Summary

  • The W3C Dataset Exchange Working Group published The Profiles Vocabulary as a First Public Working Draft on 20 August 2026. It is on the Recommendation track, but it is not a W3C-endorsed standard and may change.
  • The HTML draft declares a property chain under which a resource conforming to a profile can be inferred to conform to the specification that the profile profiles. It separately leaves the meaning and testing of conformance to individual communities.
  • A June correction fixed the order of the two properties in the chain. A separate open issue still marks inclusion of the axiom at risk, and its embedded issue text displays the old reversed order.
  • The dated RDF Turtle artifact published with the draft contained no owl:propertyChainAxiom declaration at retrieval. The implementation report names examples but its per-element evidence table still contains placeholders.
  • An open Data Shapes issue offers a completeness counterexample. Before an inherited relation enters procurement, certification or automated policy, it should carry a conformance-inheritance receipt identifying the exact artifacts, authority, tests, inference engine and result type.

A small inference with a large authority surface

On 20 August, the W3C Dataset Exchange Working Group published the First Public Working Draft of The Profiles Vocabulary, usually shortened to PROF. Its purpose is practical. Different communities routinely take a broad specification and make a profile for a jurisdiction, sector, application or implementation. The profile may narrow permitted choices, extend a model, combine several specifications or explain how a specification should be used.

PROF gives those arrangements machine-readable structure. A prof:Profile can point to a standard or another profile through prof:isProfileOf. It can also point to resources that play distinct roles: constraints, validation, schemas, vocabularies, guidance, mappings, examples or specifications. Communities may add and publish further roles. A client can therefore discover not just that a profile exists, but which artifacts are meant to make it usable.

The draft’s most consequential sentence concerns inheritance. If a data resource declares conformance to Profile X, and Profile X is a profile of Specification Y, the HTML text says the resource must conform to Y as well. The draft represents the inference as:

dct:conformsTo owl:propertyChainAxiom ( dct:conformsTo prof:isProfileOf )

In graph terms, the machine starts at the resource, follows dct:conformsTo to the profile, then follows prof:isProfileOf to the parent specification. A reasoner can add the resulting dct:conformsTo relationship. That is attractive precisely because the derived assertion can be reused by catalogs, validation tools and automated agents without every client rebuilding the path.

The publication status limits the authority of the text. This is a First Public Working Draft on the Recommendation track, not a Recommendation. W3C says a Working Draft is a work in progress published for review and historical reference. The group has agreed to publish it, but W3C and its Members have not endorsed it as a standard. Implementations can provide useful feedback; they should not turn the maturity label into an approval claim.

Inference is not the definition of conformance

The draft puts an important qualification beside the property chain. Individual communities may decide what conformance to their profiles and to the things profiled means, and how it should be tested. The vocabulary can carry a relation. It cannot settle every community’s substantive rule for earning that relation.

This distinction is easy to lose because RDF inference is exact once its inputs and axioms are fixed. A reasoner either derives a triple or it does not. The apparent precision belongs to the graph operation, not necessarily to the underlying claim. A profile might express every rule of its parent, only a chosen subset, an extension layered on top, or guidance that requires human interpretation. Its validator may be complete for one purpose and incomplete for another.

The draft’s treatment of null profiles makes the problem concrete. A null profile is intended to help test conformance to a specification that lacks its own validation mechanism. Yet the draft says no universal precise construction method is available, because the source specification’s rules may themselves be imprecise. Translating natural-language or OWL requirements into SHACL constraints is an act of interpretation and coverage design, not merely serialization.

PROF also distinguishes direct and transitive profile relationships. prof:isProfileOf records an asserted direct link. prof:isTransitiveProfileOf is designed to expose more distant ancestors so clients do not have to walk every hierarchy. If the convenience property is used, the draft says the relevant relationships should be complete. But it does not prescribe how each implementation must realize the associated inferences.

