Methodology

Start with the right kind of record

How entries are chosen

We include a record when it helps explain a distinct function, method, implementation, or standard and has a traceable source. The collection is selective. An absent product has not been judged unsuitable. Model releases follow a separate inclusion rubric in the Computational Model Studio.

Legacy spreadsheets and diagrams suggest candidates, but their descriptions are checked against current primary documentation before any public record is added. This catalog does not test vendor performance claims or rank products.

Sources and review

Public records link to the documents used for their claims and show when those documents were last checked. That date is an editorial check, not a guarantee that a product or standard remains unchanged. A record’s review state and the described product’s lifecycle are separate: a historical item can be well sourced without being available now.

Source authority depends on the claim: provider or developer documents for identity and access, specifications for terminology, technical reports for architecture, and clearly attributed analysis for interpretation. Conflicting or uncertain claims remain unasserted until resolved.

Boundaries matter

The six areas are connected lenses. A computational model’s designed function is different from what a surrounding application achieves by adding retrieval, tools, memory, or human approval. Model artifacts and hosted offerings are distinct access paths. Governance, evaluation, and assurance connect across every area.

A controlled vocabulary organizes labels; an ontology can add formal semantics. A stored vector index is not by itself a memory policy. A published framework is not proof of compliance. A documented API is not proof that a service behaves as described.

Updates and reusable data

Identifiers remain stable as records are reviewed. Material additions and classification changes appear in Updates. The export manifest identifies the data release, exact content digest, schema, and downloadable files. Each export carries source IDs so claims can be traced to the source register.

Corrections are reviewed against the cited source and recorded in the next data release. The current site does not collect visitor information or fetch catalog data from outside services when a page loads.