Conditional ODRL policy use through the DPV profile
Context and Problem Statement
Classifying data, its purpose and legal basis does not itself express a permission, prohibition or duty. Some reviewed use cases need policies about retention, changing conditions or cross-border transfer. A policy language can express those rules, but must not replace privacy classification or imply that a website enforces them.
Considered Options
- Conditionally use ODRL through the DPV Profile of ODRL.
- Adopt standalone ODRL with no DPV target or classification boundary.
- Invent permission and duty primitives within the classification vocabulary.
- Defer every policy representation.
Decision Outcome
Select the first option, limited to the three cases in R5. DPV remains the classification authority established by ODR-0054. Policy expression is neither a legal determination nor operational enforcement.
Consequences
- Good, because classification and policy expression retain distinct responsibilities.
- Good, because a bounded common vocabulary avoids incompatible policy dialects.
- Bad, because qualified humans must author and review each policy.
- Neutral, because deploying an enforcement engine is a separate implementation decision.
Confirmation
Accepted conditionally as method. OPDA must demonstrate the target/classification boundary, permitted profile and scope, negative validation cases and human approval before claiming implementation. No source-system shape or council vote proves those checks here. Neither policy coverage nor regulatory compliance is claimed.
Rules
R1: DPV profile required
Use ODRL only through the DPV Profile of ODRL. A bare policy without a DPV-typed target fails this selected profile. DPV describes the data and its privacy context; ODRL expresses policy over that data.
R2: Common Vocabulary only
Use ODRL 2.2 Common Vocabulary. Other profiles, including information-content management and media-rights profiles, are not implicitly adopted. Extending the profile requires a separately reviewed amendment.
R3: Separate vocabulary responsibilities
| Responsibility | Selected vocabulary |
|---|---|
| Data classification | DPV |
| Legal basis, purpose, actors and rights context | DPV |
| Policies, permissions, prohibitions, duties and constraints | ODRL |
| Privacy concepts used as policy operand values | DPV terms within the ODRL rule |
Do not use ODRL as a replacement classifier or DPV predicates as policy operators. The selected composition allows DPV terms as right-operand values; it does not turn DPV data classes into ODRL left-operand operators. A policy’s type alone says nothing about the nature of its target.
R4: Validate the boundary
Local SHACL MUST require each policy to have a target resolving to an appropriately
DPV-typed resource and reject the prohibited classification/operator combinations.
Both are violation-level requirements. Checking merely that odrl:target exists
does not establish the target’s type: the implementation must test that additional
condition explicitly. Validate actual rule paths, not only a convenient illustrative
policy skeleton. Failure blocks a package’s conformance claim.
R5: Exactly three conditional uses
- Retention policies expressing deadlines or elapsed-time conditions over classified storage-duration requirements.
- Permissions conditioned on dynamic state, such as consent status, an open retention window and applicable geography.
- Cross-border transfer policies composing the applicable reviewed transfer conditions.
Other policy cases require an amendment. The chosen duration or legal basis must come from qualified review; an example cannot establish either for a real property record.
R6: Composition, not automatic effect
Describe a target’s privacy context first. A human-curated policy then identifies that target, action and constraints using the selected profile. For example, a reviewed inspection-contact retention policy may express when further use is prohibited. Merely publishing that graph does not delete data, revoke access or execute a deadline. An implementation must separately connect a validated policy to an enforcement mechanism before claiming those effects.
R7: Human-curated policy only
Automated extraction and enrichment MUST NOT generate odrl:Policy,
odrl:Permission, odrl:Prohibition or odrl:Duty instances. Fields, code and
extraction confidence cannot establish regulatory intent. Policies must be manually
authored for a specific permitted use case by a person with relevant regulatory
context. Document this limit prominently with the implementation. Do not promise
that all data classes have policies or substitute a coverage percentage for evidence.
R8: No governance-ontology expansion
This decision concerns Category 11 only. Category 6 governance-ontology modelling is outside the selected OPDA concern set. This record does not authorise ODRL there or reopen that excluded category through policy examples.
R9: Reference without imports
Use canonical http://www.w3.org/ns/odrl/2/ vocabulary IRIs and the canonical DPV
IRIs selected by the adopted profile. Do not import either entire ontology with
owl:imports. Local shapes establish the delivery contract; canonical references
do not demonstrate implementation or enforcement.
More Information
- ODRL vocabulary
- Source-method record: ODR-0089, revision
67174057e6384b79d0b28b7736fe70a66e112895.
Amendments
- 2026-09-05 — OPDA adaptation. Preserves conditional adoption, profile limits, the three uses and binding human-curation restriction. Replaces source pipeline, council and execution claims with OPDA implementation obligations. Makes explicit the gap between a target-presence check and the required target-type validation.
Comments
Loading comments…
Sign in to post a comment