← All insights
RESEARCHProject Tachyon

Ironwood formal verification complete

Ironwood's balance integrity is backed by a Lean machine-checked proof of more than 2,700 theorems. It is the result of three teams working on it for over a month.

Project Tachyon announced that it has completed formal verification of Ironwood, the new shielded pool activating today as part of NU6.3. Its previous post showed that formal verification can eliminate undetectable forgery bugs, and this time it published a machine-checked proof written in Lean. The proof is in the ironwood repository, consists of more than 2,700 theorems, and was completed by three research teams working for over a month.

The goal of this work is a single security property, balance integrity: a shielded pool never pays out more value than was publicly paid in. Most of the effort went into proving the knowledge soundness of the Ironwood zk-SNARK. It analyzed the Fiat-Shamir transform in the algebraic prover model and the random oracle model, and part of the proof handles the challenges of the 11-round inner product argument (IPA). The result reduces to concrete security bounds on the discrete logarithm problem, and the author gives a conservative security target of about 123 bits of work.

To make the proof fit the real software, it sets a boundary called the verifier fingerprint. It defines the multi-scalar multiplication (MSM) returned by the Rust verifier as a canonical combination derived from the proof string, and confirmed in Lean that it matches the values recorded from a halo2 run. The logic is that even if there is an implementation error, every block contains all its proofs, so past proofs can be re-verified to detect forgery after the fact. For the circuit, it translated 395 regions into Lean and proved the Action statement holds, and also proved agreement with the verification key.

The last layer is ledger-level theorems. It proved, for every valid ledger, that every spend is a previously created output, that value is conserved, that received notes remain spendable (blocking the Faerie Gold attack), and that spends require authorization, with any violation reducing to the breaking of assumptions such as discrete log or the random oracle. Value range proofs prevent modular wraparound, guaranteeing balance at the integer level. This proof covers only forgery prevention (soundness); privacy properties are out of scope.

As contributors it mentions Gregor Mitscha-Baude of zkSecurity and Daira-Emma Hopwood, a cryptographer at ZODL, Tal Derei, who led the formalization, and Guillaume Claret and Valerii Huhnin, who worked in parallel on an independent proof. The proof uses definitions inspired by ArkLib, VCV-io and Mathlib, and used Clean to analyze the circuit. The remaining assumptions (such as BLAKE2b as a random oracle) are named explicitly in the documentation.

This summary is based on the original; check the original for exact details.