Independent expert building machine-checkable, verifier-side primitives — validation that fails visibly instead of passing silently.
Why follow: everything here is falsifiable by construction — every claim ships with a runnable reproduction and a negative control, so you can prove me wrong before you trust me. If you care about verifier-side failure semantics, agent governance, or how evidence survives context handoffs, this is the workshop.
Work submitted to ITU-T FG-TIDA (Trustworthy Identification for Autonomous Agents), written in an independent-expert capacity:
silent-failure-catalog→ submitted as a verifier-side challenge reference inFG-TIDA/use-casesPR #23: a named catalog of 14 silent-failure modes — validation that passes while nothing is checked — each with a runnable reproduction, a fix, and a negative control. Zenodo-archived (concept record 23051637, resolves to the latest version); canonical release tag per the repo's citation note.- Theme #7 — verifier-side requirements and failure semantics: discussion contributions on what a relying party may verify offline, and how verdicts stay re-verifiable without making every party a verifier. The catalog is also referenced in theme #6 (intent-based security policies).
FG-TIDA/wg1-ra-aaiPR #2 — reference-architecture chapter 4: offline re-verifiability of agent attestation results, aligned with RFC 9334 (RATS) role separation.- Merged:
FG-TIDA/use-casesPR #18 — completing the submissions index.
| Layer | Repos | One line |
|---|---|---|
| Theory | lgd-theory · xlgd · xcgs-manifesto | Machine-checkable lifecycle model: registry · evidence · gates. |
| Protocol | uibc-core · uibc-competition · silent-failure-catalog · assayance | Executable evidence/registry/validator core, the UIBC-MEM benchmark, and the verifier-side silent-failure challenge reference. |
| Applications | medxpert-reg-kb · medxpert-reg-connector · medxpert-skills · agent-skills · agent-memory-service · lgd-hub | Product integrations: medical-device compliance tooling, agent skill packs, memory service, unified entry hub. |
Four names, four layers — used consistently: XLGD (umbrella · X is a distinction mark), LGD (theory), UIBC (executable protocol), MedXpert (medical-device compliance applications — parallel to, not inside, the umbrella). All four are project identifiers only — no entity or trademark registration filed.
graph LR
A[silent-failure-catalog<br/>verifier-side challenge ref] -->|negative controls| B[FG-TIDA #6 / #7<br/>verifier-side failure semantics]
C[uibc-core<br/>evidence + registry] --> D[LGD lifecycle model]
A -->|C7 checks-in-path| E[AVS 0.3 candidates]
- 🔍 New to the idea? Start with silent-failure-catalog — read one mode, run its repro, kill it with the fix.
- 🛠 Building agent systems? uibc-core gives you registry / evidence / gates you can wire into your own lifecycle (Apache-2.0).
- 🏛 Tracking FG-TIDA? Watch this account — FG-TIDA submissions and follow-ups land here first.
- 🌐 Prefer plain language? medxpert.cn — the medical-device compliance line, in prose.
- ⭐ Star or 👀 watch a repo if a mode / primitive saves you a bug — it tells me what to deepen next.
- 🐛 Found a claim that doesn't reproduce? That is the best issue to open. Negative controls are the house style.
- 💬 Ask me about verifier-side validation, agent evidence, or medical-device compliance tooling.
- ➕ Follow for standards-track work (ITU-T FG-TIDA), new silent-failure modes, and machine-checkable governance primitives.
- Email: zhaoxinghua09@gmail.com
- ORCID: 0009-0001-0512-1237
- Site: https://medxpert.cn
- LinkedIn: https://www.linkedin.com/in/steven-zhao-9156a197
- X (Twitter): @zhaoxing32035
All account content is All Rights Reserved (text, methodology, theory terms, specs, examples, scripts). Code is released under MIT / Apache-2.0 / AGPL-3.0 per repo. Attribution-only citation is permitted with named source + link + rights holder Zhao Xinghua / Steven Zhao·China.
Brand status: MedXpert, SynomosAI, LGD, UIBC, Nomos, XLGD are project identifiers only — no entity or trademark registration filed. Their appearance is source identification, not a claim of legal-entity or trademark rights.
Maintained by the rights holder. Public information only; not legal, regulatory, or registration advice.
本账号聚焦于可机器核验的验证方构件——让校验在失效时可见，而不是静默通过。核心理念：每一句声称都自带可运行复现与反向对照，先证伪、后采信。
以独立专家身份向 ITU-T FG-TIDA（可自治智能体可信识别焦点组） 提交的工作：
- silent-failure-catalog —— 14 种"静默失败"模式的命名目录，每条含可运行复现、修法、反向对照。已作为验证方挑战参考实现提交（
FG-TIDA/use-casesPR #23）；Zenodo 归档（概念记录 23051637，解析到最新版本），规范引用 tag 以仓内引用注记为准。 - 主题 #7 验证方失败语义 的讨论贡献（依赖方可离线核验什么、判决如何保持可复验而不把每个人都变成验证者）；catalog 亦被 主题 #6（意图安全策略）引用；
wg1-ra-aaiPR #2 参考架构第 4 章：离线可复验性（对齐 RFC 9334 角色分离）。 - 已合并：
use-casesPR #18（补全投稿索引）。
理论层（lgd-theory · xlgd · xcgs-manifesto）→ 协议层（uibc-core · uibc-competition · silent-failure-catalog · assayance）→ 应用层（medxpert 系列 · agent-skills · agent-memory-service · lgd-hub）。
| 术语 | 层 | 含义 |
|---|---|---|
| XLGD | 伞 | 理论与协议线伞层标识；X 是区别符，用于标明来源 |
| LGD | 理论 | 全生命周期模型：凡自治之物，有籍·有证·有门禁 |
| UIBC | 协议 | 可执行协议：registry / evidence / gates 参考实现 + 基准 |
| MedXpert | 应用 | 医疗器械合规产品线——与 XLGD 伞并行，不在伞内 |
- 觉得某个模式帮你避了坑 → 给仓点 ⭐/👀，这决定我深挖哪条线。
- 发现哪句声称复现不了 → 开 issue，这是本仓最欢迎的贡献（反向对照是本仓家风）。
- ➕ 关注本账号：ITU-T FG-TIDA 标准线动态、新增静默失败模式、可机器核验的治理构件都会先落在这里。
- 邮箱：zhaoxinghua09@gmail.com ｜ ORCID：0009-0001-0512-1237 ｜ 官网：https://medxpert.cn
- LinkedIn：https://www.linkedin.com/in/steven-zhao-9156a197 ｜ X：@zhaoxing32035
权利宣告：本账号全部内容保留所有权利；代码依各仓许可证（MIT/Apache-2.0/AGPL-3.0）使用。署名引用须标注权利人「赵兴华 / Steven Zhao·China」。MedXpert、SynomosAI、LGD、UIBC、XLGD 等仅为项目标识，均未申请实体/商标注册。