Does digital curation begin after a project ends?

Ideally it begins during planning, because capture methods and documentation choices shape how usable the material will be later.

Is preservation part of curation?

Yes. Preservation supports continued access, while curation also includes selection, description, contextual management and reuse throughout the lifecycle.

Must all curated material be public?

No. Appropriate access can include restrictions. The collection should document who may use the material and under what conditions.

AI Glossary / Data engineering and architecture / Beginner

Digital curation

Definition

Digital curation is the active management of digital material throughout its lifecycle so that it remains understandable, trustworthy and usable for its intended purposes. It includes selecting, describing, maintaining and preserving material, as well as supporting appropriate access and reuse. Saving a file is one action within curation, not the whole process.

What it means

A spreadsheet can survive intact while becoming difficult to use because nobody remembers its units, code meanings or collection method. Digital curation preserves the context needed to interpret the contents alongside the contents themselves. It asks what future users will need and who will maintain that information.

The term is broader than choosing interesting links or arranging an online collection. In research and information management, it describes sustained responsibility for digital objects and their usefulness. Curation can begin before data are created, when teams decide formats, naming conventions, documentation and responsibilities. Later repair is often harder when essential context has already disappeared.

How it works

The team defines the material's purpose and likely users, then plans how it will be created and documented. It selects what should be retained and captures descriptive, technical and administrative information. This may include provenance, version history, access conditions and relationships among files.

Material entering a managed collection is checked against the collection's requirements. Storage, integrity checks and access arrangements support ongoing use. If formats or dependencies become difficult to maintain, preservation actions may produce a new representation while documenting what changed and why.

Curation is iterative. New uses can reveal missing documentation, and changing circumstances may require reappraisal. Some material may be retained, some transformed and some disposed of under the relevant policy. Keeping everything indefinitely without understanding it is not automatically a successful curation strategy.

Three practical examples

1. A research team deposits survey data with a codebook, collection notes and access conditions. Future users can interpret categories instead of guessing from abbreviated column names.

2. A design studio preserves project files together with fonts, linked assets and software information. It tests whether a representative project can still be opened and rendered as intended.

3. An AI team manages dataset versions, label definitions and transformation records. This helps later evaluators identify which material supported a particular experiment and which limitations still apply.

Think of it this way

Think of caring for a museum collection. Keeping an object safe matters, but so do its label, history, condition and rules for handling it. The analogy is limited because digital objects can be copied exactly while still depending on changing software, formats and access systems. Their physical survival alone does not ensure that people can interpret or use them.

Worked example

Suppose a fictional project creates 12 monthly measurement files, one per month. Every file has 1,000 rows, so the collection contains 12,000 observations. The team also creates a codebook stating that value 999 means missing, not a measured quantity.

An analyst who ignores the codebook might include 999 in an average and produce a misleading result. A curated package links each file to the correct codebook version, records units and explains the collection process. It also identifies which months used a revised sensor and where that change affects comparability.

The files' checksums can help detect later byte changes, but matching checksums do not establish that the interpretation is correct. Conversely, a documented conversion may legitimately change the bytes while preserving the information required for use. Curation brings integrity, meaning and change history together rather than relying on one technical check.

Illustrative case study

Fictional case: CurateHarbour inherits a folder of training examples from a completed project. Several files have similar names, and the label guide refers to a category that no longer appears in the latest export. The original project lead has left.

The team inventories the available material and identifies verified relationships between versions. It preserves uncertainty where it cannot establish which guide belongs to a file. It then creates a managed release with documented scope, provenance and known gaps, while keeping the source material distinguishable from the new package.

The lesson is that curation should make evidence and uncertainty visible. Inventing missing history would make the collection look tidier while making it less trustworthy.

Common misconceptions and mistakes

Do not equate curation with backup. A backup can restore stored data after loss, but it may not supply the context or access arrangements needed for future reuse. Avoid assuming that putting files in a repository makes all curation work complete.

Do not silently replace original values during cleanup. Record transformations and preserve appropriate links to their sources. Finally, do not assume open access is always the correct access model. Some collections require restrictions, and curation must support the intended users without ignoring the conditions attached to the material.

Business use

Digital curation helps organizations reuse prior work, maintain evidence and avoid repeatedly reconstructing lost context. It is especially useful when projects outlast their original teams or when datasets support several analytical purposes.

Assign ownership, maintenance effort and review points early. Specify what a future user should be able to find, understand and do with the material. For AI projects, preserve the relationship between raw sources, processed datasets, labels and experiment outputs. This supports traceability, but it does not by itself establish that a dataset is representative or that a model is suitable for deployment.

Check your understanding

A dataset opens correctly, but nobody knows whether its temperature values are Celsius or Fahrenheit. Is file readability sufficient for reuse?

Answer: No. Essential representation information is missing. Curation must address meaning and context alongside the ability to retrieve and open the file.

Related terms

Metadata
Data lifecycle
Digital preservation
provenance
data stewardship
repository
fixity
versioning
appraisal
representation information

Learn next

Learn metadata and preservation planning next. Select a small dataset and ask another person to interpret it using only the supplied documentation. Record every unanswered question, then add verified context and identify any gaps that cannot be resolved.

Metadata

Sources

Digital Curation Centre, What is digital curation?, explains active management, preservation and reuse across the data lifecycle: https://www.dcc.ac.uk/about/digital-curation

Digital Curation Centre, Curation Lifecycle Model, describes the model's role in mapping activities and responsibilities: https://www.dcc.ac.uk/guidance/curation-lifecycle-model

The explanatory pages were read. The measurement-file exercise and CurateHarbour case are original illustrations, and no specific repository certification or preservation guarantee is claimed.

Previous: Digital continuity Next: Digital divide