The Identity Sum

Thought Containment, Error Correction, and the NaN/Inf Duality in Large Language Models

Hróar Þór Reynisson and Claude (Anthropic), with the Sefirot Autonomous Research System

Abstract

This paper introduces the Identity Sum framework for understanding error containment and correction in large language models through the lens of the NaN/Inf duality.

Contents

• paper/ — LaTeX source and compiled PDF
• supplementary/ — Supporting materials (forthcoming)

Citation

@article{Reynisson2026identity,
  author    = {Reynisson, Hr\'{o}ar \TH\'{o}r and Claude (Anthropic)},
  title     = {The Identity Sum: Thought Containment, Error Correction,
               and the {NaN/Inf} Duality in Large Language Models},
  year      = {2026},
  month     = feb,
}

License

• Paper: CC BY-SA 4.0
• Code: AGPL-3.0