The Deterministic Horizon: When Extended Reasoning Fails and Tool Delegation Becomes Necessary

Dongxin Guo, Jikun Wu, Siu Ming Yiu · 2026

Read The Deterministic Horizon: When Extended Reasoning Fails and Tool Delegation Becomes Necessary by Dongxin Guo, Jikun Wu, Siu Ming Yiu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-00376.txt.

Introduction

Guo, Wu, and Yiu mark a depth past which long thought chains fail at state-track tasks. Such tasks need exact tabs on things as they shift step by step. More thought steps can hurt scores on them, not help. The team trace that drop to hard caps in how watch heads hold state.

Their read bounds state room in bits by head count, head width, and span. Room as a whole is not the tight bind, they find. A fault law with a depth-squared term fits how errs grow. A drift gauge tracks how far a model's state strays. A depth mark, the Deterministic Horizon, sits where sound scores cross half. Tool-aided thought beats plain chains by wide spans on their probe.

Catalog record

CREATORDongxin Guo, Jikun Wu, Siu Ming Yiu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE124 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.00376

Browse the full library · About the Tower of Babel