Development previewBrowse and compare a sample of the catalog. Editing and personal features aren’t available here.About this demo
Understand the data
The import, drafting, and review processes used to maintain the catalog.
This page describes the maintainer's workflow. Visitors cannot add or edit catalog records on the public demo. To suggest a correction or ask about contributing, use Get involved.
Wikidata imports. A maintainer selects a topic and runs a preview that reports potential additions, existing records, and skipped items. After review, the maintainer can run the import. This route can save records as accepted because approval occurs before the write.
Manual entry. The maintainer enters the name, kind, facts, and a source. The entry and source are saved together. A source title, URL, or both is required for creating an entry through the form.
Drafting tools. Tools prepare suggested entries, classifications, descriptions, or connections. Suggestions wait for review. Some batch workflows store records as Proposed before a separate acceptance step.
Stage 1Prepare the record
Wikidata import
Manual entry
Drafting tools
Stage 2Check identity and fields
Identity
Kind
Fields
Connections
Stage 3Apply the route's source requirements
Wikidata import
Manual entry
Drafting tools
Stage 4Approve or store as Proposed
Accepted
Proposed
| Check | Purpose |
|---|---|
| External identity | Prevent an external identifier, such as a Wikidata ID, from being assigned to two entries. |
| Kind | Require registered entry kinds. Unmatched imports are skipped and reported. |
| Field validation | Check that values fit the relevant schemas, such as numbers for numerical fields. Unsupported fields are dropped. |
| Connection rules | Check supported labels, compatible endpoints, duplicates, and required qualifiers. Causal connections require a sign. |
| Source and review requirements | Apply the requirements of the particular import, form, suggestion, or batch workflow. |
These checks validate structure and workflow. They do not determine whether a source is right, whether a classification is appropriate, or whether a historical claim is true.
| Tool | Role |
|---|---|
| Wikidata suggestions | Propose classifications, images, or links based on existing Wikidata records. |
| Connection lookup | Find mapped Wikidata relationships between entries already in the catalog. |
| Page capture | Prefill an entry draft from a webpage's metadata. |
| Description templates | Compose descriptions from recorded facts and connections. |
| Network-based suggestions | Propose similarity links using shared neighbors. |
| Optional AI concept suggester | Suggest classification concepts for a person to review. |
| AI-assisted batch preparation | Help draft records from named published sources. |
Most of these functions use structured data or fixed rules. They are not all AI features.
Wikidata identifiers are obtained through lookup rather than trusted from AI output. AI output is not treated as a source for a catalog claim.
Proposed records are stored but awaiting acceptance. This status can apply to AI-assisted batches and to connections gathered from Wikidata. It describes review progress, not which tool produced the record.
Accepted records have passed their workflow's approval step. Manually entered records and reviewed imports may go directly to this state. A queued suggestion becomes an accepted catalog change when it is approved.
Before accepting a proposed batch, the workflow checks that the records still match what was reviewed. A declined proposal can be removed; it is not a finding that the claim was false.
The catalog's record status is distinct from the suggestions queue, which also tracks decisions about drafts.
The entry for ornament illustrates an AI-assisted batch workflow recorded in August 2026:
5 Aug 2026Draft prepared
“Ornament” was listed in a batch about modernism and postmodernism, written up with AI help from three named sources, including Adolf Loos, Ornament and Crime (1910), and Robert Venturi, Complexity and Contradiction in Architecture (1966).
5 AugClassified
Ornament holds no facts of its own; it is an idea that buildings and movements relate to. So it is a concept, and it also gets an entry of kind Concept, because connections need to point at it.
5 AugIdentity matched
Wikidata was searched for “ornament” and returned three candidates: decoration (Q335261), an Estonian band, and a 2007 DVD. Q335261 was taken because its name matches and its description says “decoration”. The ID came from that search, never from the AI.
5 AugSaved as ProposedProposed
The kind, facts and connection rules passed, and none of the sources named an AI model. The entry was written with two cited connections: modernism is opposed to ornament (Loos) and ornament is part of Postmodernism (Venturi). All of it showed the dashed Proposed mark.
10 AugBatch acceptedAccepted
A person reviewed the whole batch and accepted it. Each row was re-checked first: still there, still proposed, in that batch, and unchanged since review. The Proposed mark went away.
11 AugDescription reviewed
A one-line description was generated from its connections by a fixed sentence template. An earlier version was rejected; the accepted one describes ornament as part of Postmodernism and opposed to modernism.
12 AugImage reviewed
A suggestion tool found a public-domain drawing on Wikimedia Commons through the entry's Wikidata ID. The license was recorded as stated on the file's page, and a person accepted it.
Today: an accepted entry of kind Concept, anchored to Q335261, with two cited connections, an accepted description and a public-domain picture.
Accepting the entry did not automatically approve every later description or media choice.
No. Fixed rules check whether a kind is registered and whether its fields are valid. People choose or review the meaning of classifications. The optional concept suggester can offer ideas, but its output still needs a decision.
The Import preview illustrates a planned way to turn a sourced list of names into proposed entries. It is not a working public import tool.
A new kind requires a distinct set of useful fields, suitable examples, and human approval. Its schema, registry, and relevant data changes must be released together. See Classification.