# The Lifecycle Governance Doctrine (LGD): Registry, Evidence, and Gates Across the Full Life of Autonomous Systems

**English flagship edition · v1.0 · 2026-09-06**

> **Proposed by**: MedXpert × SynomosAI — a regulated-industry practice line (medical devices) and a cross-domain governance thought line (AI & autonomous systems).
> **Author**: Zhao Xinghua / Steven Zhao · China · ORCID 0009-0001-0512-1237 · https://medxpert.cn
> **DOI (concept, always latest)**: https://doi.org/10.5281/zenodo.22456647
> **License**: CC BY 4.0 — cite the author when sharing.
> **One-liner**: Every self-governing entity — AI model, software-enabled medical device, autonomous agent — shall carry a **registry** (who it is), **evidence** (what it did and why), and **evolution gates** (who allowed it to change), traceable, provable, and accountable **across its full lifecycle, from birth to retirement**.

---

## Abstract

The regulated world already possesses a mature model of full-lifecycle governance. Its most representative example is **medical device lifecycle regulation**: classification, design control, risk management, registration, clinical evaluation, quality-system assurance, post-market surveillance, and retirement — with regulatory requirements, evidence duties, and release gates at every step.

Mainstream AI-governance efforts remain mostly **single-point**: some address *identity* ("who is it" — passports, registries, algorithm filing), some address *evidence* ("prove it" — audit trails, explainability, transparency), and some address *gates* ("who lets it grow" — change control, conformity assessment, alignment review). **Almost no one has chained these points into one lifecycle thread.**

LGD claims the missing dimension: governance is not a registration event, nor a policy paper, but an **unbroken chain from birth to retirement**. Every self-governing entity should be governed along that chain — and the medical-device paradigm provides the closest real-world reference model for how to do it.

## 1. The Doctrine — three laws

> **LGD-I · Registry (有籍)** — Every autonomous entity is registered at birth: identity, creator, governance owner, and records. Register everything; no exemption by size or private use ("凡造必登" — anything created is registered).
>
> **LGD-II · Evidence (有证)** — Every consequential act leaves evidence: decision logs, risk registers, model/device records, change records, human-oversight proof, incident ledgers. Claims carry source and signature; records are replayable.
>
> **LGD-III · Gates (有门禁)** — Every evolution passes a gate: capability upgrade, re-release, permission expansion go through trigger → assessment → release → retrospective review. High-stakes authority belongs to humans; machines request, humans authorize.

**Corollaries**
1. **Isomorphic reference model**: medical-device lifecycle regulation is the closest *real-world* template for governing AI — the gap in AI governance is structurally similar to an already-solved problem in devices.
2. **Lifecycle stage = executable instrument**: each stage maps to a checkable tool or evidence generator; tools are not loose parts but gears on the chain.
3. **Cross-domain generalization**: the same spine instantiates in every regulated domain — financial AI, electronic evidence, digital government, on-chain assets, autonomous driving, data elements, low-altitude systems, embodied AI, and more.

## 2. The reference model (medical device lifecycle)

| Stage | Public anchor (examples) | LGD transfer |
|---|---|---|
| Classification | CN Medical Device Regulation & classification catalog / EU MDR 2017/745 rules / US FDA product classification | "What risk class am I?" first |
| Design control | ISO 13485 design & development / FDA 21 CFR 820.30 | Traceable, change-controlled development |
| Risk management | ISO 14971 | Living risk register |
| Registration | NMPA / FDA / EU pathways | First release gate |
| Clinical / performance | MEDDEV / FDA clinical evidence framework | Claims need evidence |
| Quality & production | GMP / ISO 13485 systems | Auditable process |
| Post-market | PMS / vigilance / PSUR | Incident ledgers, active detection |
| Retirement | Device decommission / UDI cancellation / data destruction | Final gate — chain closes |

## 3. Putting any entity under LGD — three checks

1. **Registered?** Can it answer "who is it, who created it, who governs it"? If not, register it first.
2. **Evidenced?** Can every consequential decision produce artifacts (logs, records, oversight proof)? If not, build the evidence chain.
3. **Gated?** Did it pass assessment before growing (upgrade/re-release/permission expansion)? If not, stop its release.

A link that fails any check is a *stated gap* — LGD never claims fake completeness.

## 4. Implementation status

- **Executable instrument family** across lifecycle stages (registration-path calculators, classification, change-gate assessment, clinical-evidence management, post-market planning, retirement checks) ships as callable compliance tooling; see the MedXpert tool suite.
- **User-facing applications** (self-check consoles, AI-device compliance checkers, reference-verification bench, identity-anchor generator, algorithm-passport generator, AI-creates-AI compliance checker) make the three laws everyday actions.
- **Protocol layer**: structured schemas (algorithm passport and related) express registry–evidence–gates in machine-readable form.
- Full domain series: see `docs/` — financial AI ×2, electronic evidence, digital government, on-chain assets, autonomous driving, data elements, low-altitude/UAS, embodied AI, industrial/OT, biomedical data, education.

## 5. Global alignment

LGD is engineered to sit *on top of* the existing global standards network rather than compete with it — registry/evidence/gates map onto ISO/IEC 42001, EU AI Act (2024/1689), ISO 13485/14971, EU MDR/IVDR, FDA SaMD/TPLC/PCCP, NIST AI RMF, UNESCO Recommendation on the Ethics of AI, FATF Recommendations, and others. See `docs/LGD-Global-Standards-Alignment-v1.0.md`.

## 6. Suggested citation

> Zhao, X. (2026). *The Lifecycle Governance Doctrine (LGD): Registry, Evidence, and Gates Across the Full Life of Autonomous Systems* [Preprint]. MedXpert × SynomosAI. ORCID 0009-0001-0512-1237. DOI: https://doi.org/10.5281/zenodo.22456647

## 7. Honest boundaries

LGD is a **civil-society framework**, not legal or regulatory advice, not a certification, not a compliance guarantee. Device-regulation references are structural alignment, not an equivalence claim. LGD is the organizing layer over independent doctrines (identity, evidence, gates, real-world-reference, anti-fabrication) — it does not replace them. Where a lifecycle link has no tooling, the gap is stated.

---

© MedXpert × SynomosAI · CC BY 4.0 · Cite: Zhao, X. (2026). DOI 10.5281/zenodo.22456647