What was released
CausalForge is a research framework that treats causal-inference discovery as a formally verifiable process rather than natural-language generation alone. Its agents choose topics, propose results, encode statements in Lean and attempt proofs.
Core design
- The accompanying Causalean library contains 7,035 machine-checked declarations according to the paper.
- A CausalSmith module connects topic selection, proposal, formalization, proving and presentation.
- A statement audit checks whether the formal theorem actually matches the informal claim.
- The authors report releasing code, the library and execution records.
Why it matters
The work points toward scientific AI whose conditions and reasoning can be inspected by a proof checker. That is especially useful in causal inference, where a small change in assumptions can alter the conclusion.
Limitations
This is a July 2026 preprint and has not completed peer review. A proof establishes that a formal statement follows from its formal assumptions; it does not automatically establish that the question or assumptions accurately represent the real world.