One thing.
One name.
Computed by anyone.

Let four people add the band Radiohead to a shared knowledge base and you will get four entries: Radiohead, RADIOHEAD, one with a trailing space nobody can see, one from a database export. Every fact anyone contributes attaches to one of the four at random. Ratings split, followers split, and the knowledge base ends up knowing less than any single contributor did.

This is the default outcome of letting independent parties describe the same world — and it compounds with every new source. Central ID registries fix it by making everyone ask permission, which a decentralized protocol cannot do. Name matching merges the four Nirvanas into one. Fuzzy matching is a guess, and guesses aren't reproducible.

The Intuition Identifier takes a different route: derive the name from the thing. Take the facts you have, run them through a published, exactly-specified procedure, and out comes a string. Anyone with the same facts gets the same string. Where the facts are too thin to tell two things apart, the identifier is honest about it instead of pretending precision — because hashing does not create identity; it preserves exactly as much uniqueness as its input had.

Not all identifiers are equally trustworthy, and a specification that treated them as equal would be lying by omission. Every scheme belongs to exactly one class, and the class is visible in the identifier itself.

Each entity type declares an ordered list of identifier options, strongest first — its identity ladder. You climb as high as your facts allow. The ladder for a film:

Rung 2 sits above rung 3 deliberately: when two identifiers are equally correct, the more open one wins. An identifier only one company can verify is a weaker foundation for a public knowledge graph than one anyone can check.

This is the gen1 algorithm running in your browser: NORM-1 normalization, canonical preimage, keccak-256, first 16 bytes. Change a fact and watch the identifier change. Give it the same facts in sloppier formatting — extra spaces, curly quotes, capitals — and watch it not change.

Paste an IID — or a raw value like a hyphenated ISBN — and this will parse it, check the scheme against the registry, verify canonical form (check digits included), and tell you exactly how strong the identifier is. Validation is the spec's three steps: shape, registered scheme, canonical value.

The scheme registry is closed on purpose: an unregistered scheme means an unregistered canonicalization, and an uncanonicalized value is exactly the ambiguous string this spec exists to eliminate. Adding a scheme is a reviewed, governed change — and once ratified, its rules are frozen forever.

Every scheme has a normative document with its grammar, canonicalization algorithm, and executable test vectors in the scheme registry; every entity type's ladder lives in the entity-type registry. A machine-readable export is published as registry.json.

One real entity legitimately carries many identifiers: a band has an ISNI and a MusicBrainz ID. No deterministic function maps one to the other — that is a judgement, and judgements are claims. In Intuition, "these two are the same thing" is itself content: attestable, disputable, stakeable, and reversible.

So the layers split along the line of what is mechanical and what is not. Identifiers are deterministic and free, and collapse the overwhelming bulk of duplicates automatically. The equivalence layer handles the judgement calls in public, where they can be argued with. An IID is a claim of identity, not a certificate — correctness is established the way every other truth is established in Intuition: by people putting something behind it.

This spec governs which string names a thing — and stops there. What your backend makes of an atom is a different job: which vocabulary it maps into, which type semantics apply, how it's contextualized, rendered, and ranked. Mint an atom from an ISBN and one system may interpret it through schema.org, another through a library-science ontology, a third through a schema it invented — all three converge on the same atom and interoperate at the identity layer, because identity and interpretation are separate layers.

That modularity is a core value of the approach: Intuition is deliberately unopinionated about interpretation while providing the deterministic building blocks for decentralized coordination. The identifier layer stays strict, minimal, and frozen precisely so that interpretation engines above it can stay opinionated, swappable, and yours to choose.

The identifier space is only as useful as the world it covers, and no one team covers the world. The specification is openly licensed (MIT), governed in public, and designed so that contribution is safe: everything additive is open, and nothing you build on can be changed out from under you.