Zero-Knowledge Proofs, and the Much Narrower Thing We Actually Have
A zero-knowledge proof convinces somebody a statement is true while revealing nothing beyond its truth. The phrase covers an enormous range, from proving you know a password to proving a whole program ran correctly, and the range is where the misunderstanding lives.
What we have
A signature scheme that lets a holder prove one specific statement: that they hold an issuer's signature over a set of messages, while revealing only a chosen subset of those messages and never revealing the signature itself.
That is a real zero-knowledge proof, of a real and useful statement, and it is what makes two presentations of the same credential hard to link to each other.
The ciphersuite is pinned to a named curve and hash from an internet-draft specification, and the implementation is stated to be byte-compatible with the chain's own: the same secret key produces the same public key, and signatures cross-verify in both directions.
The overclaim this page exists to refuse
The demonstration everybody reaches for is proving you are over eighteen without revealing your birth date.
That is not what happens here. In our stack, an over-eighteen credential is a credential type: the issuer decided the question, wrote the answer as a claim, and signed it. The holder later reveals that claim and withholds others.
The predicate vocabulary appears in our code only as a credential type in tests, and never as a computation. Nothing derives the answer from a birth date at presentation time.
Both designs hide the birth date. Only one of them computes anything, and describing ours as the computing kind would be claiming a capability we do not have.
Why the distinction matters rather than being pedantic
Who decides. With a signed claim, the issuer decided and you cannot change the question. With a computed proof, the verifier can ask a question the issuer never anticipated.
What breaks. A signed claim needs re-issuance when the question changes. A computed proof needs a circuit somebody wrote, reviewed and can be held to.
What it costs. Signed claims are cheap and shipping today. General proving systems are expensive in every dimension, which is why almost nobody in this category ships them either.
And it is unaudited
The cross-implementation compatibility is stated in the package and verified between our own two implementations, which is a parity check rather than a security review.
Parity says two implementations agree. An audit says the thing they agree on is sound. We have the first and not the second.