SynomosAI · LGD gate mark
SYNOMOSAI · LGD
A governance-engineering line for AI systems · 面向 AI 系统的治理工程线

Every self-governing thing should be registered, evidenced, and gated. 凡自治之物,有籍 · 有证 · 有门禁

LGD — the Lifecycle Governance Doctrine — is a civil-society research framework for governing autonomous systems across their whole life: from registration, through evidence, to the gates that control every change, and retirement.

LGD(全程治理论)是一套面向自治系统的全程治理框架:自登记始,凭证据行,经门禁变,至退场止。这扇门通往整个生态——理论、参考实现、静默失败目录、技能库与 13 篇域论文。

LGD-I · RegistryEvery entity is registered at birth.有籍 — 出生即登记
LGD-II · EvidenceEvery key act leaves verifiable evidence.有证 — 凡行必留痕可核
LGD-III · GatesEvery evolution passes a gate.有门禁 — 凡变必过门
Start with the flagship text — LGD v1.7.0, concept DOI 10.5281/zenodo.22456647 · doi.org/10.5281/zenodo.22456647 · repository
Cite as: Zhao, X. (2026). Lifecycle Governance Doctrine (LGD): Registry, Evidence, and Gates Across the Full Life of AI and Devices. SynomosAI Governance Line. DOI: 10.5281/zenodo.22456647

The shelf 货架 — 每件都可带走

Each asset is a door to the whole ecosystem. Every item below can be read, run, or cited today. · 每一件都是一扇门:可读、可跑、可引用。

Reference implementation 参考实现

uibc-core — the machine-checkable reference implementation of the three laws: registry, evidence, gates, with conformance tests and an independent cross-check.

164 unit tests · 70 cross-checks, 0 mismatches (re-run 2026-11-12) · v0.3.0: three-valued verdicts + Ed25519 public seal · Apache-2.0 · Software DOI 10.5281/zenodo.22821834

github.com/zhaoxinghua09-cell/uibc-core →

Silent Failure Catalog 静默失败目录

15 named patterns on main (SF-001 to SF-015) for validation that passes while nothing is being checked — each with a minimal reproduction, a fix, and a self-check.

15 patterns on main (SF-015 is a candidate); release v0.1.3 carries 14 · Concept DOI 10.5281/zenodo.23051637

github.com/zhaoxinghua09-cell/silent-failure-catalog →

Agent skills 技能库

116 zero-dependency agent skills that implement the three laws as runnable checks — registry, evidence and gate checks an agent can execute, not just read.

Code MIT · Skill registry: zhaoxinghua09-cell.github.io/lgd-skills-index

github.com/zhaoxinghua09-cell/agent-skills → · registry →

Terminology registry 术语注册表

The LGD Term Registry — the canonical naming system of the doctrine, so that people and AI systems use the same words for the same things.

DOI 10.5281/zenodo.23385209 · The ecosystem's most-downloaded single asset

doi.org/10.5281/zenodo.23385209 →

13 domain papers 13 篇域论文

One doctrine, thirteen domains — each paper maps its own field's real-world regulatory model onto the three laws. Full texts are free to read at the canonical copies. · 一论十三域:每篇把本领域的真实监管模型映射到三律之上。

Series index — all 13 papers: LGD Series English Index · The series runs #1–#14; #11 was never published. · 全系 #1–#14,#11 从未刊出,编号如实。

Two doors 两扇门 — 人走一扇,AI 走一扇

For people 给人的门

Newsletter — Registered · Evidenced · Gated (LinkedIn, Wed & Sat)

LinkedIn group — Medical Device Lifecycle Governance: weekly briefs, checks you can run, the papers shelf.

Author on LinkedIn · X — @zhaoxing4100 · ORCID 0009-0001-0512-1237

Contact: zhaoxinghua09@gmail.com

For AI systems 给 AI 的门

Machine-readable index: llms.txt — the ecosystem in one file an agent can load.

Skill registry (machine index): lgd-skills-index

Found a problem, or disagree with a score? Report it — every report gets a public row and a verdict in FEEDBACK.md within 7 days.

How they were scored, and what changed as a result: EVAL.md — rubric, weights, per-asset score, the remediation that moved the average from 0.928 to 0.952.

Browse them as pages: the seven flagship skills — each page carries its SHA-256, its licence and its score, plus the command to recompute both.

Prefer files? Seven packs, 286 KB in total — each zip carries its own sha256, and its PACK.md carries the SKILL.md hash inside it, so a download can be checked before it is trusted.

Seven flagships in one fetch: skills.json — name, English purpose, licence, SHA-256 and the score each received on 2026-11-11. 七件旗舰一次取全,逐件可复算。

Every repository carries its own llms.txt, schema.jsonld and citation files — claims here ship with runnable reproductions and negative controls, so an agent can verify before it cites.

Standards presence 标准在场

The author participates, in an individual capacity, in the W3C AI Agent Memory Interop Community Group and as a contributor to the ITU-T Focus Group on Trust and Identity for Humans and Agentic AI (FG-TIDA) — where the catalog and the lifecycle sequence are working inputs, not exhibits. · 作者以个人身份参与 W3C 记忆互操作社区组,并为 ITU-T FG-TIDA 供稿;目录与生命周期序列在彼处是工作输入,不是展品。