Detecting counterfeiting after Ironwood
The Orchard undetectable-forgery bug was not the first, but Ironwood's formal verification is explained as making it the last. Wallet migrations help verify the supply.
Project Tachyon posted a piece on how to detect counterfeiting after Ironwood. Last month Taylor Hornby of Shielded Labs found a forgery vulnerability in Orchard, which was patched in a network upgrade and is believed not to have been exploited. But because exploitation would be undetectable, the community decided to create Ironwood, a fresh pool launched with the vulnerability patched. The team took on formally verifying Ironwood to rule out undetectable forgery bugs.
The key point of the piece is the observation that undetectable forgery bugs can exist only in the specification. Implementation bugs produce only detectable forgery. Because a block commits to the full contents of every transaction, including proofs, replaying the record with fixed software leaves wrongly accepted transactions as permanent evidence. Errors in a specification, on the other hand, can be ruled out mathematically, and what remains are only the unavoidable cryptographic assumptions. Zcash has been reducing its assumptions with a new pool that uses no trusted setup and no pairing-based cryptography.
It also reviewed past cases. The 2016 InternalH collision bug in the Zerocash specification and the 2018 Sprout [BCTV14] soundness flaw were specification errors and their exploitation was undetectable. This Orchard vulnerability is a soundness bug in the zk-SNARK circuit. The 2016 libsnark subgroup check omission and the 2025 halo2_proofs query collision bug would have been detectable even if exploited. The author explained that proving correctness by algebraically expressing the verifier's behavior rather than the circuit code catches bugs in both the circuit and the SNARK.
The team said it machine-checks proofs with tools such as Lean and used human-led LLMs to cut proof-writing time to a few weeks. Working with Tal Derei, Gregor Mitscha-Baude of zkSecurity and Daira-Emma Hopwood, it made significant progress over the last few weeks. The piece concludes that this is not a list of all forgery bugs but a demonstration that specification verification alone can rule out undetectable forgery. Knowledge soundness and the comparison with the real software will be covered in the next post.
This summary is based on the original; check the original for exact details.