Port /data/discover/{identifier} to content-node

[adamjohnwright]

Spec 006 T082.

This endpoint describes an event as a schema.org Dataset, for search engines. It is DiscoverController over graph-core's SchemaDataSet, reproduced field by field (schemaDataSet).

Ordering: graph-core's findById here is DatabaseObjectRepository.findByStId/findByDbId: every relationship, both ways, unsorted. Node runs that load and maps it with mapResult, as Spring Data maps it. That load decides the order of creators and citations. Sorting by r.order, by relationship id, or by a one-type expand each disagreed with Java on some event.

Lazy loads: each edit's authors, and each author's affiliations, are loaded as graph-core's LazyFetchAspect loads them: its query (lazyCollectionQuery), in relationship order, repeated by stoichiometry.

Java's answers kept:

• Jackson's NON_EMPTY drops "", null and [].
• Not found is a 404, "{id} not found". That includes an object with no relationship at all, whatever its class.
• A non-event is a 422 with a null message.
• A dbId past a long is the 500.

Parity: 50 requests against Java across two runs, all byte-identical (an error body's url aside). They cover pathways and reactions, URL, Book and multi-summation citations, an edit with no author, every id form, orphaned objects, and the 404, 422 and 500. No names were printed while checking; comparisons were by bytes and hashes.

Review: an adversarial review found:

• an orphaned non-event answering 422 where Java answers 404 (fixed, and checked live);
• the NON_EMPTY gap for "" (fixed);
• several surviving test mutations. All six are now caught, and the handler's 200 path is tested end to end with a stand-in graph.

Routing: nginx and the dev proxy; the routing tests were shown failing first.

🤖 Generated with Claude Code