EN · implementation-guideOPC Global AI 原生 Web 应用实施规范
文件状态
该文件是 OPC 的实施和治理的真实来源 全球主页和利益相关者入职体验。规范的TABLE AI × OPC Global 蓝图在该指令层之后被完整保留。
- 源蓝图:用户提供的
pasted-text.txt,1,195行和55,328字节 - OPC Global 品牌来源:https://apuch.art/api/brands/opcglobal.json
- 品牌目录清单:https://apuch.art/api/manifest.json
- 固定品牌版本:
735ff5e,发布于 2026-07-24 - 固定资产校验和:
- 英雄:
8351935e9abc74b963309f2ec5677f22a9eaf4998c0e0cb59ecd55be42e4fe40 - 矢量标志:
b58c3be7390a1b625de8ad7d5c953fe8b95d1467692977c338c9b255f266502c - 原蓝色标志:
c970696e04ab2859dd66c7b85c2c1b461d06ede4e2d65ec6a97d9c11db1396b3
- 英雄:
- 品牌警告:已发布的记录标记为
placeholder,并且没有primaryGuide。代币和资产是官方的。版式、布局、 下面的组件、运动和交互规则是产品系统的扩展。
产品目标
构建三个协调层:
- 完整的指示和治理记录。
- 简洁的双语公众主页,讲解任务价值操作 系统无需复制每个策略表。
- 面向所有公众和运营人员的双语、基于角色的入职原型 利益相关者。
- 公共标准中心和版本化 API/MCP 注册表,公开 规范蓝图、受控术语、任务规范、角色、策略、 和来源出处,而不创建第二个事实来源。
经验必须使运营模式易于理解、可审计并且 可采取行动,但从不接受身份证件、付款凭据, 机密的客户材料、真实的 KYC 信息或受监管的数据。
实施基线
- 通过 vinext 进行站点兼容的 React 和 Next 风格路由。
- 用于令牌集成的 Tailwind CSS v4,以及用于 编辑布局和状态系统。
- 默认情况下服务器渲染的内容;客户端代码与入职隔离 状态、验证、确认和设备本地持久性。
- 使用经过校验和验证的官方英雄的一种优化的本地衍生品 用于页面性能。未触及的官方来源仍然存档在旁边 它。
- 此版本仅批准了一个品牌社交预览:
public/og.png,SHA-256bf3cc07f9597550af2830b217d7c3cc126d44a53a5d6791ccc10baacdb9d8d13。 - 该网站在 JavaScript 延迟的情况下仍然有用,并且副驾驶仍然有用 没有远程模型提供者。
体验架构
路线
/重定向至/zh/en和/zh是本地化主页/en/standards和/zh/standards渲染完整的规范蓝图/en/developers和/zh/developers记录公共 API 和 MCP 边界/en/onboarding和/zh/onboarding是利益相关者选择器/[locale]/onboarding/[role]包含每个引导角色的旅程- 区域设置切换保留当前路线和设备本地进度
公共利益相关者角色
| 角色键 | 英文名称 | 中文名称 | 主要入驻结果 |
|---|---|---|---|
buyer |
组织/买家 | 企业与机构需求方 | 结构化需求、结果、数据等级、时间安排、预算范围和验收负责人 |
expert |
专家/标准所有者 | 专家与标准业主 | 领域证据、标准或 IP 范围、审查角色和可用性 |
coach |
教练/工作包主管 | 教练与工作包负责人 | 交付领导、任务分解、指导、质量责任和 OPC 级别 |
contributor |
贡献者/OPC 单位 | 贡献者与 OPC 单元 | 技能、任务证据、准备情况、可用性、数据资格和成长路径 |
ecosystem-partner |
生态系统合作伙伴 | 生态合作伙伴 | 区域节点或任务工厂范围、地域、本地能力和交付网络 |
运营利益相关者角色
| 角色关键 | 英文名 | 中文名 | 主要入职结果 |
|---|---|---|---|
task-architect |
任务架构师 | 任务架构师 | TaskGraph、TaskSpec、依赖性、努力、接受和拆分设计准备 |
quality-reviewer |
独立质量审核员 | 独立质量审查人 | 标题、独立性、冲突、证据和审查时机准备 |
compliance-lead |
数据与合规主管 | 数据与合规负责人 | 数据分类、区域控制、访问、保留和升级准备 |
settlement-lead |
结算 & ValueLedger 铅 | 分账与结算责任人 | SplitRule、分类账、发布、异常和审计准备情况 |
主页叙述
主页必须按以下顺序呈现这些想法:
1、把真实的需求变成可以接受、可以解决的工作。 2. 定义结果、设计任务、组装交付、验证证据、解决 透明地,并重用有效的东西。 3. TABLE AI、OPC Global、共享基础设施、区域节点和任务 工厂。 4. 五个公共角色进入路径。 5. TaskSpec,循证接受,透明的SplitRules,可解释 匹配、数据分类和争议权利。 6.三个初始任务工厂。 7. 经验证的结算值作为操作北极星定义。 8. 透明的服务费用和贡献权。 9. 单一清晰的入职路线。
切勿将扩展门、0.5/3/2 模型或任何目标作为 取得了成果。不得使用虚假的客户标识、捏造证据、发明的 性能数据,或“零平台费用”声明。
品牌和UI系统
官方主题令牌
| 语义标记 | 价值 | 用途 |
|---|---|---|
| 小学 | #1D3557 |
导航、主要操作、标题、结构重点 |
| 口音 | #B79B63 |
验证值、选定状态、有限亮点 |
| 中学 | #A23E3E |
仅重大风险和错误 |
| 表面 | #F3F5F6 |
页面和分组部分背景 |
| 纸 | #FFFFFF |
高架或输入表面 |
| 墨水 | #162130 |
主要文本 |
| 静音 | #596776 |
辅助文本 |
| 线路 | rgba(29, 53, 87, 0.16) |
分隔线和控制边框 |
产品系统扩展
- 本版本仅使用浅色主题。
- Geist Sans 与 Noto Sans SC 和系统回退;用于 ID 的 Geist Mono, 公式和分类账值。
- 一个 12px 圆角半径比例。
- 荧光粉图标具有一致的重量。
DESIGN_VARIANCE: 5、MOTION_INTENSITY: 3、VISUAL_DENSITY: 4。- 动作必须传达层次结构、进度、反馈或状态变化。
- 所有自动机芯必须遵守
prefers-reduced-motion。 - 避免通用的人工智能--紫色渐变、霓虹灯、过多的玻璃、发明的SVG 插图、等量三卡网格和假冒产品屏幕截图。
- 使用官方全球网络英雄图像和官方徽标资产。
本地化规则
- 中文是默认语言环境。无需重置路线即可使用英语 或入职状态。
- 所有导航、内容、角色、表单标签、验证、副驾驶输出、 隐私指南和后续步骤必须以两种语言存在。
