Skip to main content

Discovery & Verification Mechanisms (AI Context)

Purpose

This section extends the BPS DPP Discovery Mechanisms to describe how AI Products (models, datasets, evaluation pipelines, or composite AI systems) can be discovered, verified, and trusted across digital marketplaces and registries.

It defines concrete mechanisms and conventions used by AI producers and consumers to locate, retrieve, and validate Digital Product Passports (DPPs).


1. Background

AI Products exist entirely in digital ecosystems — model registries, data marketplaces, or federated MLOps platforms. Therefore, discovery is primarily URI-driven, often enhanced with visual or machine-readable trust badges.

These mechanisms ensure that:

  • Every model or dataset is traceable to its verifiable DPP;
  • Provenance, risk, and evaluation data are discoverable without manual lookup;
  • Marketplace and CI/CD systems can automatically validate authenticity and lineage.

2. Discovery Mechanisms for AI Products

MechanismDescriptionTypical PlacementExample
QR Code (Digital Display)Visual representation of the DPP URI (Lite or Full).Marketplace product card, PDF model card, research poster, web catalog.QR → https://aips.kivanura.org/dpp/model/cv-fairness-detector/v2.1?lite=true
Direct DPP URICanonical, versioned HTTPS link to the product’s passport.Registry metadata, API response, SBOM manifest.https://aips.kivanura.org/api/v1/dpp/model/cv-fairness-detector/v2.1
Embedded JSON-LD ContextInline reference in the model card or metadata file.model-card.json, model.yaml."@context": "https://kivanura.org/spec/aips/dpp/0.1/context.jsonld"
AI Manifest (SBOM / Model Manifest)Digital manifest listing dependencies, hashes, and DPP link.Build pipeline outputs, container metadata."dpp": "https://.../dpp/model/v2.1"
Registry API IntegrationNative API endpoint in model/data registries resolving to DPP.Hugging Face-like hub, internal MLOps catalog./api/v1/dpp/{modelId}/{version}
Verifiable Credential (VC)Signed credential asserting model authenticity and provenance.Enterprise trust frameworks, regulator audits.VC signed by issuer DID referencing DPP URI.

3. Lite vs Full Resolution

  • Lite DPP → flattened, human-readable; used for public discovery and preview.
  • Full DPP → includes lineage, evaluation metrics, and risk references; requires authorized access.

Both share the same schema (aips-dpp-0.1.context.jsonld) and can be resolved from the same base URI with the query parameter ?lite=true|false.


4. Verification Workflow

A post-modern AI marketplace or CI pipeline can validate a DPP as follows:

  1. Resolve the DPP URI (from QR, manifest, or registry).
  2. Verify the digital signature or content hash (dpp:signature, dpp:contentHash).
  3. Validate structure against aips-dpp-0.1.shacl.ttl shapes.
  4. Check lineage and risk artifacts for integrity.
  5. Cache or display the Lite DPP view for human users.

This workflow supports both interactive trust UX and agentic automation.


5. Relationship to BPS Discovery Mechanisms

The AIPS DPP Discovery profile inherits all generic mechanisms from BPS:

From BPSSpecialized in AIPS as
QR CodeAlso shown on digital listings (not only physical goods).
NFCOptional for edge-AI hardware (e.g., IoT models).
GS1 Digital LinkMapped to standardized AI model identifiers (future use).
URI/URLCanonical AIPS DPP endpoints.
Content NegotiationJSON-LD vs. HTML DPP views.
Blockchain AnchorOptional: record model passport hash on ledger.

6. Best Practices for AI DPP Discovery

  • Use versioned URIs with stable paths (/dpp/model/{name}/{version}).
  • Prefer HTTPS and canonical domain (e.g., aips.kivanura.org).
  • Display a scannable QR badge on every public model card or dataset page.
  • Maintain redirects from superseded URIs to the latest DPP version.
  • Include embedded context links in all AI artifacts to enable offline validation.
  • Support Lite DPP previews for transparency; Full DPP for compliance and audit.

7. Summary

AI Product DPP discovery is digital-native that is anchored in URIs, reinforced by QR visuals, and verified through signatures + schemas.
These mechanisms ensure that any consumer, be it a human agent or a non-human agent, can retrieve, verify, and trust the provenance of an AI model or dataset directly from its source.

Together, BPS and AIPS discovery layers provide a consistent, cross-domain approach to verifiable product transparency, from physical goods to AI products.