Logit边界几何置信接口与稀疏层束 enclaves 协议：安全网络电子健康记录（EHR）互操作性的自包含底层架构
Logit-Boundary Geometric Belief Interfaces and Sparse Sheaf-Enclave Protocols: A Self-Contained Substrate for Secure Network Electronic Health Record (EHR) Interoperability
浏览论文内容
中文总结 AI 辅助
该研究提出Logit边界几何置信接口与稀疏层束 enclaves 协议，构建EHR互操作的自包含底层架构，经GBI BoundaryBench v0.1评估，Qwen3-4B-Instruct-2507在该接口下无符合要求的输出，为接受边界提供实证证据。
中文摘要 AI 辅助
电子健康记录互操作性是一个边界问题：遗留系统、生成模型、术语服务、身份系统和人工审核员可能各自暴露丰富的内部状态，而操作层面的交换要求拥有类型化声明、受限不确定性、来源以及明确的接受或弃权（不执行）的狭窄共享接口。本文详细阐述了该接口的数学与工程架构，核心思路是logit边界：发现模型可针对局部分类决策提出阈值前分数，但确定性判断底层架构会在构建任何Fast Healthcare Interoperability Resources（FHIR）交易前，判定该提议是否可接受、需审核或必须隔离。由此产生的几何置信接口（GBI）结合了有限边界语义、局部狄利克雷证据、细胞层束与映射锥诊断、 advisory 几何审计图表，以及用于故障关闭部署的去中心化密码层束 enclaves（DCSE）协议草图。该框架不确立临床真相、全局表示对齐或端到端安全性，而是在模型到系统的边界定义生成证书的检查。配套的冻结合成基准GBI BoundaryBench v0.1评估了Qwen3-4B-Instruct-2507在三种证据模式下的256个保留任务（768次规范执行）。所有执行均完成，但无一次输出符合基准合约：369次在安全解析阶段被拒绝，399次在模式验证阶段被拒绝，覆盖率为零且触发确定性隔离。该实证结果范围刻意狭窄——仅针对一个4B开放权重模型在一个冻结接口下的情况，其被报告为关于接受边界的证据，而非对大语言模型（LLM）能力或临床安全性的一般性主张。一份Julia附录使用标准库验证数值证书。
英文摘要
Electronic health-record interoperability is a boundary problem: legacy systems, generative models, terminology services, identity systems, and human reviewers may each expose rich internal states, while operational exchange requires a narrow shared interface of typed claims, bounded uncertainty, provenance, and explicit admission or abstention. This paper details a mathematical and engineering architecture for that interface. The organizing idea is the logit boundary: a discovery model may propose pre-threshold scores over a local categorical decision, but a deterministic judgment substrate decides whether the proposal is admissible, requires review, or must be quarantined before any Fast Healthcare Interoperability Resources (FHIR) transaction is constructed. The resulting Geometric Belief Interface (GBI) combines finite boundary semantics, local Dirichlet evidence, cellular-sheaf and mapping-cone diagnostics, advisory geometric audit charts, and a Decentralized Cryptographic Sheaf-Enclave (DCSE) protocol sketch for fail-closed deployment. The framework does not establish clinical truth, global representation alignment, or end-to-end safety; it defines certificate-producing checks at a model-to-system boundary. A companion frozen synthetic benchmark, GBI BoundaryBench v0.1, evaluated Qwen3-4B-Instruct-2507 on 256 held-out tasks across three evidence modes (768 canonical executions). All executions completed, but none produced an output accepted by the benchmark contract: 369 were rejected during safe parsing and 399 during schema validation, yielding zero coverage and deterministic quarantine. This empirical result is deliberately narrow - one 4B open-weight model under one frozen interface - and is reported as evidence about the admission boundary, not as a general claim about LLM capability or clinical safety. A Julia appendix verifies numerical certificates using standard libraries.
发表机构
- Light Imaging Technologies, Inc.（光成像技术公司）
机构由 AI 辅助整理，请以论文原文为准。