- 区域设置切换仅更改显示方式。它不得重置答案。
- TaskSpec、TaskGraph、VSV、SplitRule 和 ValueLedger 等标识符 跨区域保持稳定并收到翻译后的解释。
结构化入职副驾驶
副驾驶不是一个开放式的聊天界面。它是一个确定性的、 类型化的入职状态之上的提供者中立的指导层。
允许的行为:
- 推荐或确认角色;
- 识别缺失的非敏感信息;
- 计算准备状态;
- 识别政策或数据风险;
- 提出下一步的准备行动;
- 引用相关蓝图部分。
要求的行为:
- 将建议明确标记为提案;
- 让用户控制确认;
- 在人工智能服务不可用时工作;
- 在浏览器存储中仅保存非敏感状态;
- 提供重置动作;
- 切勿颁发凭证、释放资金、更改 SplitRules、完成 KYC,或 提交最终生产申请。
公共接口
type Locale = "en" | "zh";
type PublicRole =
| "buyer"
| "expert"
| "coach"
| "contributor"
| "ecosystem-partner";
type OperationalRole =
| "task-architect"
| "quality-reviewer"
| "compliance-lead"
| "settlement-lead";
type Role = PublicRole | OperationalRole;
interface OnboardingState {
schemaVersion: 1;
role: Role;
answers: Record<string, string | string[] | boolean>;
evidenceRefs: string[];
currentStep: 0 | 1 | 2;
readiness: "starting" | "building" | "ready";
riskFlags: string[];
nextActions: string[];
confirmations: string[];
lastSavedAt: string;
}
interface AIRecommendation {
summary: string;
suggestedRole: Role;
missingInformation: string[];
riskFlags: string[];
blueprintCitations: string[];
confidence: "low" | "medium" | "high";
confirmationRequired: true;
}
interface BrandSnapshot {
brandSlug: "opcglobal";
status: string;
sourceVersion: string;
synchronizedAt: string;
tokens: Record<string, string | string[]>;
assets: Array<{
id: string;
filename: string;
sha256: string;
}>;
}
interface StandardsManifest {
schemaVersion: string;
contentVersion: string;
sourceCommit?: string;
instructionSha256: string;
blueprintSha256: string;
canonicalLanguage: "zh";
}
接口术语定义{
id:字符串;
canonicalZh:字符串;
canonicalEn: 字符串;
别名:字符串[];
定义Zh:字符串;
定义En: 字符串;
节 ID:字符串;
}
MCP 边界
Use apuch.art’s published get_brand, list_assets, get_asset, and
request_asset_url capabilities through the brand-sync adapter, with its
public brand JSON as the deterministic fallback.
Published OPC resources:
opc://manifest/currentopc://glossary/currentopc://standards/{sectionId}opc://roles/{role}opc://taskspec/{taskCode}opc://policy/{policyId}opc://brand/current
Permitted read or draft tools:
resolve_termvalidate_standard_referencevalidate_taskspeccompare_standard_versionsdraft_alignment_reportrecommend_roledraft_onboarding_profileassess_readinesspropose_task_path
Do not expose autonomous payment release, credential issuance, SplitRule
modification, or final onboarding submission. Remote MCP uses stable protocol
version 2025-11-25 over stateless Streamable HTTP and may negotiate
future versions only after they become stable. The 2026-07-28 revision is an
RC at the time of implementation and must not be a first-release dependency.
Protocol status must be rechecked against the
official MCP releases.
The build-time brand adapter should attempt the published MCP tools first and fall back to public brand JSON without weakening checksum validation. Runtime AI integration must sit behind a provider-neutral adapter. Every future model-assisted provenance event records the model, prompt or Skill version, tool calls, source data, output proposal, and explicit human confirmation.
无障碍与性能
- WCAG AA contrast for all text and controls.
- Visible keyboard focus on all interactive elements.
