Awesome Auto Research Hub Papers · Datasets · Projects
← Back to papers

CausalForge: A Formally Grounded, Self-Improving Agentic Framework for Automated Research in Causal Inference

arXiv 2026 73.5 method

TLDR

CausalForge automates theoretical research in causal inference using a Lean proof assistant and a self-improving agentic pipeline.

Reasoning

The paper presents a novel framework that combines a formal library (Causalean) with an agentic pipeline (CausalSmith) for automated theorem proving and research. Strengths include formal verification via Lean and a self-improving loop, while weaknesses are the narrow focus on causal inference and potential subjectivity in the statement audit step.

Read-first score

Read-first score 73.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 69.

Recency 8%
100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Reproducibility 25%
85

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=artifact,code,github

Methodology quality 25%
80

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=evaluation,metric,result

Topical relevance 42%
57.5

Uses existing LLM keyword relevance scores normalized to 0-100. AI scientist,automated scientific discovery,autonomous research agent,automated research,literature review agent,survey generation,automated experimentation,experiment design agent,AI for scientific research,paper writing agent,research automation,scientific discovery agent

Field roles

FrontierMethodology anchorReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 37.

Keyword Scores

automated research
9
research automation
9
AI scientist
8
automated scientific discovery
8
autonomous research agent
8
AI for scientific research
8
scientific discovery agent
8
paper writing agent
5
literature review agent
2
automated experimentation
2
survey generation
1
experiment design agent
1

Deep Analysis

Innovations

  • CausalForge: a formally grounded framework for automated theoretical research in causal inference using the Lean proof assistant
  • Causalean: a foundational Lean library with 7,035 machine-checked declarations for causal inference, developed with LLM assistance under human design and review
  • CausalSmith: a self-improving agentic pipeline that autonomously selects research topics, proposes results, formalizes statements, constructs proofs, and outputs artifacts for human inspection
  • Statement audit: a mechanism that compares formal theorems against the informal claims they intend to express, addressing the gap between formal proof and scientific meaning

Methodology

CausalForge combines the Causalean library and the CausalSmith pipeline. CausalSmith selects topics, generates results, formalizes them in Lean, constructs machine-checked proofs, and presents artifacts; a statement audit then checks alignment between formal theorems and informal claims. The system is evaluated on artifacts from completed autonomous research runs.

Key Results

The system produces artifacts from autonomous research runs, with formal verification ensuring proof correctness and statement audits mitigating the gap between formalization and intended scientific claims.

Limitations

  • Machine-checked proofs only guarantee correctness under formal assumptions, not that the formalization faithfully captures the intended scientific claim, and the statement audit may not fully eliminate this gap.
  • The pipeline relies on human inspection of the final artifacts, limiting full autonomy.
  • Evaluation is limited to artifacts from autonomous runs, with no quantitative metrics reported in the abstract.

Tags