cyber": "问卷与日志记录", "gap": "自报会系统性低估或高估，且偏差方向与人群相关", "evidence": "A", "evidence_note": "自报偏差是营养流行病学的公开议题；可用客观测量交叉校验", "buyer": "研究设计者", "decision": "这份数据能否作为证据", "product": "数据质量标注" }, { "id": "LG-014", "title": "睡眠测量：分期算法与金标准的距离", "domain": "longevity", "layer": "capture", "physical": "真实的睡眠结构与时长", "cyber": "由体动、心率推算的睡眠分期", "gap": "推算分期与多导睡眠图差异明显，个体间差异更大", "evidence": "B", "evidence_note": "消费级睡眠分期精度有公开研究；结论依赖设备与人群", "buyer": "产品负责人", "decision": "睡眠相关的 claim 能否成立", "product": "睡眠指标口径" }, { "id": "LG-015", "title": "基因与表观遗传输入：上限在数据里", "domain": "longevity", "layer": "capture", "physical": "真实的生物学状态与年龄", "cyber": "甲基化位点、多基因评分", "gap": "输入质量与人群代表性决定了输出的天花板，模型再复杂也补不回来", "evidence": "B", "evidence_note": "表观遗传时钟有公开发表；跨人群适用性仍在讨论", "buyer": "科学负责人", "decision": "这个评分能否用于个体建议", "product": "输入质量检查" }, { "id": "LG-016", "title": "生物年龄：一个数字骗了所有人", "domain": "longevity", "layer": "represent", "physical": "真实的健康寿命", "cyber": "单一生物年龄数值", "gap": "不同模型给出不同年龄，且都无法直接对应个体预后", "evidence": "B", "evidence_note": "生物年龄模型有公开研究；与个体结局的关联强度有限", "buyer": "品牌与消费者", "decision": "这个概念能否对外使用", "product": "模型对比说明" }, { "id": "LG-017", "title": "参考区间：别人的正常不等于你的正常", "domain": "longevity", "layer": "represent", "physical": "特定人群的真实分布", "cyber": "通用参考区间", "gap": "用通用区间判断个体，异常可能只是人群差异", "evidence": "A", "evidence_note": "参考区间按人群建立是公开规范；细分人群数据常缺失", "buyer": "临床顾问", "decision": "这个值算不算异常", "product": "区间适用性标注" }, { "id": "LG-018", "title": "综合评分：权重从哪来", "domain": "longevity", "layer": "represent", "physical": "多个系统的真实状态", "cyber": "加权总分", "gap": "权重没有依据，总分还会掩盖单项恶化", "evidence": "C", "evidence_note": "综合评分设计无统一标准；本条为常见问题的框架归纳", "buyer": "产品负责人", "decision": "这个评分是否有意义", "product": "评分设计原则" }, { "id": "LG-019", "title": "功能终点：用户真正在意的东西", "domain": "longevity", "layer": "verify", "physical": "走路、握力、平衡等真实生活能力", "cyber": "功能测试指标", "gap": "功能指标更接近用户感受，却比实验室指标更少被采集", "evidence": "B", "evidence_note": "功能评估有公开量表；执行标准化程度影响可比性", "buyer": "研究与产品负责人", "decision": "该测什么才说明问题", "product": "功能终点清单" }, { "id": "LG-020", "title": "研究设计：观察性能说明什么", "domain": "longevity", "layer": "verify", "physical": "真实的因果效应", "cyber": "研究设计类型与偏倚控制", "gap": "观察性研究更便宜也更多，但无法排除混杂与反向因果", "evidence": "A", "evidence_note": "研究设计分级是公开规范；单篇研究的结论强度需逐项判断", "buyer": "科学审阅", "decision": "这条证据能支持多强的结论", "product": "证据金字塔标注" }, { "id": "LG-021", "title": "效应量：显著不等于有用", "domain": "longevity", "layer": "verify", "physical": "用户真实可感知的改善", "cyber": "p 