Beyond the Attention Stability Boundary: Agentic Self-Synthesizing Reasoning Protocols

2026-04-27

AI Summary

This work addresses the challenge of large language models failing to respond to mid-conversation instructions and losing goal-directed behavior in multi-turn, nonlinear dialogues due to attention latch—a phenomenon formally characterized for the first time in this study. To overcome the limitations imposed by attentional stability, the authors propose the Self-Synthesized Reasoning Protocol (SSRP), which decouples high-level architectural planning (Architect) from turn-by-turn execution (Executive). The framework integrates a metacognitive architecture, introduces novel evaluation metrics such as Aggregated Pivot Accuracy (APA), and leverages information-bottleneck-guided granularity ablation, recursive reflection baselines, and program completeness auditing. Experimental results demonstrate that SSRP achieves up to a 715-fold improvement in robustness on models including GPT-5.4 and significantly outperforms baselines on MultiWOZ 2.2, attaining 98.8% program completeness.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

As LLM agents transition to autonomous digital coworkers, maintaining deterministic goal-directedness in non-linear multi-turn conversations emerged as an architectural bottleneck. We identify and formalize a systemic failure mode termed the Attention Latch in decoder-only autoregressive Transformers. This phenomenon, a behavioral manifestation of Information Over-squashing, occurs when the cumulative probabilistic weight of historical context overrides mid-task updates, causing agents to remain anchored to obsolete constraints despite explicit contradictory instructions. We propose Self-Synthesizing Reasoning Protocols (SSRP), a metacognitive framework that implements a discrete separation between high-level architectural planning (Architect) and turn-by-turn procedural execution (Executive). We evaluate SSRP across 9K trajectories using the MultiWOZ 2.2 dataset and the Aggregate Pivot Accuracy (APA), a novel metric we validate by mapping its scores to the U-shaped 'Lost in the Middle' curve. We present 3 experimental tiers: a shallow recency-based retrieval pilot, a high-entropy SOP, and a semantic hijacked 3-hop Multi-Fact Synthesis task. Our results empirically locate the Attention Stability Boundary, where stateless Vanilla ReAct baselines for GPT 5.4 collapse to 0.1% success while SSRP achieves a 715X Resilience Lift. We demonstrate statistically significant gains across Gemini 3.1 Pro, Claude Sonnet 4.6 and DeepSeek V3.2. Audits confirm SSRP necessity by proving attentional lapse via a recursive reflexion baseline (100% success); decoupling the latch from positional bias through equidistant stress testing (90% accuracy); and formalizing SSRP via the Information Bottleneck principle and granularity ablations. Procedural Integrity audit (98.8% adherence) reveals a Grounding Paradox where high-stability models fail by refusing to hallucinate under retrieval-reasoning contamination.

Problem

• Attention Latch
• Information Over-squashing
• goal-directedness
• multi-turn conversations
• autoregressive Transformers

Innovation

• Self-Synthesizing Reasoning Protocols
• Attention Latch
• Information Over-squashing
• Metacognitive Framework
• Aggregate Pivot Accuracy