Licenses for Datasets and Data

[Ludee]

Licensing of Composite Scientific Datasets

Problem Statement

Metadata standards such as DataCite define licensing at the dataset (DOI) level.

In practice, however, many datasets:

• Contain multiple tables (“resources”)
• Integrate third-party data
• Include public-domain materials
• Combine analytical content with raw source data
• Apply dual licensing

This creates a structural mismatch between metadata representation and legal reality.

---

1. Dataset-Level License (Metadata Perspective)

In DataCite:

• rights / rightsURI apply to the dataset as a whole.
• The model assumes a coherent licensing statement for the citable object.

This works well if:

• The entire dataset is owned by one rights holder.
• All components are released under the same license.

It becomes problematic if licensing differs across resources.

---

2. Common Licensing Scenarios

Scenario A – Uniform License

All components:

• Created by the same rights holder
• Released under a single license (e.g., CC-BY-4.0)

→ Straightforward.
→ Dataset-level license accurately reflects legal situation.

---

Scenario B – Dual Licensing (True Dual License)

The same dataset is offered under two alternative licenses:

• Example: CC-BY-4.0 or CC0-1.0

Conditions:

• The rights holder controls all rights.
• Licenses are alternative offers.

This is legally coherent and still dataset-level uniform.

---

Scenario C – Multi-License Aggregation

The dataset contains different components under different licenses:

Resource	Source	License
Table A	Public law texts	CC0-1.0
Table B	Own analytical content	CC-BY4.0
Table C	Third-party data	Restricted license

This is not dual licensing.

This is a composite dataset with heterogeneous licensing.

In this case:

• No single license fully describes all components.
• A dataset-level statement risks oversimplification.

---

Scenario D – Public-Domain Content + New Structure

Example:

• Public statutes (e.g., DataCite)
• Newly created classification system
• Analytical summaries

Legal structure:

• Statutes: public domain
• Summaries and structure: protected
• Database structure: possibly protected

The dataset-level license may apply to:

• The compilation
• The analytical additions

But not to the underlying public-domain texts.

---

3. Core Distinction

It is essential to distinguish between:

• Container license (applies to dataset as a packaged database)
• Component licenses (apply to individual tables or materials)

These are conceptually different layers.

---

4. Legal Risks of Oversimplification

If a dataset-level license implies uniformity while components differ:

• Users may assume reuse rights that do not exist.
• Third-party license obligations may be violated.
• Public-domain content may be incorrectly presented as licensed.

Transparency is legally safer than simplification.

---

5. Recommended Model: Layered Licensing

Layer 1 – Dataset-Level License (Container)

Define a primary license that governs:

• The database structure
• Newly created content
• The dataset as a compiled object

This is the license declared in DataCite.

Example wording:

“The dataset as a compiled database is licensed under CC-BY-4.0.”

---

Layer 2 – Resource-Level Licensing

For each resource:

• Specify license
• Identify rights holder
• Indicate source
• Clarify derivative status

This should be documented in:

• OEMetadata
• Machine-readable fields (if supported)

---

Layer 3 – Explicit Exceptions

Include a clarification statement such as:

“Individual resources may be subject to separate licensing conditions. See resource-level metadata for details.”

---

6. Terminological Precision

Distinguish clearly between:

• Dual licensing → Same material, alternative licenses.
• Multi-license aggregation → Different components, different licenses.
• Public domain inclusion → Not licensed.
• Container license → Applies to the database structure as a whole.

---

7. Practical Recommendation

For composite scientific datasets:

1. Always declare one primary dataset-level license.
2. Document deviations explicitly.
3. Avoid implying uniform licensing if it does not exist.
4. Separate third-party materials clearly.
5. Prefer structural transparency over legal ambiguity.

---

Conclusion

Metadata systems are DOI-centric and assume uniform licensing.
Real-world datasets are often composite legal objects.

The robust solution is a layered licensing model:

• Dataset-level (container)
• Resource-level (component)
• Public-domain differentiation
• Third-party transparency