- Labels above inputs; helper text optional; errors below inputs.
- No placeholder-only labels.
- Explicit loading, empty, validation, completion, recovery, and reset states.
- Explicit mobile layouts at 375px, 768px, 1024px, and 1440px.
- Hero image dimensions must be reserved to prevent layout shift.
- Target LCP below 2.5 seconds, INP below 200ms, and CLS below 0.1.
- Honor reduced motion and avoid scroll listeners tied to React state.
可追溯性矩阵
| 蓝图部分 | 主页 | 入职 | 未来产品模块 | 治理/政策 |
|---|---|---|---|---|
| 一、核心结论 | 英雄与价值循环 | 所有角色介绍 | 指挥中心 | 平台章程 |
| 二、愿景、使命与战略定位 | 使命与VSV | 所有角色目的陈述 | KPI层 | 战略章程 |
| 三、TABLE AI 与 OPC Global 的角色分工 | 运营网络 | 生态系统合作伙伴 | 组织和权限 | 品牌及运营模式 |
| 四、从职业目录转向任务图谱 | TaskGraph 说明 | 角色与技能匹配 | TaskGraph / SkillGraph | 分类标准 |
| 五、任务体系与最小交易颗粒度 | 价值环细节 | 买家和任务建筑师 | TaskGraph | 任务粒度策略 |
| 六、标准任务卡 TaskSpec | 信任架构 | 买家和任务建筑师 | TaskSpec 库 | 任务标准 |
| 七、第一批标准任务目录 | 任务工厂示例 | 相关角色示例 | 标准任务市场 | 版本化目录 |
| 八、价值交付标准 | 证据和质量 | 买家和评论家 | 品质引擎 | 验收政策 |
| 九、工作量、报价与价值分离 | 透明的商业模式 | 买家及结算线索 | 报价引擎 | 定价政策 |
| 十、智能分账 | 透明的贡献权 | 结算线索 | ValueLedger | 分拆及资产政策 |
| 十一、Demand Radar | 任务工厂上下文 | 买家和生态系统合作伙伴 | 需要工作室 / Demand Radar | 机会得分 |
| 十二、AI时代新岗位 | 角色进入路径 | 专家、教练、贡献者 | TalentGraph | 角色塑造标准 |
| 十三、OPC UNI | 有偿成长 | 教练和贡献者 | OPC UNI | 凭证政策 |
| 十四、人才匹配与信誉体系 | 可解释的匹配 | 交付角色 | 匹配引擎 | 上诉和声誉政策 |
| 十五、交付、验收与争议机制 | 信任架构 | 买家、教练、评论家 | 工作室/质量引擎 | 争议政策 |
| 十六、产品系统架构 | 共享基础设施 | 运营角色 | 所有命名模块 | 技术治理 |
| 十七、基础架构 | 区域网络 | 生态系统合作伙伴和合规性 | 信任与合规 | 区域和数据政策 |
| 十八、商业模式 | 透明的服务模式 | 买家及合作伙伴 | 报价/计费 | 商业政策 |
| 十九、首期三个任务工厂 | 任务工厂部分 | 买家及合作伙伴 | 垂直任务工厂 | 发射焦点 |
| 二十、90天落地计划 | 未作为成就呈现 | 下一步指导 | 交付路线图 | 项目治理 |
| 二十一、拓展闸门 | 标记验证标准 | 运营准备 | 指挥中心 | 扩张政策 |
| 二十二、管理指标体系 | VSV 和假设框架 | 角色准备指标 | 分析 | 测量标准 |
| 二十三、最容易失败的十个地方 | 信任保障 | 禁止行为 | 风险监控 | 风险登记册 |
| 战略最终表达 | 最终号召性用语 | 完成总结 | 产品叙述 | 战略章程 |
验收检查
- 下面的规范蓝图保留了所有源标题、表格、任务代码、 公式、基准、门限和风险项目。
- 每个角色都有完整的英文和中文内容。
- 每个角色页面包括目的、职责、先决条件、证据、 禁止的行动、准备情况、下一步行动和蓝图引用。
- 品牌资产与记录的 SHA-256 校验和相匹配。
- 没有可见的网站副本包含破折号。
- 没有仅生产或敏感的操作被呈现为可操作的。
- 所有路由的构建和渲染都没有客户端错误。---
TABLE AI × OPC Global 规范蓝图
统一组织、身份、结算模块扩展
此版本通过受保护的操作扩展了公共任务价值系统
工作区。公式迁移源为
ref/经营与分账模型v3.html,固定在 SHA-256
76ca4c88c9c1778f1842a1c03de2a74d38ab2a6d68b7de2116266a4cba740d79。
来源是运营预测计算器,而不是会计分类账。其
示例成为黄金测试,而生产规则则添加版本控制,
参与者、幂等性、整数舍入、批准、逆转、争议和
仅附加出处。
身份、组织和权利持有者
- TableUser Hosted OIDC 是身份验证机构。自然人使用
taid作为唯一稳定的跨系统密钥;登录sub、电子邮件和用户名 绝不能成为业务外键。 - 组织是协作和授权空间。每个分割 项目属于一个组织,可能涉及多个法人实体。
- 每个结算基准线都属于一个法人实体。各实体 计算、批准并过账到独立的分类账;项目级 视图汇总但绝不会隐式合并实体资金。
- 一方可以是受 TAID 约束的自然人,也可以是经过验证的法人实体。
每一方都有一个全球分拆账户。
ledger_ready不是payout_ready。 - 外部受邀者仅成为项目参与者。他们不会成为 组织成员。邀请使用一次性、可撤销、散列令牌, 默认7天后过期,登录后需要补发 电子邮件不同。
TableUser 是资源授权机构,flay01 存储版本化的
仅投影。正式写入实时调用TableUser,失败时关闭
该服务不可用。资源补助涵盖organization,
legal_entity、split_project 和 equity_ledger;角色名称为:
platform:admin、opc:org:owner、opc:org:admin、
opc:settlement:approver、opc:equity:admin、