值与效应量", "gap": "大样本下微小差异也会统计显著，但对用户没有意义", "evidence": "A", "evidence_note": "效应量报告是公开规范；可感知阈值需按指标定义", "buyer": "科学顾问", "decision": "这个结果值不值得宣传", "product": "效应量报告模板" }, { "id": "LG-022", "title": "发表偏倚：没人发表的那一半", "domain": "longevity", "layer": "verify", "physical": "全部研究结果", "cyber": "已发表文献", "gap": "阴性结果发不出来，文献整体偏乐观", "evidence": "A", "evidence_note": "发表偏倚是公开的方法学议题；可用漏斗图与注册库交叉检查", "buyer": "科学审阅", "decision": "这个领域的结论有多可信", "product": "偏倚评估" }, { "id": "LG-023", "title": "亚组分析：切得够细总能显著", "domain": "longevity", "layer": "verify", "physical": "全体人群的真实效应", "cyber": "事后挑选的亚组结论", "gap": "反复切分数据，总会找到一组看起来有效", "evidence": "A", "evidence_note": "亚组分析需预设并校正，是公开统计规范", "buyer": "科学顾问", "decision": "这个亚组结论能否使用", "product": "分析计划检查" }, { "id": "LG-024", "title": "日内波动：单点采样说明不了趋势", "domain": "longevity", "layer": "capture", "physical": "一天内真实变化的指标", "cyber": "单点采样值", "gap": "日内波动可能大于干预效应，两次单点比较毫无意义", "evidence": "B", "evidence_note": "多数指标的日内变异有公开数据；具体幅度因指标而异", "buyer": "研究设计者", "decision": "采样方案是否合理", "product": "采样时间规范" }, { "id": "LG-025", "title": "依从性：无效还是没人照做", "domain": "longevity", "layer": "deploy", "physical": "用户真实执行情况", "cyber": "计划剂量与实际摄入", "gap": "效果不佳常被归因于产品无效，实际是没人按计划执行", "evidence": "B", "evidence_note": "依从性测量有公开方法；自报数据存在系统性偏差", "buyer": "产品与运营负责人", "decision": "效果不佳的原因是什么", "product": "依从性追踪" }, { "id": "LG-026", "title": "Claim 措辞：一条线的两边", "domain": "longevity", "layer": "decide", "physical": "真实的作用机制", "cyber": "结构、功能、疾病三类表述", "gap": "同一份证据，措辞跨过一条线就从合规变成违规", "evidence": "A", "evidence_note": "各国对健康声称的监管框架是公开的；具体判定需按市场核对", "buyer": "法规与市场负责人", "decision": "这句话能不能说", "product": "措辞分级表" }, { "id": "LG-027", "title": "监管分类：路径决定成本", "domain": "longevity", "layer": "decide", "physical": "产品的真实用途与风险", "cyber": "食品、补充剂、器械、药品的分类", "gap": "分类不同，证据要求与上市路径差一个数量级", "evidence": "A", "evidence_note": "分类规则公开；边界产品需专业判断", "buyer": "创始人", "decision": "按什么路径上市", "product": "分类判断树" }, { "id": "LG-028", "title": "风险收益：对健康人做干预", "domain": "longevity", "layer": "decide", "physical": "真实人群的获益与伤害", "cyber": "风险收益评估", "gap": "对健康人做干预，风险容忍度远低于对病人，但营销话术常照搬临床", "evidence": "B", "evidence_note": "风险收益框架公开；健康人群的绝对风险基数低，判断更依赖长期数据", "buyer": "医学顾问", "decision": "这个产品该不该面向健康人", "product": "风险收益表" }, { "id": "LG-029", "title": "个体化边界：群体证据落到一个人身上", "domain": "longevity", "layer": "decide", "physical": "个体的真实状况", "cyber": "基于群体数据的推荐", "gap": "群体平均效应对某个个体可能正好相反", "evidence": "B", "evidence_note": "个体化建议在营养与运动领域证据薄弱；需明确边界", "buyer": "产品与合规", "decision": "能不能给这个人这个建议", "product": "个体化边界说明" }, { "id": "LG-030", "title": "长期依从：产品假设用户会一直用", "domain": "longevity", "layer": "deploy", "physical": "真实的长期使用行为", "cyber": "留存曲线与使用模式", "gap": "效果依赖长期坚持，而绝大多数产品在几个月内就失去用户", "evidence": "B", "evidence_note": "留存数据需按产品自测；行业公开数据仅作参考", "buyer": "增长与产品负责人", "decision": "这个产品形态能否支撑长期使用", "product": "留存诊断" }, { "id": "LG-031", "title": "长期安全：短期试验看不见的东西", "domain": "longevity", "layer": "deploy", "physical": "长期使用后的真实不良