ARCANA: A Reflective Multi-Agent Program Synthesis Framework for ARC-AGI-2 Reasoning
Quick Answer
ARCANA is a multi-agent framework designed for ARC AGI 2 tasks, optimizing reasoning efficiency through a structured approach involving perception, hypothesis generation, and reflective refinement.
Quick Take
ARCANA is a framework designed for 2 tasks, optimizing reasoning efficiency through a structured approach involving perception, hypothesis generation, and reflective refinement. It employs a shared differentiable blackboard and a learned meta controller to enhance solution quality under strict constraints.
Key Points
- ARCANA decomposes tasks into perception, hypothesis generation, symbolic execution, and reflective refinement.
- A perceptual grounding agent creates object-centric scene graphs from raw data.
- A latent program policy proposes diverse domain-specific language (DSL) programs.
- The framework improves reasoning efficiency and solution quality on complex tasks.
- Agents communicate via a shared differentiable blackboard managed by a learned meta controller.
Paper Resources
📖 Reader Mode
~2 min readAbstract: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.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2607.09059 [cs.AI] |
| (or arXiv:2607.09059v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.09059 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Kejian Tong [view email]
[v1]
Fri, 10 Jul 2026 03:03:42 UTC (6,052 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →Adversarial Social Epistemology for Assemblies of Humans and Large Language Models
The paper introduces Adversarial Social Epistemology (ASE) to analyze how agents manipulate trust in public communications, highlighting mechanisms that undermine the reliability of testimony and inference. It critiques existing frameworks like epistemic bubbles and misinformation diffusion, proposing a new language for understanding trust breaches and auditing inferential chains in densely interactive environments involving humans and large language models.