FRACTAL: A Self-Similar Multi-Agent Orchestration Architecture with Formal Guarantees (arXiv 2610.22561)

Paper Overview

• Field: Machine Learning
• Authors: Ohad Dan, Abhinav Taliyan
• Published: 2026-10-19
• arXiv: 2610.22561

Abstract

We present FRACTAL, a self-similar multi-agent architecture for recursive goal decomposition, in which the same five-component organizational pattern -- Orchestrator, Ensemble, Knowledge Ledger, Decision Maker, Meta-Observer -- recurs at every level of a decomposition tree. We make four formal contributions: (1) a communication bound showing intra-node cost is linear in ensemble size and round count (O(kr)) under a two-phase blind-proposal/stress-test protocol, versus quadratic cost for free-form debate, with an explicit, stated assumption on fixed evaluator count; (2) a two-mode orchestration criterion separating generative and evaluative phases via a measurable diversity-saturation threshold; (3) a supermajority convergence guarantee with explicit probability bounds at stated agreement levels, and a formal depth bound under an operationalized goal-complexity measure; (4) a Bayesian inter-layer escalation mechanism unifying hysteresis and contract-breach signaling under one posterior update. We illustrate the architecture's execution qualitatively on a worked example and a multi-layer case study; we do not report quantitative performance claims in this paper, and we explicitly discuss the formal and inductive-bias limitations of the model, including an unimplemented but architecture-compatible direction toward bounded self-improvement. We use "fractal" as an architectural metaphor for self-similarity, not a measured mathematical property. Empirical validation of the architecture's comparative performance is left to future work.

Key Points

• Architecture: five components (Orchestrator, Ensemble, Knowledge Ledger, Decision Maker, Meta-Observer) repeat recursively across every level of a decomposition tree.
• Communication cost: linear O(kr) under blind-proposal/stress-test protocol, versus quadratic for free-form debate (assuming a fixed evaluator count).
• Orchestration criterion: switches between generative and evaluative modes at a measurable diversity-saturation threshold.
• Guarantees: supermajority convergence with explicit probability bounds; formal depth bound tied to an operationalized goal-complexity measure.
• Escalation: Bayesian inter-layer mechanism unifies hysteresis and contract-breach signals in a single posterior update.
• Limitations: no quantitative performance claims; "fractal" is a self-similarity metaphor, not a measured property; empirical validation and bounded self-improvement remain future work.