SciGroveBeta
Climate

CMIP-Forge: An Agentic System that Retrieves, Computes, and Self-Reviews Climate Science

Dmitrii Pantiukhin, Boris Shapkin, Ivan Kuznetsov, Thomas Jung, Nikolay Koldunov

Featured June 18, 2026

Original Paper

This analysis was generated by SciGrove. Upload your own PDFs or enter a DOI — and get the same AI breakdown on any paper.

Get started

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

A smart computer system called CMIP-Forge reads thousands of climate papers and uses that knowledge to automatically analyze live climate data, with AI reviewers checking its work to catch mistakes.

In depth
The paper introduces CMIP-Forge, an agentic system designed to automate complex climate research workflows by integrating scientific literature with live climate data analysis. It leverages a hybrid retrieval-augmented generation (RAG) system over a vast corpus of CMIP6 papers, enabling an autonomous worker agent to access domain knowledge and execute cloud-native Python workflows. A crucial innovation is its multi-layered Defense-in-Depth architecture, which includes static code analysis and an autonomous adversarial peer-review protocol to enforce physical and methodological invariants, thereby mitigating common LLM-generated scientific errors.

Key Takeaways

  • 1
    CMIP-Forge integrates literature-grounded domain knowledge from 6,581 CMIP6 papers with live climate data analysis, addressing a critical bottleneck in climate science.
  • 2
    The system employs a multi-layered Defense-in-Depth architecture, featuring Abstract Syntax Tree (AST) static analysis and audited scientific primitives, to programmatically enforce geophysical guardrails and prevent subtle physical errors in generated code.
  • 3
    An autonomous adversarial peer-review protocol uses independent LLM agents to audit the worker's analytical pipelines, providing a self-correcting feedback loop for scientific rigor, while also exposing new failure modes like sycophantic regression.

Conceptual Flow

HIGH LEVEL
1
Automating Climate Research with AI Agents

The system reads climate papers, finds data, runs analysis, and has other AI agents check its work to make sure it's correct.

User Question
Climate Papers
Live Climate Data
AI Agent Plans & Acts
Analysis Code
Climate Figures
Reviewer Feedback
2
Reliable Autonomous Climate Insights

The system successfully performs complex climate studies, finding important patterns and even identifying its own AI review process's weaknesses.

Complex Climate Questions
AI-Generated Analysis
System Finds Answers & Self-Corrects
Validated Climate Findings
Identified AI Limitations