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

Source: [arXiv](https://arxiv.org/abs/2607.22511v1)  
Feed7 permalink: https://feed7.dev/p/2607-22511v1-0mgsdh3  
Published: 2026-07-24T17:32:35.000Z  
Trust: Needs Review (needs_review)

## Why Included

CausalForge pairs a Lean-verified causal-inference library with an autonomous research pipeline and a semantic statement audit. Formal proof checks derivation, not whether the theorem matches the intended claim.

## Source Summary

CausalForge combines **7,035 machine-checked declarations** in the Causalean Lean library with CausalSmith, an agent pipeline that selects topics, proposes results, formalizes statements, builds proofs, and presents artifacts for review.

## Practical Implication

Builders of research agents should separate mechanical validity from semantic validity. The framework uses Lean for proof checking and a **statement audit** to compare each formal theorem with the informal scientific claim it is meant to encode.

## Agent-Ready Context

CausalForge combines **7,035 machine-checked declarations** in the Causalean Lean library with CausalSmith, an agent pipeline that selects topics, proposes results, formalizes statements, builds proofs, and presents artifacts for review.

Builders of research agents should separate mechanical validity from semantic validity. The framework uses Lean for proof checking and a **statement audit** to compare each formal theorem with the informal scientific claim it is meant to encode.

A kernel-checked proof establishes only that a statement follows from its assumptions. It cannot establish that the formalization captures the right scientific question, so human inspection remains part of the presented workflow.

## Context Map

- Layer: agent
- Domains: research, data
- Topics: harness-engineering, agent-reliability, reasoning

## Uncertainty

- A kernel-checked proof establishes only that a statement follows from its assumptions. It cannot establish that the formalization captures the right scientific question, so human inspection remains part of the presented workflow.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.