opc:project:admin|editor|viewer、opc:project:participant 和
opc:auditor。
平台管理员可以配置、暂停、排除故障和起草,但 不能单独批准组织应付账款分类账。组织管理员可以 创建项目、发布规则、邀请参与者并准备结算。 组织所有者或结算审批人批准正式应付账款。安 只有一名有效批准者的组织可以自行批准低于其自己的批准者 未来有效限制。
经营推演与正式结算相互隔离
OperatingPlan 模型服务、客户、收入周期、收集、
退款和成本预测;所有输出都是建议性的。 Settlement仅消耗
通过来源证据确认实际金额。预测退款或收款
利率永远不会产生或改变应付义务。
第一个不可变模板是 OperatingFourPoolV3@1:1. 汇总一个法人实体已确认的收入或批准的收付实现制。
2. 收付实现制首先保护所有非劳动收入和退款风险。
3.扣除实际退款、坏账、税金、直接成本、明确平台
费用。
4、计算正毛利率和准备金;损失变成实体
赤字且绝不是参与者负应付款项。
5. 分配获取、交付、产品和平台池,其比率总和
精确到 100%。
6. 合作第二年的收购权下降至50%,合作年起下降至20%
三;剩余的回流到平台池。
7. 产品权利在首次交付后三年或之后到期(以较早者为准)
累计拨款相当于研究投入的五倍。
8.申请预支工资抵消、外部税务处理、固定成本、贷款
还款,以及明确的平台瀑布。
9. 外部成员的资格由规则版本控制,而不是由规则版本控制。
内部/外部硬编码区别。
10. 在每个池内独立标准化参与者权重。
货币使用 ISO 4217 货币和整数小单位。费率使用精确的小数 代表。最大剩余分配保证了守恒, 稳定的派对 ID 作为决胜局。退款和损失永远不会自动生成 债务;未来的抵消需要一个预先存在的、公认的、有争议的规则。
规则、外汇、审核和账本生命周期
参考外汇最初使用欧洲央行每日参考利率和确定性欧元 三角测量。 UI 必须将其标记为参考速率,而不是执行速率。 周末或节假日使用最新公布的汇率,不晚于 参考日期。系统管理员或组织管理员可以创建手册 仅针对未经批准的运行进行覆盖,并附有理由、证据、差异、影响 范围和参与者通知。运行冻结原始响应、源日期、对、 散列和覆盖;历史永远不会以当前的速率重新计算。
规则版本遵循 draft → published → active → superseded。他们影响
仅未来期间。新的和受影响的参与者必须接受当前的
进入新时期之前的版本。历史规则、计算快照、
并且批准的账本是不可变的。
结算运行如下
draft → calculated → participant_review → ready_for_approval → approved。
参与者审核持续五个工作日。一条有争议的线路冻结了
无可争议的线路继续;有效期记录无异议而非捏造
批准。已批准的运行无法编辑。修正创建调整运行
具有链接的冲销或调整条目。
运营股权台账
每个符合条件的法人实体都可以拥有一个仅追加的运营股权台账
涵盖类别、持有者、单位、认购和支付金额、期权池、
归属、发行、转让、注销、回购、调整事件。
派生余额永远不会被直接覆盖。内部审批可能会影响
经营权和利润分配却始终显示
legalRegistrationStatus。中国大陆和香港有各自的
管辖权概况;所有其他司法管辖区在此版本中均为只读。
该模块默认关闭,仅在初始持有快照后打开
和解了。它不能取代法定股东名册、法律备案、
或合法转让文书。
REST 和 MCP 合约
REST 正式写入需要 TableUser OIDC JWT 或 M2M 令牌、资源授予、
Idempotency-Key 和 If-Match 用于版本更新。响应返回
资源版本、审核事件 ID 和输入/计算哈希。
MCP 资源包括:
opc://organizations/{id}opc://split-templates/{version}opc://split-projects/{id}opc://settlement-runs/{id}opc://equity-ledgers/{legalEntityId}
MCP 工具仅限于读取和草稿操作:
simulate_operating_plan、draft_split_project、validate_split_rule、
calculate_draft_settlement、explain_settlement、
compare_rule_versions 和 draft_invitation。 MCP 绝不能发布规则,
覆盖 FX、批准分类账、发布股权事件、提交最终入职,或
执行付款。每个 AI 或 MCP 提案都会记录模型、提示或技能
版本、工具调用、源数据和来源事件中的人工确认。
数据、部署和显式排除
生产数据保留在腾讯云南京。目标数据平面是
PostgreSQL 16 具有单独的 flay01_core 和 tableuser_authz 数据库,
独立的运行时和迁移角色、TLS、静态加密、自动化
备份,以及至少十四天的时间点恢复。 Postgres 发件箱
工作人员无需添加 Redis 即可处理计算、电子邮件和通知。
私有站点部署是经过清理的 UI 审查平面,并且从不连接
到生产分类帐。
试点能力目标是 100 个组织,每个项目 100 名参与者, 每次运行 5,000 行;计算 p95 不到五秒并且在组织内 授权p95低于200ms。大型账本使用服务器分页。
本次发布不执行付款、收集身份证件、银行 凭证、KYC 材料、机密客户上传、颁发凭证、 进行纳税申报或扣缴,提供任意公式设计者, 自动追偿债务,或声称 UI 行动完成了法定权益 转移。
TABLE AI × OPC Global 全球透明任务协作与智能分账体系蓝图
查看规范原文
OPC Global AI-Native Web App Instruction
Document status
This file is the implementation and governance source of truth for the OPC Global homepage and stakeholder onboarding experience. The canonical TABLE AI × OPC Global blueprint is preserved in full after this instruction layer.