That leaves three different questions where one word—conforms—can appear. Is the relationship directly asserted or inferred? Did a validator run, and what did it cover? Does a particular purchaser, regulator or certifier accept that evidence for its decision? A vocabulary can help connect those layers. It should not collapse them.

The axiom was corrected and remains at risk

The public issue record shows why the inference surface needs its own provenance. Issue 58, opened in March 2026, identified that an earlier expression placed the two properties in the wrong order. The correction changed the chain so the path begins with the resource’s dct:conformsTo edge and then follows the profile’s prof:isProfileOf edge. A commit applied that change across the introduction, example, definition table and feature-at-risk appendix. The issue closed on 26 June.

That correction did not close Issue 29. The older issue remains open with an at-risk label and says inclusion of the property-chain axiom is unresolved. Two May comments connect it to the completeness of null-profile validation and to a counterexample in the Data Shapes repository.

The 20 August draft therefore carries two visible states at once. Its principal prose, worked example, definition table and feature-at-risk appendix show the corrected property order. The inline issue text imported from Issue 29 still quotes the old reversed order. A careful reader can reconstruct the intended version. A client ingesting one fragment, a copied issue summary or an earlier cached artifact may not.

The broader appendix is equally candid. It lists the vocabulary’s classes, properties, roles and conformsTo axiom as features at risk pending implementation evidence. A note says the list began with every feature because formal evidence had not yet been gathered into the specification, and that the list should shrink as evidence is documented. “At risk” does not mean rejected. It means publication is asking implementers to turn a design assertion into a tested record.

The HTML and Turtle surfaces do not yet say the same thing

W3C’s announcement links a dated prof.ttl file and describes it as the OWL vocabulary encoded in RDF Turtle. That file is especially important for machines: it is the artifact an implementation can download, hash and load.

At retrieval, the 20 August Turtle file defined prof:isProfileOf, made it a subproperty of prof:isTransitiveProfileOf, and described the intended inheritance of constraints. It did not contain an owl:propertyChainAxiom declaration. The HTML draft, by contrast, presented the axiom in its introduction, example and property table.

The omission may be deliberate while the axiom is at risk. The draft’s RDF-representation section also notes that additional axioms may live in separate artifacts. But the public release surfaces inspected for this Article did not make the disposition explicit. A human reading the HTML can believe the axiom is part of the vocabulary; a reasoner loading the dated Turtle file will not receive it from that file.

That difference is not a typographical curiosity. It changes execution. One implementation may add the axiom manually from the HTML. Another may treat the Turtle artifact as authoritative and derive nothing. A third may use a living namespace artifact from a different date. All three can describe themselves as implementing the same draft while producing different graphs.

The minimum repair is not necessarily to insert the axiom. The group may decide to remove it, revise its semantics or publish it as an optional module. The governance requirement is simpler: the human specification, downloadable ontology, issue state and test expectations should reveal the same decision, or clearly declare why they do not.

The implementation ledger has examples but not feature proof

The FPWD makes its conceptual model, vocabulary specification and Test Suite appendix normative for compliance. The appendix says a software suite uses SHACL validation templates and instructions to test implementations. The separate implementation report goes further: it describes a method, names five examples and groups them into three claimed independent implementation groups.

One example is Cheka, a Python validation tool described as traversing a profile hierarchy to find validators belonging to the profile and its ancestors. That is close to the operational mechanism at issue. Other examples include a LOCI ontology, ODRL profiling guidance, an OGC hierarchy and a persistent profile catalog.

Yet the report’s decisive table—implementation claims per vocabulary element—still contains xx placeholders for the element, label, implementer and online location. The displayed repository link for the test suite returned HTTP 404 at retrieval, although a testsuite directory exists on the repository’s gh-pages branch. These observations do not show that no tests or implementations exist. They show that the public path from a named implementation to verified support for each at-risk feature is not finished on the cited evidence surface.

