Box Maze: A Process-Control Architecture for Reliable LLM Reasoning

TL;DR

The paper introduces the Box Maze framework, a process-control architecture with explicit layers to improve the reliability and consistency of LLM reasoning, especially under adversarial conditions.

Contribution

It proposes a novel layered architecture for LLM reasoning that explicitly enforces process integrity, addressing limitations of existing behavioral safety methods.

Findings

Boundary failure rates reduced from ~40% to below 1% in simulations.

Explicit control layers improve consistency in adversarial scenarios.

Preliminary simulation results support the potential of process-level control.

Abstract

Large language models (LLMs) demonstrate strong generative capabilities but remain vulnerable to hallucination and unreliable reasoning under adversarial prompting. Existing safety approaches -- such as reinforcement learning from human feedback (RLHF) and output filtering -- primarily operate at the behavioral level and may lack explicit architectural mechanisms for enforcing reasoning process integrity. This paper proposes the Box Maze framework, a conceptual process-control architecture that decomposes LLM reasoning into three explicit layers: memory grounding, structured inference, and boundary enforcement. We introduce preliminary simulation-based evaluation involving progressive boundary erosion scenarios across multiple heterogeneous LLM systems (DeepSeek-V3, Doubao, Qwen). Results from n=50 adversarial scenarios suggest that explicit cognitive control layers may improve consistency in boundary maintenance, with architectural constraints reducing boundary failure rates from approximately 40% (baseline RLHF) to below 1% under adversarial conditions. While current validation is simulation-based, these preliminary results indicate that process-level control may offer a promising direction for improving reliability in large language model reasoning.