AIPS DPP Alignment with BPS DPP Core

Purpose

This document defines how the AI Product DPP Profile (AIPS DPP) extends and semantically aligns with the Base Product Specification (BPS) DPP Core.
The goal is to ensure that any AIPS DPP instance remains a valid BPS DPP instance, while introducing additional AI-specific properties needed to describe machine learning and model-based products.

---

1. Design Relationship

The AIPS DPP Profile is an extension, not a replacement, of the BPS DPP Core.

Layer	Description
BPS DPP Core	Provides universal structure and vocabulary for all product passports — identity, issuance, integrity, and verification primitives.
AIPS DPP Profile	Adds AI-specific constructs such as model cards, evaluation results, training data lineage, and risk attributes.

All AIPS DPP artifacts (context, schema, SHACL) import or reference the corresponding BPS DPP Core artifacts.

---

2. Context Inclusion Model

Every AIPS DPP JSON-LD instance must include both contexts:

"@context": [
  "https://kivanura.org/spec/bps/dpp/0.1/bps-dpp-core-0.1.context.jsonld",
  "https://kivanura.org/spec/aips/dpp/0.1/aips-dpp-0.1.context.jsonld"
]

The BPS context establishes foundational predicates such as:

• dpp:subject, dpp:issuer, dpp:issuedAt, dpp:status
• Inline and ByRef pattern (schemaInline, schemaByRef, etc.)
• Common datatypes (xsd:string, xsd:dateTime, etc.)

The AIPS context introduces domain-specific extensions:

• aips:modelCardInline, aips:evalInline, aips:riskInline
• aips:trainingDataByRef, aips:evalByRef, etc.
• Additional quality, bias, and fairness properties aligned to AI lifecycle.

---

3. Semantic Mapping Table

The table below illustrates the semantic correspondence between BPS DPP Core and AIPS DPP terms.

BPS DPP Core Term	AIPS DPP Term	Relationship	Description
dpp:Passport	aips:Passport	rdfs:subClassOf	AIPS Passport is a subclass of DPP Passport.
dpp:schemaInline	aips:modelCardInline	rdfs:subPropertyOf	Model card acts as the schema summary for AI systems.
dpp:lineageInline	aips:trainingDataInline	rdfs:subPropertyOf	Training dataset provenance aligns with general lineage semantics.
dpp:lineageByRef	aips:trainingDataByRef	rdfs:subPropertyOf	References to data manifests or external datasets.
dpp:qualityInline	aips:evalInline	rdfs:subPropertyOf	Evaluation metrics extend general quality indicators.
dpp:policyInline	aips:policyInline	owl:equivalentProperty	Governance and compliance reporting maintain identical structure.
dpp:kgByRef	aips:kgByRef	owl:equivalentProperty	External graph reference mechanism remains identical.
dpp:provenanceInline	aips:trainingPipelineInline	rdfs:subPropertyOf	Describes execution lineage for training pipelines.
dpp:status	aips:status	owl:equivalentProperty	Lifecycle state inherits directly from BPS Core.

This ensures that AIPS-specific predicates can be reasoned about under the same semantic model as BPS Core without breaking inheritance.

---

4. Structural Alignment

Dimension	BPS Core	AIPS Extension
Product scope	Generic — any product (physical, digital, or hybrid).	Specialized — AI or ML-based digital products.
Integrity model	Hash + signature.	Same (inherits BPS).
Verification API	/api/v1/dpp/{id}	Same endpoint, extended with ?profile=aips.
Inline/ByRef pattern	Present for all logical sections.	Preserved — used for model card, eval, training data, and risk sections.
SHACL constraints	Core structural validation.	Extends with additional shapes for AI fields.

AIPS does not redefine identity or trust layers; it extends the payload domain only.

---

5. Example: Context and Inference

When an RDF/Linked Data consumer loads an AIPS DPP document, inference rules apply naturally due to subclass and subproperty relationships:

aips:Passport rdfs:subClassOf dpp:Passport .
aips:modelCardInline rdfs:subPropertyOf dpp:schemaInline .
aips:evalInline rdfs:subPropertyOf dpp:qualityInline .

As a result:

• All aips:modelCardInline statements are also valid dpp:schemaInline statements.
• Any SPARQL query for dpp:qualityInline will match both generic and AI-specific metrics.

This enables federated querying and reasoning across mixed DPP domains (e.g., BPS + AIPS + future verticals).

---

6. Validation Stack Integration

Layer	Validation File	Purpose
BPS DPP Core	bps-dpp-core-0.1.schema.json	Base structural constraints.
AIPS DPP	aips-dpp-0.1.schema.json	Adds AI-specific fields; includes $ref to BPS schema.
BPS SHACL	bps-dpp-core-0.1.shacl.ttl	Defines universal integrity and issuance rules.
AIPS SHACL	aips-dpp-0.1.shacl.ttl	Extends validation for modelCard, eval, risk, and training data.

Both validation layers can be executed independently or in cascade:

rdf-validate-shacl --shapes bps-dpp-core-0.1.shacl.ttl aips-example.ttl
rdf-validate-shacl --shapes aips-dpp-0.1.shacl.ttl aips-example.ttl

---

7. Backward Compatibility

• Any AIPS DPP that passes AIPS validation will automatically pass BPS Core validation for the common fields.
• Removing the AIPS-specific sections yields a syntactically valid BPS DPP Core instance.
• No breaking changes are introduced to the BPS namespace.

---

8. Versioning and Namespace Coherence

Element	BPS Namespace	AIPS Namespace
Namespace Root	https://kivanura.org/spec/bps/dpp/0.1/	https://kivanura.org/spec/aips/dpp/0.1/
Ontology File	bps-dpp-core-0.1.ttl	aips-dpp-0.1.ttl
Context File	bps-dpp-core-0.1.context.jsonld	aips-dpp-0.1.context.jsonld

AIPS DPP maintains strict version coupling with the underlying BPS version. For example, AIPS DPP 0.1 is bound to BPS DPP Core 0.1. Future releases (e.g., AIPS DPP 0.2) may track BPS 0.2.

---

9. Summary

• AIPS DPP extends BPS DPP Core through clean, RDF-aligned inheritance.
• All AIPS predicates are subproperties or equivalents of BPS Core predicates.
• This ensures semantic interoperability, validation consistency, and reasoning compatibility across all product domains.
• Future domain profiles (e.g., Industrial, HealthTech) can reuse the same BPS foundation following this alignment model.

---