Dataset, Distribution and Build Attachment: Three Entries in an Additive Handover

Consider a hypothetical handover: imagine opening a document directory that belongs to a manufacturing project. There are no filenames to read, no dates to sort by, no file sizes to compare. What remains is a skeleton of relationships: some items describe whole collections of data, other items describe particular accessible forms of those collections, and still other items describe who did what and when. The vocabulary needed to talk about that skeleton comes from two complementary traditions. One tradition, expressed through W3C DCAT, helps answer the question: what is being shared, and in what accessible form? The other, expressed through the W3C PROV Primer, helps answer a different question: what happened, to what, and under whose responsibility? Neither tradition, on its own, guarantees that the data inside the files is correct. They are vocabularies for organising descriptions, not for proving truth.

What a Dataset Is — and What It Is Not

In the catalogue vocabulary, a dataset is an organised collection of data that is treated as a single unit for purposes of description and discovery. The key idea is that a dataset is a conceptual object: it can be cited, versioned, and discussed even when no one is currently downloading anything. A common misconception is to assume that a dataset is identical to a particular file sitting on a server. That confusion collapses two different layers of meaning. The dataset is the conceptual collection; the distribution is one concrete, accessible form of it. According to DCAT, a distribution represents an accessible form of a dataset, such as a downloadable file. Once that distinction is accepted, it becomes possible to say that a single dataset may have more than one distribution — different formats, different resolutions, different subsets — without the dataset itself multiplying into several datasets. The conceptual unity stays intact even as the accessible forms vary.

This matters for a document directory because without filenames or dates, one way to document that two entries belong to the same dataset is to have a catalogue record that says so. The catalogue description is the anchor. If the directory contains only bare distributions with no linking description, then from a catalogue perspective it is not yet clear whether the entries belong to one dataset or several. The vocabulary does not resolve ambiguity automatically; it simply gives a structure in which the ambiguity can be addressed explicitly.

Distributions: Accessible Forms, Not Guarantees of Content

DCAT describes a distribution as an accessible form of a dataset. A description of such a form does not establish whether a particular reader currently has permission or working access, and it does not validate the content. A file may be perfectly accessible and still contain errors, omissions, or approximations. DCAT's concepts do not themselves prove data accuracy. This is a point that is easily overlooked when a directory is browsed at speed: the presence of a downloadable form can create an impression that the content has been validated. The vocabulary resists that impression. It separates the question "Can I get it?" from the question "Is what I get correct?" The distribution description helps a reader investigate the first question; actual access and content quality still require their own checks.

In a hypothetical handover directory, this separation becomes visible when one imagines two distributions that belong to the same dataset but were produced by different processes. One might be a raw export; the other might be a filtered or transformed version. Both are accessible. Both refer back to the same dataset through the catalogue description. Neither, by virtue of being a distribution, carries a built-in certificate of accuracy. The vocabulary keeps the two questions apart, and that discipline is useful when reviewers later ask what was actually handed over and whether it can be trusted.

Provenance: Entities, Activities and Agents Without Truth Claims

The second vocabulary layer addresses a different need. Where catalogue vocabulary organises what is shared, provenance vocabulary organises what happened. The W3C PROV Primer introduces entities, activities and agents for describing provenance: things, processes involving things, and responsible actors. An entity is a thing — a file, a report, a configuration snapshot. An activity is a process that uses, generates, or transforms entities. An agent is the person or system responsible for an activity. Together, these three categories allow a description of how a particular distribution came to exist in its current form.

Again, a misconception is common: writing a provenance description can feel like establishing a chain of truth. It does not. Provenance descriptions record relationships; writing such a description does not itself establish that every assertion is true. The description may say that a certain activity used a certain entity and produced a certain output, but the vocabulary does not verify that the activity actually ran as described, or that the input really was what the record claims. Provenance is a record of relationships, not a proof of correctness. In a document directory without filenames or dates, a provenance entry might say that a filtering activity generated a particular distribution from a particular dataset snapshot, under the responsibility of a particular agent. That statement is useful for traceability, but it does not, by itself, validate the content of the resulting file.

Where Catalogue and Provenance Meet in a Handover Directory

A handover directory can be imagined as having at least two layers of description layered over the bare entries. The catalogue layer says: these entries belong to this dataset; these are its distributions. The provenance layer says: this distribution was generated by this activity, using this input, under this agent's responsibility. Neither layer replaces the other, and neither layer substitutes for inspection of the actual content.

When a reviewer approaches such a directory, several useful questions arise that do not depend on any particular filename or timestamp. Is there a catalogue description that identifies the dataset, or are the distributions floating without a conceptual anchor? For each distribution, is there a provenance entry that records how it came to exist? Do the provenance entries reference entities that can be matched to catalogue records, so that the two layers can be cross-checked? Are there distributions without provenance, or provenance entries that refer to entities not present in the catalogue? These questions are structural; they can be asked even when the directory is stripped of all surface detail.

This is where an internal technical perspective becomes relevant. A team maintaining the site’s technical articles might choose to require that every distribution in a handover be linked to both a catalogue record and at least one provenance entry before the directory is considered complete. That is an editorial convention, not a guarantee of accuracy. It ensures that the vocabulary layers are present and cross-referenced, but it does not inspect the bytes inside the files. The convention is a discipline of description, not a substitute for content validation.

What the Vocabulary Cannot Do — and Why That Matters

It is worth stating explicitly what these vocabularies are not. They are not validation frameworks. They are not accuracy proofs. They are not manufacturing specifications. They are descriptive structures that allow a team to say, with shared meaning: this is the dataset, this is one accessible form of it, this is how that form came to exist, and this agent was responsible for the activity. That shared meaning is valuable precisely because it is limited. It does not overpromise. It does not claim that a file is correct because it has a provenance entry, or that a dataset is trustworthy because it has a catalogue record.

For an article archive that stores handover directories over time, the practical implication is that the archive should preserve not only the distributions themselves but also the catalogue descriptions and provenance entries that give them context. A future reviewer who encounters a bare file may need additional evidence to establish which dataset it belongs to, how it was generated, or who was responsible. The vocabulary layers are lightweight, but their absence is costly: it forces the future reviewer to reconstruct context that could have been recorded at the time of handover.

In a hypothetical handover review, the most productive use of this vocabulary is to ask structural questions before diving into content. Are the layers present? Are they cross-referenced? Are there gaps — distributions without provenance, provenance without catalogue anchors? Once those structural questions are answered, the substantive work of evaluating the data can begin on a clearer footing. The vocabulary does not replace that work, but it makes it harder to skip the foundational step of understanding what was handed over and how it came to be. That discipline, maintained consistently, is what gives a directory its long-term intelligibility — even when filenames, dates, and measurements are stripped away.