意图漂移对齐与意图漂移度:SEI/ORBIT 框架下的人机协作跨层指标

本文提出两项新指标——意图漂移对齐(Intent Drift Alignment, IDA)与意图漂移度(Intent Drift Score, IDS),用于量化人机协作中的跨层意图对齐问题。现有 RCS-µ 指标(残差率、修复延迟、切换稳定性、复现度)主要聚焦于微观残差修正与任务切换,而 IDA/IDS 则扩展至多轮交互中的意图漂移捕捉与对齐。IDS 与 RD/CL/SS/RS 存在交互关系:高残差率或修复延迟往往导致 IDS 升高,而高 IDA 可抵消部分漂移效应。 基于 SEI v1.4 与 ORBIT v1.3 框架,本文提出的指标为微观残差机制与宏观认知增益(CAf)之间建立了桥梁,使得在非线性工作流中,对元认知、语义残留与组织共振动态的系统性评估成为可能。该研究为 RCS-µ 提供补充,确立了测量与缓解意图漂移的方法论,同时也是可独立应用的通用分析工具。 This paper introduces two novel metrics—Intent Drift Alignment (IDA) and Intent Drift Score (IDS)—for quantifying cross-layer alignment in human–AI collaboration. While existing RCS-µ indicators (Residual Density, Correction Latency, Switch Stability, Reproducibility Score) address micro-level task switching and residual correction, IDA/IDS extend the framework to capture intent-level drift and alignment across multi-turn interactions. Grounded in SEI v1.4 and ORBIT v1.3, the proposed metrics provide a bridge between micro residual mechanisms and macro collaborative amplification (CAf), enabling systematic evaluation of meta-cognition, semantic residue, and resonance dynamics in nonlinear task flows. This contribution complements RCS-µ by establishing a methodology to measure and mitigate intent drift, offering an independent, generalizable tool for research and system design.