- Source blueprint: user-provided
pasted-text.txt, 1,195 lines and 55,328 bytes - OPC Global brand source: https://apuch.art/api/brands/opcglobal.json
- Brand directory manifest: https://apuch.art/api/manifest.json
- Pinned brand version:
735ff5e, published 2026-07-24 - Pinned asset checksums:
- hero:
8351935e9abc74b963309f2ec5677f22a9eaf4998c0e0cb59ecd55be42e4fe40 - vector logo:
b58c3be7390a1b625de8ad7d5c953fe8b95d1467692977c338c9b255f266502c - primary blue logo:
c970696e04ab2859dd66c7b85c2c1b461d06ede4e2d65ec6a97d9c11db1396b3
- hero:
- Brand caveat: the published record is marked
placeholderand has noprimaryGuide. The tokens and assets are official. Typography, layout, component, motion, and interaction rules below are product-system extensions.
Product objective
Build three coordinated layers:
- This complete instruction and governance record.
- A concise bilingual public homepage that explains the task-value operating system without reproducing every policy table.
- A bilingual, role-based onboarding prototype for all public and operational stakeholders.
- A public standards center and versioned API/MCP registry that expose the canonical blueprint, controlled terminology, TaskSpecs, roles, policies, and source provenance without creating a second source of truth.
The experience must make the operating model understandable, auditable, and actionable while never accepting identity documents, payment credentials, confidential customer material, real KYC information, or regulated data.
Implementation baseline
- Sites-compatible React and Next-style routing through vinext.
- Tailwind CSS v4 for token integration, with scoped product CSS for the editorial layout and state system.
- Server-rendered content by default; client code is isolated to onboarding state, validation, confirmation, and device-local persistence.
- One optimized local derivative of the checksum-verified official hero is used for page performance. The untouched official source remains archived beside it.
- Exactly one branded social preview is approved for this release:
public/og.png, SHA-256bf3cc07f9597550af2830b217d7c3cc126d44a53a5d6791ccc10baacdb9d8d13. - The site remains useful with JavaScript delayed and the copilot remains useful without a remote model provider.
Experience architecture
Routes
/redirects to/zh/enand/zhare localized homepages/en/standardsand/zh/standardsrender the complete canonical blueprint/en/developersand/zh/developersdocument the public API and MCP boundary/en/onboardingand/zh/onboardingare stakeholder selectors/[locale]/onboarding/[role]contains each guided role journey- Locale switching preserves the current route and device-local progress
Public stakeholder roles
| Role key | English name | Chinese name | Primary onboarding outcome |
|---|---|---|---|
buyer |
Organization / Buyer | 企业与机构需求方 | A structured need, outcome, data grade, timing, budget band, and acceptance owner |
expert |
Expert / Standard Owner | 专家与标准所有者 | Domain evidence, standard or IP scope, review role, and availability |
coach |
Coach / Work-package Lead | 教练与工作包负责人 | Delivery leadership, task decomposition, mentoring, quality responsibility, and OPC level |
contributor |
Contributor / OPC Unit | 贡献者与 OPC 单元 | Skills, task evidence, readiness, availability, data eligibility, and growth path |
ecosystem-partner |
Ecosystem Partner | 生态合作伙伴 | Regional-node or task-factory scope, territory, local capability, and delivery network |
Operational stakeholder roles
| Role key | English name | Chinese name | Primary onboarding outcome |
|---|---|---|---|
task-architect |
Task Architect | 任务架构师 | TaskGraph, TaskSpec, dependency, effort, acceptance, and split-design readiness |
quality-reviewer |
Independent Quality Reviewer | 独立质量审查人 | Rubric, independence, conflict, evidence, and review-timing readiness |
compliance-lead |
Data & Compliance Lead | 数据与合规负责人 | Data classification, regional controls, access, retention, and escalation readiness |
settlement-lead |
Settlement & ValueLedger Lead | 分账与结算负责人 | SplitRule, ledger, release, exception, and audit readiness |
Homepage narrative
The homepage must present these ideas in this order:
- Turn real demand into work that can be accepted and settled.
- Define outcomes, design tasks, assemble delivery, verify evidence, settle transparently, and reuse what works.
- TABLE AI, OPC Global, shared infrastructure, regional nodes, and task factories.
- The five public role entry paths.
- TaskSpec, evidence-based acceptance, transparent SplitRules, explainable matching, data classification, and dispute rights.
- The three initial task factories.
