TDCA Weekly 特刊：制度诊断 10 篇论丛摘要合集（AI 审计/Skill/思维协议/许可/溯源/安全/框架/ACPs/算力/Token）

[henyi-tdca]

TDCA Weekly 特刊：制度诊断 10 篇论丛摘要合集

Special Issue: Institutional Diagnosis Series — 10 Essay Abstracts (Bilingual)

本期特刊覆盖 10 个行业热点：AI 审计降噪 / Skill 赛道 / 思维协议 / 开源许可 / AI 溯源 / Agent 安全 / Agent 框架 / 国标互操作 / 算力经济 / Token 经济。每篇给出核心论点摘要（中英双语）。全部内容为研究性观点（simulated/ID92 标注口径沿用本仓库惯例）。

This special issue maps 10 industry hotspots to TDCA's institutional-diagnosis lens. Each entry is a bilingual abstract of one essay from the series.

---

1. 从机器证明到制度大模型 ｜ From Machine Proof to the Institutional Model

热点：AI 审计降噪

本文提出「制度孪生」与「函数大模型」两个概念，经全球文献考古确认二者尚属空白。核心论点：AI 治理的下一步不是更强的模型，而是可被机器验证的制度层——把治理规则以形式化证明实例化为可执行函数，审计噪音才能在源头被消除。TDCA 是该路径的首个自证工程案例，给出了从「机器证明」到「制度大模型」的完整论证链。

This essay introduces two concepts — institutional twin and function-scale model — verified as gaps in the global literature. Its thesis: the next step of AI governance is not stronger models but a machine-verifiable institutional layer, where governance rules are instantiated as executable functions via formal proof, eliminating audit noise at the source. TDCA is the first self-demonstrating engineering case of this path.

2. 蒸馏 Skill vs 编译思维协议 ｜ Distilling Skills vs Compiling Thinking Protocols

热点：Skill 赛道

批判「把书蒸馏为 Skill」的流行路径：L1 知识未经确权直接进入 L2 流转，构成配置权掠夺；蒸馏本质上是认知暴力的结构化。主张以「编译」替代「蒸馏」——思维协议（COP）必须先完成确权（MOU）才可进入调用与流转，知识贡献者的配置权在制度层被保留。

A critique of the popular "distill books into Skills" approach: L1 knowledge flowing into L2 circulation without rights confirmation constitutes configuration-rights appropriation; distillation is the structuralization of cognitive violence. The alternative is compilation — a Thinking Protocol (COP) must complete rights confirmation (MOU) before it can be invoked or circulated, preserving contributors' configuration rights at the institutional layer.

3. 默会知识的制度协议化 ｜ Institutional Protocolization of Tacit Knowledge

热点：思维协议

企业 AI 落地困境的根源是默会资产被系统性忽略。文章指出波兰尼意义上的「两套系统」与黑箱反馈闭环同构；出路不是知识提取，而是配置权编译——把默会知识转化为可确权、可计价、可分润的制度协议，让「会做但说不出」的经验进入可信协作链。

The root of enterprise AI's landing difficulties is the systematic neglect of tacit assets. The Polanyian "two systems" are isomorphic to the black-box feedback loop; the way out is not knowledge extraction but configuration-rights compilation — turning tacit knowledge into institutional protocols that can be rights-confirmed, priced, and revenue-shared, bringing "know-how that cannot be articulated" into the trusted collaboration chain.

4. 开源许可证的制度经济学重读 ｜ Re-reading Open-Source Licenses via Institutional Economics

热点：开源许可

基于 arXiv:2507.05270 综述重读开源许可证体系：许可证本质上是配置权协议的教科书级案例——它不转移所有权，只精确配置使用权。TDCA 文档以 CC BY-SA 4.0 发布，正是这一理论自觉的实践：用制度工具而非技术围墙来组织协作。

Re-reading the open-source license system through the lens of arXiv:2507.05270: a license is essentially a textbook case of a configuration-rights protocol — it transfers no ownership, only precisely configured usage rights. Publishing TDCA documents under CC BY-SA 4.0 is the practical expression of this theoretical self-awareness: organizing collaboration with institutional tools rather than technical walls.

5. 制度水印与技术水印 ｜ Institutional Watermarks vs Technical Watermarks

热点：AI 溯源

从 Anthropic 强制水印事件切入：技术水印是 H→M 的单向监控逻辑，制度水印（NCA 存证链）则是协作确权逻辑——前者回答「这是不是 AI 生成的」，后者回答「这次生成由谁、在什么授权下、承担什么责任」。文章给出 ID29 的再验证：溯源的终点不是识别，而是可归责。

Starting from Anthropic's mandatory watermarking: technical watermarks encode an H→M one-way monitoring logic, while institutional watermarks (the NCA attestation chain) encode collaborative rights-confirmation — the former asks "was this AI-generated?", the latter asks "by whom, under what authorization, and with what accountability." A re-validation of ID29: the endpoint of provenance is not identification but attributability.