事件", "cyber": "主动监测与上报机制", "gap": "低频长期风险在短期试验里几乎不可能出现", "evidence": "B", "evidence_note": "上市后监测在药品领域成熟；保健品体系覆盖程度不一", "buyer": "质量与法规负责人", "decision": "是否需要上市后监测", "product": "安全监测方案" }, { "id": "LG-032", "title": "配方变更：名义相同，实际不同", "domain": "longevity", "layer": "deploy", "physical": "原料来源与工艺的真实变化", "cyber": "版本记录与等效性说明", "gap": "换供应商后成分名义相同，实际纯度与生物利用度可能不同", "evidence": "B", "evidence_note": "变更控制是公开的质量体系要求；等效性需检测而非声明", "buyer": "供应链与质量负责人", "decision": "这个变更要不要重新验证", "product": "变更评估表" }, { "id": "LG-033", "title": "跨境合规：一个国家能说，另一个不能", "domain": "longevity", "layer": "deploy", "physical": "不同市场的真实监管要求", "cyber": "成分、宣称与标签的本地化规则", "gap": "在本国合规的产品，到另一个市场可能直接违规", "evidence": "A", "evidence_note": "各市场监管框架公开；具体条文需按目标市场核对", "buyer": "出海负责人", "decision": "这个市场能不能进", "product": "市场准入清单" }, { "id": "LG-034", "title": "消费者理解：相关被读成因果", "domain": "longevity", "layer": "deploy", "physical": "用户真实的理解水平", "cyber": "产品页面与说明文案", "gap": "用户把辅助读成治疗，把相关读成因果", "evidence": "B", "evidence_note": "消费者理解研究有公开方法；文案测试比专家评审更可靠", "buyer": "市场与客服负责人", "decision": "这段文案会不会被误解", "product": "文案审查清单" }, { "id": "LG-035", "title": "基因数据：泄露是不可撤销的", "domain": "longevity", "layer": "deploy", "physical": "个人真实身份与遗传信息", "cyber": "存储、共享与二次使用记录", "gap": "基因数据无法像密码一样更换，泄露后果是终身的", "evidence": "A", "evidence_note": "遗传数据保护在主要司法辖区有专门规定；执行细节按地区核对", "buyer": "法务与安全负责人", "decision": "这些数据能否存、能否用", "product": "数据治理清单" }, { "id": "CX-009", "title": "采集协议：每个项目重新发明一次", "domain": "cross", "layer": "capture", "physical": "不同团队采集时的真实条件", "cyber": "统一采集协议与最小字段", "gap": "每个项目都自定义采集，数据之间无法合并", "evidence": "B", "evidence_note": "采集协议在科研与工业均有公开范式；跨域统一仍需裁剪", "buyer": "数据负责人", "decision": "这批数据能否与其他批次合并", "product": "采集协议模板" }, { "id": "CX-010", "title": "元数据最小集：缺一个字段就废掉一批数据", "domain": "cross", "layer": "capture", "physical": "数据产生时的完整上下文", "cyber": "必填元数据字段", "gap": "当时觉得不重要的字段，半年后决定这批数据还能不能用", "evidence": "B", "evidence_note": "元数据标准公开；最小集需按用途定义", "buyer": "数据与工程负责人", "decision": "这份数据能否入库", "product": "元数据规范" }, { "id": "CX-011", "title": "脱敏与再识别：单看都安全，合起来不然", "domain": "cross", "layer": "capture", "physical": "真实个人或工厂身份", "cyber": "脱敏后的数据", "gap": "多个看似无关的字段交叉起来，仍可唯一识别", "evidence": "A", "evidence_note": "再识别风险有公开研究；k-匿名等方法可用于评估", "buyer": "法务与安全负责人", "decision": "这份数据能否对外提供", "product": "脱敏检查" }, { "id": "CX-012", "title": "术语映射：同一个东西，八个名字", "domain": "cross", "layer": "represent", "physical": "同一个真实对象或过程", "cyber": "各系统里的不同命名", "gap": "术语不统一，跨系统比较与聚合都做不了", "evidence": "B", "evidence_note": "本体工程有公开方法；维护成本主要在共识而非技术", "buyer": "架构师", "decision": "两套系统能否互通", "product": "术语映射表" }, { "id": "CX-013", "title": "时序对齐：对齐方式决定结论", "domain": "cross", "layer": "represent", "physical": "同时发生的多个事件", "cyber": "不同采样率的时间序列", "gap": "用不同方式对齐，同一批数据能得出不同结论", "evidence": "B", "evidence_note": "重采样与对齐方法公开；选择需与物理过程的时间尺度匹配", "buyer": "数据工程师", "decision": "这批序列能否一起分析", "product": "对齐规范" }, { "id": "CX-014", "title": "概率表示：下游分不清确定与瞎猜", "domain": "cross", "layer": "represent", "physical": "真实存在的不确定性", "cyber": "概率分布与置信度", "gap": "只给点估计，下游无法判断这个数该信几分", "evidence": "B", "evidence_note": "不确定性量化方法公开；商业输出常为可读性省略区间", "buyer": "算法与决策方", "decision": "这个输出有多可信", "product": "输出格式规范" }, { "id": "CX-015", "title": "成本模型：只算算力不算人", "domain": "cross", "layer": "represent", "physical": "真实的时间、人力与设备成本", "cyber": "成本估算模型", "gap": "AI 项目的账很少把人力和返工算进去，导致预算系统性偏低", "evidence": "B", "evidence_note": "成本估算方法公开；人力与返工比例需按项目历史校准", "buyer": "项目负责人", "decision": "这个方案值不值得做", "product": "成本模板" }, { "id": "CX-016", "title": "第三方复核：只有自己能复现的结论不算结论", "domain": "cross", "layer": "verify", "physical": "真实发生的结论", "cyber": "可被外部复算的证据包", "gap": "内部能复现但外部复算不了，客户就无法真正采信", "evidence": "B", "evidence_note": "可复现交付在科研与审计中有公开范式；商业交付中较少标准化", "buyer": "客户与审计方", "decision": "这个结论能否被采信", "product": "复核包规范" }, { "id": "CX-017", "title": "审计日志：出事之后才想起要有", "domain": "cross", "layer": "verify", "physical": "真实发生的操作与事件", "cyber": "不可篡改的操作记录", "gap": "需要回溯时才发现日志没开、字段不全或已被覆盖", "evidence": "A", "evidence_note": "日志规范公开；留存周期与不可篡改要求按合规场景确定", "buyer": "合规与运维负责人", "decision": "能否还原当时发生了什么", "product": "日志最小集" }, { "id": "CX-018", "title": "基准过拟合：排行榜不再代表现实", "domain": "cross", "layer": "verify", "physical": "真实世界的任务分布", "cyber": "固定基准集与排名", "gap": "大家开始针对基准优化，基准就与社会需要的真实能力脱钩", "evidence": "A", "evidence_note": "基准饱和与过拟合是公开议题；需定期更换或扩充基准", "buyer": "技术评估方", "decision": "这个排名还有意义吗", "product": "基准更新机制" }, { "id": "CX-019", "title": "跨域迁移：换个场景，结论默认失效", "domain": "cross", "layer": "verify", "physical": "新场景或新人群", "cyber": "在原域验证过的模型", "gap": "迁移后原有验证结论不再成立，但很少有人重新验证", "evidence": "B", "evidence_note": "域适应有公开研究；迁移后的验证不可省略", "buyer": "产品与科学负责人", "decision": "能否直接迁移", "product": "迁移验证清单" }, { "id": "CX-020", "title": "风险分级：不是所有问题都值得同等证据", "domain": "cross", "layer": "decide", "physical": "真实后果的严重程度", "cyber": "风险等级与所需证据强度", "gap": "按同一标准处理所有问题，要么过度投入，要么严重不足", "evidence": "B", "evidence_note": "风险分级在安全与医疗领域有公开框架；跨域映射需自定义", "buyer": "决策者", "decision": "这个风险需要多少证据", "product": "风险分级表" }, { "id": "CX-021", "title": "停止规则：什么时候该承认失败", "domain": "cross", "layer": "decide", "physical": "真实投入的沉没成本", "cyber": "事先约定的 kill