- Verified Settled Value as the operating north-star definition.
- Transparent service fees and contribution rights.
- A single clear route into onboarding.
Never present the expansion gates, the 0.5 / 3 / 2 model, or any target as an achieved result. Do not use fake customer logos, fabricated proof, invented performance data, or “zero platform fee” claims.
Brand and UI system
Official theme tokens
| Semantic token | Value | Usage |
|---|---|---|
| Primary | #1D3557 |
Navigation, primary actions, headings, structural emphasis |
| Accent | #B79B63 |
Verified value, selected states, limited highlights |
| Secondary | #A23E3E |
Critical risks and errors only |
| Surface | #F3F5F6 |
Page and grouped section background |
| Paper | #FFFFFF |
Elevated or input surfaces |
| Ink | #162130 |
Primary text |
| Muted | #596776 |
Secondary text |
| Line | rgba(29, 53, 87, 0.16) |
Dividers and control borders |
Product-system extensions
- Light theme only for this release.
- Geist Sans with Noto Sans SC and system fallbacks; Geist Mono for IDs, formulas, and ledger values.
- One 12px corner-radius scale.
- Phosphor icons with a consistent weight.
DESIGN_VARIANCE: 5,MOTION_INTENSITY: 3,VISUAL_DENSITY: 4.- Motion must communicate hierarchy, progress, feedback, or state change.
- All automatic movement must respect
prefers-reduced-motion. - Avoid generic AI-purple gradients, neon glows, excessive glass, invented SVG illustration, equal three-card grids, and fake product screenshots.
- Use the official global-network hero image and official logo assets.
Localization rules
- Chinese is the default locale. English is available without resetting route or onboarding state.
- All navigation, content, roles, form labels, validation, copilot output, privacy guidance, and next steps must exist in both languages.
- Locale switching changes presentation only. It must not reset answers.
- Identifiers such as TaskSpec, TaskGraph, VSV, SplitRule, and ValueLedger remain stable across locales and receive a translated explanation.
Structured onboarding copilot
The copilot is not an open-ended chat surface. It is a deterministic, provider-neutral guidance layer over the typed onboarding state.
Allowed behaviors:
- recommend or confirm a role;
- identify missing non-sensitive information;
- calculate a readiness state;
- identify policy or data risks;
- propose the next preparation action;
- cite the relevant blueprint section.
Required behaviors:
- clearly label recommendations as proposals;
- keep the user in control of confirmation;
- work when AI services are unavailable;
- save only non-sensitive state in browser storage;
- provide a reset action;
- never issue credentials, release money, alter SplitRules, complete KYC, or submit a final production application.
Public interfaces
type Locale = "en" | "zh";
type PublicRole =
| "buyer"
| "expert"
| "coach"
| "contributor"
| "ecosystem-partner";
type OperationalRole =
| "task-architect"
| "quality-reviewer"
| "compliance-lead"
| "settlement-lead";
type Role = PublicRole | OperationalRole;
interface OnboardingState {
schemaVersion: 1;
role: Role;
answers: Record<string, string | string[] | boolean>;
evidenceRefs: string[];
currentStep: 0 | 1 | 2;
readiness: "starting" | "building" | "ready";
riskFlags: string[];
nextActions: string[];
confirmations: string[];
lastSavedAt: string;
}
interface AIRecommendation {
summary: string;
suggestedRole: Role;
missingInformation: string[];
riskFlags: string[];
blueprintCitations: string[];
confidence: "low" | "medium" | "high";
confirmationRequired: true;
}
interface BrandSnapshot {
brandSlug: "opcglobal";
status: string;
sourceVersion: string;
synchronizedAt: string;
tokens: Record<string, string | string[]>;
assets: Array<{
id: string;
filename: string;
sha256: string;
}>;
}
interface StandardsManifest {
schemaVersion: string;
contentVersion: string;
sourceCommit?: string;
instructionSha256: string;
blueprintSha256: string;
canonicalLanguage: "zh";
}
interface TermDefinition {
id: string;
canonicalZh: string;
canonicalEn: string;
aliases: string[];
definitionZh: string;
definitionEn: string;
sectionId: string;
}
MCP boundary
Use apuch.art’s published get_brand, list_assets, get_asset, and
request_asset_url capabilities through the brand-sync adapter, with its
public brand JSON as the deterministic fallback.
Published OPC resources:
opc://manifest/currentopc://glossary/currentopc://standards/{sectionId}opc://roles/{role}opc://taskspec/{taskCode}opc://policy/{policyId}opc://brand/current
Permitted read or draft tools:
resolve_termvalidate_standard_referencevalidate_taskspeccompare_standard_versionsdraft_alignment_reportrecommend_roledraft_onboarding_profileassess_readinesspropose_task_path
Do not expose autonomous payment release, credential issuance, SplitRule
modification, or final onboarding submission. Remote MCP uses stable protocol
version 2025-11-25 over stateless Streamable HTTP and may negotiate
future versions only after they become stable. The 2026-07-28 revision is an
RC at the time of implementation and must not be a first-release dependency.
Protocol status must be rechecked against the
official MCP releases.
The build-time brand adapter should attempt the published MCP tools first and fall back to public brand JSON without weakening checksum validation. Runtime AI integration must sit behind a provider-neutral adapter. Every future model-assisted provenance event records the model, prompt or Skill version, tool calls, source data, output proposal, and explicit human confirmation.
Accessibility and performance
- WCAG AA contrast for all text and controls.
- Visible keyboard focus on all interactive elements.
- Labels above inputs; helper text optional; errors below inputs.
- No placeholder-only labels.
- Explicit loading, empty, validation, completion, recovery, and reset states.
- Explicit mobile layouts at 375px, 768px, 1024px, and 1440px.
- Hero image dimensions must be reserved to prevent layout shift.
- Target LCP below 2.5 seconds, INP below 200ms, and CLS below 0.1.
- Honor reduced motion and avoid scroll listeners tied to React state.
Traceability matrix
| Blueprint section | Homepage | Onboarding | Future product module | Governance / policy |
|---|---|---|---|---|
| 一、核心结论 | Hero and value loop | All role introductions | Command Center | Platform charter |
| 二、愿景、使命与战略定位 | Mission and VSV | All role purpose statements | KPI layer | Strategy charter |
| 三、TABLE AI 与 OPC Global 的角色分工 | Operating network | Ecosystem partner | Organization and permissions | Brand and operating model |
| 四、从职业目录转向任务图谱 | TaskGraph explanation | Role and skill matching | TaskGraph / SkillGraph | Classification standard |
| 五、任务层级与最小交易颗粒度 | Value-loop detail | Buyer and task architect | TaskGraph | Task granularity policy |
| 六、标准任务卡 TaskSpec | Trust architecture | Buyer and task architect | TaskSpec library | Task standard |
| 七、第一批标准任务目录 | Task-factory examples | Relevant role examples | Standard task market | Versioned catalog |
| 八、价值交付标准 | Evidence and quality | Buyer and reviewer | Quality Engine | Acceptance policy |
| 九、工作量、报价与价值分离 | Transparent commercial model | Buyer and settlement lead | Quote Engine | Pricing policy |
| 十、智能分账 | Transparent contribution rights | Settlement lead | ValueLedger | Split and asset policy |
| 十一、Demand Radar | Task-factory context | Buyer and ecosystem partner | Need Studio / Demand Radar | Opportunity scoring |
| 十二、AI时代新岗位 | Role entry paths | Expert, coach, contributor | TalentGraph | Role formation standard |
| 十三、OPC UNI | Paid growth | Coach and contributor | OPC UNI | Credential policy |
| 十四、人才匹配与信誉体系 | Explainable matching | Delivery roles | Match Engine | Appeal and reputation policy |
| 十五、交付、验收与争议机制 | Trust architecture | Buyer, coach, reviewer | Workroom / Quality Engine | Dispute policy |
| 十六、产品系统架构 | Shared infrastructure | Operational roles | All named modules | Technical governance |
| 十七、全球化架构 | Regional network | Ecosystem partner and compliance | Trust & Compliance | Regional and data policy |
| 十八、商业模式 | Transparent service model | Buyer and partner | Quote / Billing | Commercial policy |
| 十九、首期三个任务工厂 | Task-factory section | Buyer and partner | Vertical task factories | Launch focus |
| 二十、90天落地计划 | Not presented as achievement | Next-step guidance | Delivery roadmap | Program governance |
| 二十一、扩张闸门 | Labeled validation criteria | Operational readiness | Command Center | Expansion policy |
| 二十二、管理指标体系 | VSV and hypothesis framing | Role readiness metrics | Analytics | Measurement standard |
| 二十三、最容易失败的十个地方 | Trust safeguards | Prohibited actions | Risk monitoring | Risk register |
| 最终战略表达 | Final CTA | Completion summary | Product narrative | Strategy charter |
Acceptance checks
- The canonical blueprint below retains all source headings, tables, task codes, formulas, benchmarks, gates, and risk items.
- Every role has complete English and Chinese content.
- Every role page includes purpose, responsibilities, prerequisites, evidence, prohibited actions, readiness, next actions, and blueprint citations.
- Brand assets match the recorded SHA-256 checksums.
- No visible website copy contains an em dash.
- No production-only or sensitive action is presented as operational.
- All routes build and render without client-side errors.
Canonical TABLE AI × OPC Global Blueprint
Unified organization, identity, and settlement module extension
This release extends the public task-value system with a protected operations
workspace. The formula migration source is
ref/经营与分账模型v3.html, pinned at SHA-256
76ca4c88c9c1778f1842a1c03de2a74d38ab2a6d68b7de2116266a4cba740d79.
The source is an operating forecast calculator, not an accounting ledger. Its
examples become golden tests, while production rules add versioning,
participants, idempotency, integer rounding, approval, reversal, disputes, and
append-only provenance.
Identity, organization, and rights holders
- TableUser Hosted OIDC is the authentication authority. Natural persons use
taidas the only stable cross-system key; Logtosub, email, and username must never become business foreign keys. - An Organization is a collaboration and authorization space. Each Split Project belongs to one Organization and may involve several Legal Entities.
- Every Settlement Basis Line belongs to exactly one Legal Entity. Each entity calculates, approves, and posts to an independent subledger; project-level views aggregate but never merge entity funds implicitly.
- A Party is either a natural person bound to TAID or a verified Legal Entity.
Each Party has one global Split Account.
ledger_readyis notpayout_ready. - External invitees become Project Participants only. They do not become Organization members. Invitations use single-use, revocable, hashed tokens, expire after seven days by default, and require reissue when the signed-in email differs.