6. Agent 安全指南的制度盲区诊断 ｜ Diagnosing the Institutional Blind Spot in Agent Security Guidelines

热点：Agent 安全

诊断腾讯云×信通院与 TC260 两份 Agent 安全指南：二者均为技术安全能力目录，覆盖漏洞、注入、越权等层面，但缺失 L2 配置权语言——「影子智能体」这类风险（智能体在未授权配置下行动）无法在纯技术框架内被表达，更无法被计价与归责。

A diagnosis of two Agent security guidelines (Tencent Cloud × CAICT, and TC260): both are catalogs of technical security capabilities covering vulnerabilities, injection, and privilege escalation, but both lack an L2 configuration-rights vocabulary — risks like the "shadow agent" (an agent acting under unauthorized configurations) can neither be expressed nor priced nor attributed within a purely technical framework.

7. 飞连对比诊断 ｜ Comparative Diagnosis of Volcano Engine Feilian

热点：Agent 框架（闭源）

字节飞连的零信任体系是技术防御层的最优解之一，但制度层缺席：NHI（非人类身份）管理解决了「谁在访问」，未解决「以什么配置权访问、收益如何分配」。本文同时是「闭源不挂载但可诊断」方法论的示例——不触碰源码，仅做制度层对照分析。

Volcano Engine's Feilian zero-trust system is among the best technical defense layers, yet the institutional layer is absent: NHI (Non-Human Identity) management answers "who is accessing" but not "with what configuration rights, and how proceeds are shared." The essay also demonstrates the "no mounting on closed-source, but diagnosis is allowed" methodology — institutional-layer comparison without touching source code.

8. ACPs 深度对比 ｜ A Deep Comparison with ACPs (GB/Z 185-2026)

热点：国标互操作

北邮 ACPs（GB/Z 185-2026）提供了可用的通信与身份底座，但制度交付与结算层缺席。文章对比「协作智能」与「群体智能」两条路线：前者要求每个智能体携带可确权、可结算的协作协议，后者侧重规模涌现——TDCA 站在前者一侧。

BUPT's ACPs (GB/Z 185-2026) provides a usable communication and identity substrate, but institutional delivery and settlement layers are absent. The essay contrasts "collaborative intelligence" with "swarm intelligence": the former requires every agent to carry rights-confirmable, settleable collaboration protocols; the latter emphasizes emergence at scale. TDCA sides with the former.

9. 算力空转危机 ｜ The Idle-Compute Crisis

热点：算力经济

行业 GPU 利用率普遍不足 20% 的根源不在技术而在制度：L1 所有权与 L2 配置权市场脱节，算力只能被囤积而不能被调度。文章主张以配置权调度替代所有权囤积——这是算力经济从「拥有逻辑」转向「调用逻辑」的制度解法。

The root of sub-20% industry GPU utilization is institutional, not technical: L1 ownership and the L2 configuration-rights market are disconnected, so compute can be hoarded but not dispatched. The essay advocates replacing ownership-hoarding with configuration-rights dispatch — the institutional solution for shifting the compute economy from an ownership logic to an invocation logic.

10. Token 经济幻觉 ｜ The Token-Economy Illusion

热点：Token/AIGC

工业 AI 三幻觉、Token 四部曲与 AIGC 价值稀释同源：效用未闭环——Token 被生产、被消耗，但从未被确权与结算，「使用量」因此无法沉淀为「价值量」。「调用即确权」是走出幻觉的制度路径：每一次调用同时是一次 NCA 存证，价值在调用点直接锚定。

The three illusions of industrial AI, the four-act Token drama, and AIGC value dilution share one root: utility is not closed-looped — tokens are produced and consumed but never rights-confirmed or settled, so "usage volume" never settles into "value volume." "Invocation-as-attestation" is the institutional way out: every call is simultaneously an NCA attestation, anchoring value directly at the point of invocation.

---

收尾 ｜ Closing

本系列与已发布的《思维协议库认知资产调用规则》（#35）及《思维协议基础设施白皮书》（docs/whitepapers/TDCA-COG-WP-001.md）构成同一条主线：行业热点的问题清单，都能在制度层找到对应的解法提案。欢迎就任何一篇展开讨论。

This series, together with the Cognitive-Asset Invocation Rules (#35) and the Thinking-Protocol Infrastructure Whitepaper, forms one continuous thread: every hotspot on the industry's problem list has a corresponding institutional-layer proposal. Discussion on any entry is welcome.

— TDCA Weekly 特刊（simulated/ID92）｜ 2026-08-28