X-PD-001 Evidence Presentation and Disclosure¶
A connected evidence set can serve several recipients without giving each recipient every claim. These draft rules apply across asset, valuation and condition profiles. Signed end-to-end presentations have not yet verified the design.
Three separate layers¶
- Evidence composition: independently issued credentials, subjects, versions and typed relationships form an evidence graph. It need not be centrally stored, universally visible or signed as one document.
- Presentation selection: a holder or authorised agent chooses credentials and relationship evidence for a request.
- Claim disclosure: within each selected credential, the holder reveals only claims that the issuer made selectively disclosable.
SD-JWT, RFC 9901 supplies the third layer. It does not establish graph integrity, completeness, access permissions or institutional acceptance. The documentation tree, evidence graph and recipient's disclosed view are distinct. Combining credentials never makes one issuer attest to another issuer's claims.
Disclosure rules¶
The issuer fixes disclosure granularity at issuance, including nested claims where supported. Deleting a cleartext signed claim cannot turn it into a valid hidden claim. Credential specifications must describe their intended granularity.
This is the intended disclosure model. The current reference prototype supports selective disclosure only for address_text; full granular disclosure and signed end-to-end evidence selection remain implementation work. Examples below describe a possible recipient view, not additional implemented disclosure capabilities.
Requests identify purpose, recipient context, profile and policy versions. Distinguish required, optional and not-yet-applicable evidence. A request does not itself authorise disclosure; consent, access control and secure transport remain separate requirements.
Relationships need issuer-authenticated references or another defined binding. A package manifest alone cannot prove that two credentials concern the same asset, loan or version. If a needed binding is hidden, the relationship remains unverifiable. The privacy-preserving relationship mechanism is still implementation work.
Withheld evidence is neither false nor verified. A valid signature on a subset does not prove graph completeness, and non-disclosure cannot establish non-existence. Status, freshness, authority, supersession and dependency checks still apply.
Results and examples¶
A synthetic valuation review might disclose valuation amount, date, basis, asset binding and relevant condition evidence. A lending assessment might additionally require security and borrower claims linked by issuer-authenticated loan_ref values. A bank's role does not automatically widen access. Funding evidence before funding is not yet applicable; evidence becomes required when funding is asserted.
If the review requires observation.date and it is withheld, required_claims_disclosed is unavailable, the affected policy layer is UNKNOWN, and the technical projection is DENY. It is never PASS or ALLOW. This means insufficient evidence, not a finding of falsity or an invalid signature. The same principle applies to a hidden asset_ref needed for asset binding.
The proposed Verification Result Contract carries requested_paths, withheld_required_paths and over_disclosed_paths. These are paths only, never values, and are diagnostics rather than cryptographic proof. Non-empty withheld_required_paths forbids a satisfied disclosure check.
Privacy limits¶
Identifiers, document types and graph edges can correlate recipients' views. Minimise metadata and separate recipient-visible results from detailed diagnostics; do not log undisclosed values by default. Ordinary SD-JWT provides neither unlinkability nor predicates such as “value exceeds a threshold”, and cannot retract knowledge already disclosed. Reused credentials retain their original observation and issuance dates.
A local tool, website or agent connection has its own processing and retention boundary. Disclosure support alone does not establish that boundary or the security of an implementation. See Integration profiles.