RDF Policy Reviewer

@ia2-dev/rdf-policy-reviewer is a generic web component that observes an HTML/RDF document, validates its current dataset against a separately supplied SHACL profile, and explains RDF statement changes.

<iframe id="draft" src="contract.html"></iframe>
<ia2-rdf-policy-reviewer
  source-frame="#draft"
  profile-root="#counsel-policy"
  diff-graphs="https://example.test/contract#runtime"
></ia2-rdf-policy-reviewer>

The component understands RDF, SHACL Core, SHACL-SPARQL, labels, and optional Web Annotations whose oa:hasBody is a source shape and whose oa:hasTarget identifies one or more presentation locations. It contains no contract or legal vocabulary. The profile remains an independent RDF artifact and can stay internal when the reviewed document is distributed.

Local roots are observed with MutationObserver. Framed documents can also publish ia2-rdf-dataset-change. Override the whitespace-separated change-events attribute when another host has its own dataset-change event.

Assign Document, DocumentFragment, or Element values to the sourceRoot and profileRoot properties when CSS selectors are not appropriate. Profile extraction errors stop review rather than producing a misleading conformance result.

For SHACL, the HTML/RDF adapter deliberately projects each extracted dataset's named graphs into one explicit RDF graph. The headless validatePolicy(data, shapes) export accepts ordinary RDF/JS DatasetCore graphs and has no HTML dependency.

import { validatePolicy } from "@ia2-dev/rdf-policy-reviewer/validation";
import { diffQuads } from "@ia2-dev/rdf-policy-reviewer/diff";

Those subpath exports do not register a custom element or require browser globals. Import the package root only when using <ia2-rdf-policy-reviewer>.

Semantic diffs require stable named resources. Configure diff-graphs to a runtime or application graph without blank nodes. diffQuads rejects blank-node-bearing inputs instead of treating extraction-local blank-node labels as persistent identity. The component keeps the dataset observed when review begins as its baseline, so transient keystrokes do not become the next comparison point. A single-valued replacement is presented as one change; multivalued RDF retains separate additions and removals.