Born at a permanent address

There is a debt you can take on in the first ten minutes of drafting an ontology and pay off for years. It looks like this:

@prefix rho: <http://example.org/record-harm-ontology#> .

Nobody means it permanently. It is a placeholder, to be replaced once the real address exists. But an ontology's IRIs are not a naming convenience; they are the identity of every term in it. The moment anyone imports your vocabulary, cites a term, or stores an annotation, the placeholder is load-bearing. Changing it later is not a rename — it is a breaking release that every downstream adopter has to be dragged through.

The Record Harm Ontology was drafted that way, and it is still paying. What is served from this site is a rewritten copy: the staging script substitutes the permanent base and then refuses to finish if a single example.org IRI survives the rewrite.

if grep -rInE "https?://[^ <>]*example\.org" "$DEST"; then
  echo "error: an example.org IRI is present in staged output" >&2
  exit 1
fi

That check is not paranoia. A hosted copy that quietly drifted back to the placeholder would publish two different names for the same term, which is exactly the failure the ontology itself calls Forgery of Provenance.

The Record Ontology was started later, and started at its final address:

@prefix rec: <https://www.epistemic-ontology.net/record#> .

The consequence is small and pleasant. Its hosted copy is byte-identical to the source file in its repository. There is no rewrite step, nothing to verify, and no v3.0 migration waiting in its future. Its staging script only copies, and then asserts that the permanent namespace is present.

Why 303 and not 302

Ask this site for a namespace IRI and you get a 303 See Other.

curl -sI -H "Accept: text/turtle" https://epistemic-ontology.net/record | head -2

The distinction matters more than it looks. https://…/record names an ontology. It does not name a document. There is no file that "is" the Record Ontology; there are serializations of it — Turtle today, possibly RDF/XML and JSON-LD tomorrow — and a documentation page, and none of them is the thing itself.

A 302 would say "the resource you asked for is temporarily over here", which asserts that the ontology is a document that has moved. A 303 says something weaker and true: the thing you asked about is not a document, but here is a document about it. That is the whole content of the httpRange-14 resolution, and it is one of the few places where a status code carries a genuine ontological commitment.

The same document then serves every term, because the namespaces here are hash namespaces. …/record#Inference never sends the fragment to the server at all; you fetch the one document and your client resolves the fragment locally. One round trip gets you the complete vocabulary. For vocabularies of this size that is plainly the right trade; slash namespaces earn their keep at thousands of terms, not dozens.

The redirect chain, and why the old name still works

The IRIs were minted at www.epistemic-ontology.net. The canonical host is now the apex, epistemic-ontology.net. Four names exist — the apex, www, and both .org variants — and three of them permanently redirect to the fourth, as does all plain HTTP.

So a minted term IRI resolves through one 301 to the apex, and then the 303 negotiation. Two hops, both permanent, both cacheable.

It would have been tidier to mint at the apex. It is not worth "fixing". A permanent identifier that you rewrite whenever a tidier form occurs to you is not a permanent identifier — it is a mutable one with good intentions. The minted form stays valid forever, because that is the only thing "permanent" can mean.

What follows for anyone citing this

Cite the IRI, never the file path:

url = {https://www.epistemic-ontology.net/record}

not …/record/record-ontology.ttl. The IRI is what is promised. File layout is an implementation detail, and this site has already moved files once: the worked examples now live under /record/examples/, each at its own dereferenceable namespace, because every one of them declares an IRI of the form

@prefix ex: <https://www.epistemic-ontology.net/record/examples/saccheri#> .

and an address that a file declares as its own name ought to be an address that answers. The old paths still redirect. They will keep redirecting.

The general rule

Pick the permanent name before the first term is written. If you cannot — if the domain is not registered, if the project might be renamed — then you do not yet know what you are publishing, and it is cheaper to wait than to migrate.

The alternative is what this site demonstrates in miniature: one vocabulary that is copied straight through, and one that is rewritten in flight, checked, and still carrying a breaking release on its horizon. Both work. Only one of them was free.