The provenance model

MatSci-SAM publishes contribution activities, agents, entities, and derivations as W3C PROV-O. Each term has a graph and timeline at its readable path plus /provenance. Append .ttl or .jsonld for RDF. The history describes the term across candidates and revisions.

What the record contains

A definition revision is a prov:Entity and a prov:specializationOf its stable definition. prov:wasRevisionOf identifies the predecessor. A restoration or derived candidate also names its source revision separately from the chronological predecessor.

A suggested revision creates a new definition whose first revision states prov:wasDerivedFrom the source revision. A replacement proposal identifies the stable definition it should supersede. Both retain the original.

Publication is a prov:Activity associated with the person who published the revision. The revision is attributed to that person and, for an accepted model draft, the model. Suggestions store the requested term, contributor context, model output, prompt, and tag. Revision suggestions also store the source revision and critique. The published candidate links to that exact suggestion record. Discarded drafts remain outside the vocabulary.

An example identifies the stable definition and the exact revision displayed at contribution time. Its text and attribution are immutable. A featured selection records the selector and its active interval. The active example set can change independently of definition revisions.

Imported records state their limitations. Partial revisions can lack an editor or change note. Legacy examples lack independent author, exact source revision, publication time, and selection provenance. Imported comments use an inferred revision association from the recorded time. Imported votes use the revision current at migration. Unknown facts remain omitted.

People and models

People are prov:Person agents. Models are prov:SoftwareAgent agents. Authors, editors, commenters, and assertion authors remain attributed even when their profiles are private.

The per-term view labels votes "A community member" and omits voter identity. The dataset graph names a vote agent only for a public profile or AI account. The voting act remains in the graph when its agent is omitted.

A person uses a fragment node such as {document-IRI}#user_{id}. The account number is consistent across documents, but the document-specific IRI is not a public profile address. Model agents in the dataset graph use resolvable /models/{slug} IRIs. Per-term records identify models by their recorded runtime names.

Assertions, vote events and studies

Each ledger assertion is a matsci:Assertion and prov:Entity at {subject-IRI}#statement-{key}. It reifies a triple using rdf:reifies and an RDF 1.2 triple term, and records attribution and generation time. A retraction adds invalidation time and matsci:retractedBy. Only active assertions contribute their triples to current SKOS exports. Derived triples have no independent assertion rows.

A vote event is a matsci:VoteEvent and prov:Activity at {revision-IRI}#vote-event-{id}. It records the revision, time, matsci:voteKind of up, down, or withdrawn, and matsci:actorKind. Direction changes and withdrawals append events.

The vote-event backfill created one event per standing vote that lacked an event. matsci:backfilled identifies it. Older votes with an inferred revision association also use matsci:legacyAssociationInferred. Their recorded time may be the definition creation time. The backfill does not reconstruct a complete earlier sequence.

A study is a matsci:Study and prov:Activity with title, window, and collection under matsci:worklist. Study votes and comments name that study with matsci:study. A proposed definition records the Position step, with a source derivation only when it came from a suggested revision. Community rosters and invitations remain outside the RDF.

Metadata access lists graph documents. The repository shapes/ directory contains their SHACL constraints.

The two views

SKOS publishes current meaning and classification. PROV-O publishes the activities and attribution behind those records. A topic assignment is dcterms:subject in SKOS and an attributed assertion in the provenance graph. The active examples are all exported, regardless of the featured choice.