The active charter supplies the target. Rec-track republication of PROF is a tentative deliverable. For normative specifications to advance beyond Candidate Recommendation, the charter expects at least two independent interoperable implementations of every feature, verified through open tests. It also asks for testing plans from the earliest drafts and coordination with the Data Shapes Working Group. The group is chartered until April 2028, so the unfinished record is a commissioning signal, not a deadline failure.

A counterexample asks what a pass actually covers

Data Shapes Issue 882 sharpens the semantic question. Its author constructs a profile that implements a valid subset of checks for an OWL specification. A data graph passes those SHACL checks. The parent OWL model, however, contains another constraint that the incomplete shapes graph does not test; after entailment, the data contains membership in two disjoint classes.

The profile-level validator has not lied about the checks it ran. It is sound for the subset. It is incomplete as evidence of the parent specification. If the property chain automatically turns “conforms to this profile” into “conforms to that parent,” the inferred conclusion can outrun the validator’s coverage.

This is an open issue submitted to a public W3C repository, not a Working Group resolution. It does not prove that profile inheritance is generally unsound. It identifies the condition that the final design must make testable: what completeness threshold, if any, allows a profile-level result to support a parent-level conformance relation?

Possible answers include requiring complete validators for inheritance, separating tested conformance from descriptive profiling, narrowing the axiom, attaching entailment regimes and test coverage, or leaving inference to community-specific modules. The Article does not select the semantic remedy. It argues that downstream decision systems cannot safely guess which remedy was intended.

A conformance-inheritance receipt keeps the layers apart

Every consequential inferred relation should travel with a compact receipt. It should name the exact profile and parent versions, their immutable hashes and the direct prof:isProfileOf edges used. It should list the machine-readable vocabulary artifact and the axioms actually loaded, not merely the human-readable specification title.

The receipt should then identify who defined conformance and for what purpose. Role-tagged constraints, validators, schemas, guidance and mappings should be versioned. The record should name the reasoner, validation engine, entailment regime and configuration; the test suite, feature coverage and known blind spots; independent implementation evidence; open at-risk issues and counterexamples; and the dates or changes that trigger revalidation.

Most importantly, it should label the output. An asserted relation, an inferred RDF triple, a validator pass, a certification and a legal determination are not interchangeable. A procurement system may decide that one is sufficient for a low-risk inventory field and another is required for a contractual warranty. The machine should not choose that authority level by dropping provenance.

This follows Heng Lu’s agency lens. The standards group acts as a delegated technical agent. So do the profile community, tool developer and automated reasoner. The affected principal cannot evaluate the delegation if the result hides which agent defined the rule, which artifact executed it and which actor converted it into a decision.

What the evidence does not establish

The FPWD is not a Recommendation, and this Article does not predict whether the axiom will survive. Absence from the retrieved Turtle file proves neither mistake nor intent. It establishes that the human and machine publication surfaces did not encode the same rule at the time checked.

The implementation report does contain named examples. Its placeholder table does not justify saying that there are no implementations, and a broken displayed link does not show that no tests exist. The issue counterexample is a serious input, not group consensus. No identified implementer is accused here of claiming more than its evidence supports.

The proposed receipt is Daniel Kade’s governance recommendation. It is not an announced W3C requirement, certification program or substitute for the group’s technical work.

Sources

  1. W3C News — First Public Working Draft: The Profiles Vocabulary
  2. W3C — The Profiles Vocabulary, FPWD of 20 August 2026
  3. W3C — Dated PROF RDF Turtle artifact
  4. W3C — The Profiles Vocabulary publication history
  5. W3C DXWG repository — Issue 29: conformsTo axiom at risk
  6. W3C DXWG repository — Issue 58: fix the property-chain axiom
  7. W3C Data Shapes repository — Issue 882 counterexample
  8. W3C DXWG — The Profiles Vocabulary Implementation Report
  9. W3C — Dataset Exchange Working Group Charter
  10. W3C — Standards and other document types
  11. W3C — Process Document, 18 August 2025
  12. Heng Lu — On the Agency Problem at the Core of Internet Governance