TableUser is the resource-authorization authority and flay01 stores a versioned
projection only. Formal writes call TableUser in real time and fail closed when
the service is unavailable. Resource grants cover organization,
legal_entity, split_project, and equity_ledger; role names are:
platform:admin, opc:org:owner, opc:org:admin,
opc:settlement:approver, opc:equity:admin,
opc:project:admin|editor|viewer, opc:project:participant, and
opc:auditor.
Platform administrators can provision, suspend, troubleshoot, and draft but cannot alone approve an Organization payable ledger. Organization admins can create projects, publish rules, invite participants, and prepare settlements. An Organization owner or settlement approver approves formal payables. An Organization with only one active approver may self-approve below its own future-effective limit.
Separate planning and settlement contexts
OperatingPlan models services, customers, revenue cycles, collection,
refunds, and cost forecasts; all output is advisory. Settlement consumes only
confirmed actual amounts with source evidence. Forecast refund or collection
rates can never create or alter a payable obligation.
The first immutable template is OperatingFourPoolV3@1:
- Aggregate confirmed revenue or an approved cash basis for one Legal Entity.
- A cash basis first protects all unearned revenue and refund exposure.
- Deduct actual refunds, bad debt, tax, direct cost, and explicit platform fees.
- Calculate positive gross margin and reserve; a loss becomes an entity deficit and never a participant negative payable.
- Allocate acquisition, delivery, product, and platform pools whose ratios sum to exactly 100%.
- Acquisition rights decay to 50% in cooperation year two and 20% from year three; the remainder reflows to the platform pool.
- Product rights expire at the earlier of three years after first delivery or cumulative allocation equal to five times research investment.
- Apply salary-advance offsets, external tax treatment, fixed costs, loan repayment, and the explicit platform waterfall.
- External-member eligibility is controlled by the Rule Version, never by an internal/external hard-coded distinction.
- Normalize participant weights independently within each pool.
Money uses ISO 4217 currency and integer minor units. Rates use exact decimal representation. Largest-remainder allocation guarantees conservation, with stable Party ID as the tie-breaker. Refunds and losses never generate automatic debt; future offsets require a pre-existing, accepted, disputable rule.
Rule, FX, review, and ledger lifecycle
Reference FX initially uses ECB daily reference rates and deterministic EUR triangulation. The UI must label it as a reference rate, not an execution rate. Weekend or holiday dates use the latest published rate not later than the reference date. A system admin or Organization admin may create a manual override for an unapproved run only, with reason, evidence, difference, impact scope, and participant notice. A run freezes raw response, source date, pair, hash, and override; history never recalculates with a current rate.
Rule versions follow draft → published → active → superseded. They affect
future periods only. New and impacted participants must accept the current
version before entering a new period. Historical rules, calculation snapshots,
and approved ledgers are immutable.
Settlement runs follow
draft → calculated → participant_review → ready_for_approval → approved.
Participant review lasts five business days. A disputed line freezes while
undisputed lines continue; expiry records no objection rather than fabricated
approval. Approved runs cannot be edited. Corrections create an AdjustmentRun
with linked reversal or adjustment entries.
Operational equity ledger
Each eligible Legal Entity may have one append-only operational equity ledger
covering classes, holders, units, subscribed and paid amounts, option pools,
vesting, issue, transfer, cancellation, repurchase, and adjustment events.
Derived balances are never overwritten directly. Internal approval may affect
operating rights and profit allocation but always shows
legalRegistrationStatus. Mainland China and Hong Kong have separate
jurisdiction profiles; all other jurisdictions are read-only in this release.
The module is off by default and opens only after an initial holding snapshot is
reconciled. It does not replace a statutory shareholder register, legal filing,
or legal transfer instrument.
REST and MCP contract
REST formal writes require a TableUser OIDC JWT or M2M token, a resource grant,
Idempotency-Key, and If-Match for versioned updates. Responses return the
resource version, audit event ID, and input/calculation hashes.
MCP resources include:
opc://organizations/{id}opc://split-templates/{version}opc://split-projects/{id}opc://settlement-runs/{id}opc://equity-ledgers/{legalEntityId}
MCP tools are limited to read and draft operations:
simulate_operating_plan, draft_split_project, validate_split_rule,
calculate_draft_settlement, explain_settlement,
compare_rule_versions, and draft_invitation. MCP must never publish a rule,
override FX, approve a ledger, post equity events, submit final onboarding, or
execute payment. Every AI or MCP proposal records model, prompt or Skill
version, tool calls, source data, and human confirmation in provenance events.
Data, deployment, and explicit exclusions
Production data remains in Tencent Cloud Nanjing. The target data plane is
PostgreSQL 16 with separate flay01_core and tableuser_authz databases,
separate runtime and migration roles, TLS, encryption at rest, automated
backups, and at least fourteen days of point-in-time recovery. Postgres outbox
workers process calculations, email, and notifications without adding Redis.
The private Sites deployment is a sanitized UI review plane and never connects
to the production ledger.
The pilot capacity target is 100 Organizations, 100 participants per project, and 5,000 lines per run; calculation p95 is under five seconds and in-Organization authorization p95 is under 200ms. Large ledgers use server pagination.
This release does not execute payment, collect identity documents, bank credentials, KYC material, confidential customer uploads, issue credentials, perform tax filing or withholding, provide an arbitrary formula designer, automatically pursue debt, or claim that a UI action completes a legal equity transfer.