Solidus pairs a verified Yul compiler backend with a Solidity specification challenge

A Solidity developer needs a compiler to preserve a contract’s behavior as it turns source code into instructions for the Ethereum Virtual Machine (EVM). In a report published July 24, 2026, at 23:00 UTC+8, Paradigm introduced Solidus, a pre-alpha project to build a formally verified Solidity compiler in Lean, a programming language and proof assistant. It is not audited or production-ready.

What has been proved—and what remains

Solidus has a built and formally verified backend that compiles Yul, the intermediate representation used by the Solidity compiler, to EVM code. Paradigm says its proof shows that the compiled program produces the same outputs as the input Yul program for all inputs, apart from EVM-specific details such as gas and memory layout. The model covers the state of the individual contract being compiled; its treatment of external calls assumes a deterministic external world.

The Solidity-to-Yul frontend has not been built. Paradigm has proposed a formal semantics for Solidity 0.8.35 as a starting point, but expects it may contain gaps. That distinction matters: verification of the Yul backend does not establish that a future complete Solidity compiler is correct.

Two challenges for contributors

Paradigm launched two optimization puzzles alongside the project:

  • Spec Hunt asks contributors to find differences between the proposed Solidity semantics and Solidity’s behavior, with a proof of concept. Paradigm says submissions are tested using Foundry, then reviewed by an AI agent that either merges a fix or escalates the case for human review.
  • Compiler Optimization asks contributors to improve the Yul-to-EVM backend while preserving its correctness theorem. Paradigm reports that, on its test suite, the current backend produces nearly three times as much bytecode as solc. It also cautions that the current compiler guarantees soundness only; completeness and performance are assessed heuristically with a holdout test suite.

Paradigm says AI tools did the code-writing work, while people made decisions about the specification and proof architecture. It reports more than 1,700 hours of total Codex /goal time and estimates a cost of about $150,000 at API rates. Paradigm also warns that challenge submissions may exploit gaps in the verification to game scores or introduce undetected problems. For builders, Solidus is therefore a research project to follow—not a compiler to substitute into a deployment workflow.

Original report: Paradigm, “Formally Verifying a Compiler Using Automated Research”