ARCANA: A Reflective Multi-Agent Program Synthesis Framework for ARC-AGI-2 Reasoning

2026-07-09

AI Summary

This work addresses abstract reasoning tasks under the stringent time and hardware constraints of ARC-AGI-2 by proposing a reflective multi-agent architecture that enables efficient program synthesis through an iterative loop of perception, hypothesis generation, symbolic execution, and reflection-based refinement. The approach integrates structured program search with adaptive multi-round correction, leveraging object-centric scene graphs, domain-specific language (DSL) strategies, symbolic execution for verification, and a failure-driven feedback mechanism. A learnable meta-controller dynamically orchestrates the interaction among these components to optimize performance. Experimental results demonstrate that the proposed framework significantly outperforms existing methods in both reasoning efficiency and solution quality.

Abstract

We present ARCANA, a collaborative multi agent framework for solving ARC AGI 2 tasks under strict test time and hardware constraints. ARCANA decomposes each task into iterative perception, hypothesis generation, symbolic execution, and reflective refinement. A perceptual grounding agent builds object centric scene graphs from raw grids, a latent program policy proposes diverse DSL programs, a symbolic executor verifies candidates on demonstrations, and a reflective agent synthesizes failure driven feedback for the next turn. These agents communicate through a shared differentiable blackboard and are scheduled by a learned meta controller. The design combines structured program search with adaptive multi turn correction, improving reasoning efficiency and solution quality on challenging abstract transformation tasks.

Problem

• ARC-AGI-2
• program synthesis
• abstract reasoning
• multi-agent framework
• symbolic execution

Innovation

• multi-agent framework
• program synthesis
• reflective reasoning
• symbolic execution
• differentiable blackboard