Summary

  • RFC 3187 carried the ISBN namespace into URNs, but an ISBN named an edition of a monographic work—not one file, one physical copy, one URL or one resolver response.
  • Resolution could legitimately return many instances, only a bibliographic holding for physical material, or multiple records after mistaken ISBN reuse. Syntax, assignment, catalogue evidence, access and provenance stayed separate.

Published in October 2001, Using International Standard Book Numbers as Uniform Resource Names was Informational. It registered ISBN for use under the URN architecture of RFC 2141 and RFC 2611. In 2017 the IESG moved RFC 3187 to Historic, and RFC 8254 explained that a newer ISBN namespace template under RFC 8141 had made the old registration obsolete. The old document remains evidence of the original design, not the current operational template.

The design began with an inherited object model. An ISBN identified an edition of a monographic work. That scope was narrower than the intellectual work across all editions and broader than one manufactured copy. It also differed from one downloadable file. Embedding the number in urn:isbn: did not change that granularity.

This matters because durable syntax can make a category look more exact than its assignment practice. Publishers distinguished new editions from revised printings, but RFC 3187 admitted that the boundary could be subjective and practice varied. The URN framework reflected those decisions; it did not create a central machine rule that could reconstruct what an edition should have been.

Electronic publishing added another edge. The document said each digital format should receive a separate ISBN. A print edition, PDF and another electronic format could therefore be closely related in content while remaining distinct assignments. Similar bytes or the same title did not authorize software to merge their identities.

The syntax itself was simple. The example URN:ISBN:0-395-36341-1 needed no percent encoding under the registration. For lexical comparison, hyphens were removed and any x was uppercased. That rule answered whether two written forms denoted the same ISBN string. It did not validate the publisher, edition statement or resolver output.

Even the ISBN-10 check digit had a bounded role. Modulus 11 could detect many malformed strings and used X for ten. A valid check digit proved only an arithmetic relationship among characters. It did not prove that the number had been assigned, assigned once, assigned to the displayed title, or copied from an authoritative record.

Resolution made the layers visible. Some ISBN resources existed only in print or another physical medium. For them, the best response might be bibliographic data from a national bibliography plus the location of a holding. A successful resolver had answered “what record and where is a copy?”, not delivered the book.

For electronic material, RFC 3187 expected an ISBN URN often to resolve to many instances and many URLs. That multiplicity followed the edition-level name. Several libraries, publishers or repositories could expose legitimate instances. Returning more than one location was not automatically duplication or corruption.

The proposed authority graph was distributed. National bibliographies were primary candidates because ISBN group agencies often sat in national libraries and their databases covered most ISBN-bearing publications. Group and publisher components could help select a resolver. Publisher-operated services and union catalogues could supplement the path.

Nothing in the ISBN characters froze those responses. Resolver discovery, database selection, holding lookup and content access were separate acts run by different institutions. A correct ISBN could meet a stale catalogue, an unavailable URL or an access restriction without losing its identity.

RFC 3187 confronted human error directly. An ISBN, once assigned, was not supposed to be reused. Publishers nevertheless sometimes reused one for another book. The expected consequence was not a forced rewrite of history. A national bibliography might return two or more records, allowing a user to choose using title, author and other metadata.

Catalogues could also provide cross-references from an incorrect ISBN to the correct one. That correction was an additive claim: this recorded number was erroneous, and this other number should be used. It did not prove that every catalogue, citation, book copy or resolver had converged, and it did not make the original assignment event vanish.

Persistence was similarly bounded. RFC 3187 described a correctly assigned ISBN as staying with a publication from production onward. Persistence did not mean perpetual availability, permanent metadata correctness or a resolver that never failed. It meant that changing location should not require renaming the edition.

Assignment authority operated at international, group-agency and publisher levels. Publishers commonly received blocks and selected numbers for new items. A publisher component could help route a query, but it was not a cryptographic signature. The number did not authenticate the publisher or the database responding today.

The security section excluded precisely the assurances a product might be tempted to imply. RFC 3187 did not specify secure or authenticated resolution. It did not validate integrity, source or provenance of ISBN URNs. Intellectual-property rights in the identified object and rights in resolver databases remained outside scope.

An ISBN URN therefore could not grant reading rights, license a copy, prove a file genuine or establish that a resolver was authorized. Identity, discovery, access, provenance and permission were adjacent but independent systems.

The 2017 transition reinforces that lesson. RFC 8254 preserved ISBN-10, added the ISBN-13 world and let the namespace evolve with ISO rules through the registered template. The persistent namespace survived by changing its governance document. Treating RFC 3187 itself as timeless authority would confuse the durable name with one historical registration.

The best evidence model keeps the whole chain: assignment event, agency or publisher, edition and format claim, lexical form, resolver endpoint and timestamp, returned records, holdings, URLs, retrieved assets, authentication context and correction links. Only then can a stable number support a changing bibliographic world without pretending that every layer is the same object.