本仓库使用自定义许可，不是 MIT / Apache-2.0。平台显示为
Other（NOASSERTION）， 属识别算法的正常结果，不代表本仓库处于无许可状态。
- 权利状态：全部内容保留所有权利（All Rights Reserved）。未经书面许可， 不得复制、改编、再分发、公开传播或用于衍生作品。
- 可否引用：可以。允许在注明出处的前提下引用与学术、公共讨论；
引用时请同时标注仓库名、原文链接
https://github.com/zhaoxinghua09-cell/culpachain与权利人「赵兴华 / Steven Zhao·China」。 - 完整条款：见仓库根目录 LICENSE。
- 联系：zhaoxinghua06@126.com ｜ ORCID 0009-0001-0512-1237
- 品牌状态限定：MedXpert、SynomosAI、LGD 等为相关项目标识， 均未申请实体注册、未申请商标注册；出现仅作来源标识， 不构成对法人实体或商标权的任何主张。
- 免责：本仓库内容不构成法规意见或注册代理服务；关键数据以监管机构最新发布为准。
多智能体问责治理 · Multi-Agent Accountability
A coined theoretic term in the LGD (凡自治之物 · 全程治理论) theory stack. Theory-layer zero-collision: the descriptive English name is settled by others; the coined term is ours.
Culpachain — the accountability chain across a chain of autonomous agents: when several agents act in sequence or in concert, which one answers?
Single-agent governance has a simple answer — the agent's owner is liable. Multi-agent systems break this. Agent A plans, agent B executes, agent C retries after failure, and the outcome is an emergent product of the chain. The harm is real; the culprit is distributed across the chain.
Culpachain holds that accountability must be chain-resolved, not agent-resolved:
- Registry (LGD-I) — every link in the chain is registered with its role, authority scope, and handoff boundary.
- Evidence (LGD-II) — every handoff leaves a trace: input, decision, output, and the delegation act itself.
- Gate (LGD-III) — responsibility is gated at handoff: a link may not accept a task whose authority it cannot discharge, and may not pass on a task whose outcome it cannot account for.
Multi-Agent Accountability is already claimed in the literature — accountability boundaries theory and the Accountable AI Framework (AAF) occupy the descriptive phrase. The theory layer is closed to newcomers using that name.
Culpachain names the structural object — the chain along which culpability is distributed and must be re-resolved — rather than competing for the phrase. The descriptive subtitle carries search; the coined term carries the claim.
| Angle ID | TH-ANG-003 |
| Domain code | ANG (cross-domain governance angles) |
| Parent | TH-ANG-000 跨域治理角度矩阵 (Cross-Domain Governance Angle Matrix) |
| Related | LGD-II 有证 (six evidence artifact classes) · audit-trail-keeper |
| Collision verdict | Theory layer 🔴🔴 (accountability boundaries theory + AAF) → Coined term 🟢 |
| Domain | Referent | Mapping |
|---|---|---|
| Agent protocols | ADCS specification | delegation and identity in agent chains |
| Agent messaging | IETF drafts (3) on agent interoperation | handoff trace format |
| Frontier labs | DeepMind multi-agent safety work | emergent-behaviour accountability |
| Runtime monitoring | SentinelAgent | chain-level observation |
| Product liability | joint-and-several liability doctrines | the legal analogue for distributed fault |
If you use or cite this term, please reference:
Zhao, S. (2026). Culpachain: Multi-Agent Accountability — LGD Angle Matrix #3. SynomosAI. https://github.com/zhaoxinghua09-cell/culpachain
- LGD — 凡自治之物 · 全程治理论 (Lifecycle Governance Doctrine) — the parent framework: 有籍 · 有证 · 有门禁 (registered · evidenced · gated) across the full life of an autonomous thing.
- Cross-domain angle matrix — 5 angles × 13+ domains; this repo is angle #3.
- Companion angles: Terminance · Mnemoship · Runtigil · Assurability.
赵兴华 / Steven Zhao·China ORCID 0009-0001-0512-1237 · https://medxpert.cn
Proposed and named 2026-09-10. Co-created with Lao Paidang (老拍档), AI research partner. Human authorship, judgment, and final responsibility rest with the author.
Theory names and definitions — all rights reserved by the author. Any citation, use, or extension shall trace back to this source.