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

Short summary

ARCANA is a collaborative multi-agent framework for solving ARC-AGI-2 tasks under strict time and hardware constraints. It decomposes tasks into perception, hypothesis generation, symbolic execution, and reflective refinement, with agents communicating through a shared differentiable blackboard and scheduled by a learned meta-controller. The design combines structured program search with adaptive multi-turn correction to improve reasoning efficiency on challenging abstract transformation tasks.

• •Multi-agent framework for ARC-AGI-2 with perception, hypothesis, execution, and reflection agents
• •Shared differentiable blackboard with learned meta-controller scheduling
• •Combines structured program search with adaptive multi-turn correction

Generated with AI, which can make mistakes.