Multi-Agent Coherence in AI Systems: Experimental Study of Five Roles, Language Architecture, and Self-Organization Mechanisms Based on the ODTOE Formalism

Anton Pankratov(independent)·
multi-agentAIcoherenceRound Tablefive rolesLLMphantom coherenceLambda problembilingual architectureprompt engineeringB-score

Abstract

Аннотация

RU

Результаты экспериментов по исследованию мультиагентной когерентности в системах ИИ на основе формализма ODTOE. Пятиролевая архитектура (Визионер, Аналитик, Строитель, Валидатор, Когерент), работающая через протокол Round Table. Четыре ключевых эксперимента: анализ фреймворка 25 агентами со сжатием промптов в 2,5 раза; A/B-эксперимент, показавший, что англоязычные промпты дают B-score на 48% выше для практических задач; анализ архитектуры точки входа; исследование реальной сессии развёртывания (157 tool calls), выявивш��е Lambda-проблему. Введена скорректированная когерентность S_adjusted = S_team × B̄, обнаруживающая фантомную когерентность. Двуязычная архитектура: английский для широты, русский для глубины.

Abstract

EN

Experimental results on multi-agent coherence in AI systems based on the ODTOE formalism. A five-role architecture (Visionary, Analyst, Builder, Validator, Coherencer) operating via the Round Table protocol. Four key experiments: 25-agent framework analysis with 2.5x prompt compression; A/B experiment showing English prompts yield 48% higher B-scores for practical tasks; entry point architecture analysis; real deployment session study (157 tool calls) uncovering the Lambda problem. Introduces adjusted coherence S_adjusted = S_team × B̄ detecting phantom coherence. Bilingual architecture: English for breadth, Russian for depth.

摘要

ZH

基于ODTOE形式主义的AI系统多智能体相干性实验研究。五角色架构(愿景者、分析师、构建者、验证者、相干者)通过圆桌协议运行。四个关键实验:25个智能体框架分析,提示压缩2.5倍;A/B实验表明英语提示在实践任务中B分数高48%;Lambda问题分析。引入调整后的相干性S_adjusted = S_team × B̄,检测幻象相干性。

Subjects & Identifiers

Subjects:
Physics and Society (physics.soc-ph) · multi-agent · AI · coherence · Round Table · five roles · LLM · phantom coherence · Lambda problem · bilingual architecture · prompt engineering · B-score
Category:
Social Applications · Социальные приложения
Authors:
Anton Pankratov (independent researcher)
Submitted:
Last modified:
Languages:
Russian (primary), English
Permanent URL:
https://odtoe.org/en/articles/multiagent-coherence
Journal:
Observer-Dependent Theory of Everything (ODTOE Corpus)
Comments:
For research collaboration or corrections, contact via /contact. Citations and academic engagement welcome.

Cite this article

Select the text below to copy citations in your preferred format.

Plain text

APA-like
Pankratov A. "Multi-Agent Coherence in AI Systems: Experimental Study of Five Roles, Language Architecture, and Self-Organization Mechanisms Based on the ODTOE Formalism." Observer-Dependent Theory of Everything, odtoe.org, 2026. https://odtoe.org/en/articles/multiagent-coherence
BibTeX[ click to expand ]
@article{pankratov2026multiagentCoherence,
  author    = {Pankratov, Anton},
  title     = {Multi-Agent Coherence in AI Systems: Experimental Study of Five Roles, Language Architecture, and Self-Organization Mechanisms Based on the ODTOE Formalism},
  journal   = {Observer-Dependent Theory of Everything},
  year      = {2026},
  month     = {Apr},
  url       = {https://odtoe.org/en/articles/multiagent-coherence},
  publisher = {odtoe.org}
}
RIS (EndNote / Reference Manager)[ click to expand ]
TY  - JOUR
AU  - Pankratov, Anton
TI  - Multi-Agent Coherence in AI Systems: Experimental Study of Five Roles, Language Architecture, and Self-Organization Mechanisms Based on the ODTOE Formalism
JO  - Observer-Dependent Theory of Everything
PY  - 2026
DA  - 2026-04-10
UR  - https://odtoe.org/en/articles/multiagent-coherence
PB  - odtoe.org
ER  - 
Multi-Agent Coherence in AI Systems: Experimental Study of Five Roles, Language Architecture, and Self-Organization Mechanisms Based on the ODTOE Formalism