criteria", "gap": "没有停止规则，项目会一直再试一版，直到没人敢提", "evidence": "B", "evidence_note": "预设停止规则是公开的决策实践；执行难点在组织而非方法", "buyer": "项目负责人", "decision": "什么时候该停", "product": "停止条件模板" }, { "id": "CX-022", "title": "验收条款：满足客户要求等于没有条款", "domain": "cross", "layer": "decide", "physical": "真实交付的质量", "cyber": "可判定的验收条款", "gap": "条款写得含糊，争议时双方都能解释成对自己有利", "evidence": "C", "evidence_note": "合同条款属法律判断；本条为常见问题的框架归纳", "buyer": "采购与法务", "decision": "什么情况下算违约", "product": "条款模板" }, { "id": "CX-023", "title": "保险与责任转移：新风险常在保单之外", "domain": "cross", "layer": "decide", "physical": "事故造成的真实损失", "cyber": "保险覆盖范围与责任转移", "gap": "既有保单通常不覆盖新形态的技术风险，出事才发现没有兜底", "evidence": "C", "evidence_note": "保险条款需按实际保单核对；本条为框架假设", "buyer": "法务与财务负责人", "decision": "风险能否被转移", "product": "保险核对清单" }, { "id": "CX-024", "title": "发布与回滚：没有退路就别上线", "domain": "cross", "layer": "deploy", "physical": "线上真实运行的系统", "cyber": "版本号、变更记录与回滚包", "gap": "没有回滚能力时，任何更新都是一次赌博", "evidence": "B", "evidence_note": "发布与回滚是成熟的工程实践；物理系统的回滚成本更高", "buyer": "运维负责人", "decision": "这个版本能否上线", "product": "发布检查单" }, { "id": "CX-025", "title": "退役与数据留存：项目结束之后", "domain": "cross", "layer": "deploy", "physical": "系统停止服务后的真实状态", "cyber": "数据留存期限与销毁记录", "gap": "项目结束没人管数据，既留下合规风险，也丢掉本可复用的资产", "evidence": "B", "evidence_note": "数据留存要求按合规场景确定；可复用的部分需提前约定", "buyer": "客户 IT 与法务", "decision": "数据该留多久、怎么销毁", "product": "退役清单" }, { "id": "CX-026", "title": "例外处理：AI 替你猜的那些事", "domain": "cross", "layer": "decide", "physical": "业务里真实存在的例外与边界情形", "cyber": "例外规则表与兜底路径", "gap": "没写下来的例外，AI 会替你猜；猜得像，错了你也看不出来", "evidence": "B", "evidence_note": "例外管理在业务规则与流程设计中有公开实践；覆盖度需逐场景梳理", "buyer": "业务负责人与一线主管", "decision": "这个系统能不能在真实业务里跑", "product": "例外清单与兜底路径" }, { "id": "CX-027", "title": "脏材料怎么进：非标准输入的处理", "domain": "cross", "layer": "capture", "physical": "真实的脏数据、脏材料与缺失字段", "cyber": "输入校验与降级规则", "gap": "演示用的是干净样例，现场一半输入是缺的、错的、格式不对的", "evidence": "B", "evidence_note": "输入校验与降级是公开工程实践；真实分布必须现场统计", "buyer": "数据与业务负责人", "decision": "系统在非理想输入下是否还能用", "product": "输入质量报告" }, { "id": "CX-028", "title": "以后怎么改：版本、回退与问题定位", "domain": "cross", "layer": "deploy", "physical": "系统上线后的真实变更需求", "cyber": "版本管理、回退包与问题定位责任人", "gap": "手搓的工具现场能用，但没人能改、不敢改、改坏回不去", "evidence": "B", "evidence_note": "版本与回退是成熟工程实践；责任归属需在合同与交接中明确", "buyer": "IT 与业务负责人", "decision": "这套系统能不能长期维护", "product": "维护责任表" }, { "id": "CX-029", "title": "权限：谁能看、谁能改、谁只能查", "domain": "cross", "layer": "decide", "physical": "组织里真实存在的三类角色", "cyber": "角色与权限矩阵", "gap": "自己搓的时候通常只有一个角色：都能看；进了组织就是三类人", "evidence": "B", "evidence_note": "访问控制在信息安全体系中有公开标准；角色划分需按组织实际确定", "buyer": "IT 与合规负责人", "decision": "这套系统能不能交给组织用", "product": "权限矩阵" }, { "id": "CX-030", "title": "数据安全：数据放在哪，出事谁负责", "domain": "cross", "layer": "deploy", "physical": "客户名单、合同金额、员工信息等敏感数据", "cyber": "存储位置、导出控制与事件处置", "gap": "扔进一个私人账号跑着方便，出事就是事故", "evidence": "B", "evidence_note": "数据安全要求由合规场景决定；具体条款需按地区与行业核对", "buyer": "IT 与法务负责人", "decision": "这套方案是否允许上线", "product": "数据安全清单" } ] }