An institution can hold a detailed research archive, an active policy library and years of administrative records while still struggling to answer a straightforward question. The difficulty is often the boundary between collections. A project has one identifier in the research office, another in a repository and a familiar abbreviation in departmental documents. Searching each collection separately leaves the person asking the question to reconcile the results.
A connected knowledge layer should make those relationships explicit without flattening the distinctions that matter. A draft policy is not an approved policy. A research output is not the dataset behind it. A public catalogue entry does not make its underlying files public. Start by deciding which institutional questions the connection should help people answer and which boundaries it must preserve.
Step 1
Choose one question that crosses repositories
A useful starting question might concern the outputs of a particular project or the current procedure for a particular administrative activity. Identify the collections that contain the evidence and the people who maintain them. Ask those owners to describe the difference between authoritative records, working copies and historical material. This discussion usually reveals more useful requirements than a general request to make every document searchable.
Build a small inventory for the selected scope: repository, record type, owner, identifier, access rules and update process. Include known gaps and collections that cannot yet be connected. Treat the inventory as an operational document with an owner, rather than a one-time discovery spreadsheet. A search experience should tell users what it covers, especially when an incomplete collection could change how they interpret an answer.
Step 2
Connect identities and preserve context
Agree on the entities the institution needs to distinguish: projects, people, units, publications, datasets and policies, for example. Preserve the original identifiers alongside any shared identifier. Names alone are a weak basis for matching records because names change, abbreviations collide and organizational structures evolve. When a match is uncertain, expose it for review instead of treating it as a confirmed relationship.
The W3C Data Catalog Vocabulary offers a reference for describing datasets and their distributions across catalogues. It is useful when planning interoperable metadata, but adopting a vocabulary does not resolve institutional ownership by itself. Define who may publish a relationship, correct a classification and retire an obsolete record. Keep source links and status information close to the result so the person reading it can inspect the underlying evidence.
Step 3
Validate with the people who know the records
Prepare questions with librarians, research administrators and departmental staff. Include questions with no available answer, questions about superseded material and questions whose evidence is restricted. Assess whether the system finds the right records and explains their status. A fluent summary that merges a draft with an approved policy is a failure even if every sentence resembles something in the collection.
Test updates as well as initial ingestion. Change a record's status, remove a user's access and replace a source document. Check what appears in search results, summaries and citations afterward. Assign responsibility for unresolved matches and stale sources. The service becomes useful when staff can trust its scope and understand its limitations, not simply when a large number of files has been processed.
How sDEN approaches it
Connected knowledge. Inspectable evidence.
Apply these principles to a defined collection and a practical decision.
Within this scope
Connect existing repositories
sDEN Foundation connects documents and systems through a governed knowledge graph. Institutional teams can retain their repositories while making relationships between records available to search and private AI.
Within this scope
Keep access attached
Existing identities and permissions remain part of the knowledge boundary. Faculties, services and project teams should receive only the context they are authorized to use.
Within this scope
Make evidence inspectable
Source-linked answers let people inspect the document behind a response. Expert validation can establish which classifications and relationships are suitable for institutional use.
A useful first delivery
What your team should be able to use
A useful first delivery is a connected collection with a clear question set, named owners and documented access boundaries. Staff can follow a result to its source, distinguish current from historical information and report a mistaken relationship. The institution also knows what is outside the collection. That provides a practical basis for adding the next repository without repeating the same identity and ownership decisions from scratch.
Questions
Institutions & research, answered.
Must every repository use the same metadata model?
No. A shared layer can map selected fields while preserving source metadata. The mapping needs explicit definitions and ownership, particularly where two repositories use the same term differently.
Should all institutional records become searchable together?
Only within appropriate access boundaries. Start with a bounded use case and test permissions before broadening coverage. Connecting records does not imply granting everyone access to them.
Sources
