Verifiable
evidence.

Signed records. Checkable receipts.

SCITT provides an architecture for registering signed statements with a transparency service. The service checks its registration policy, records an accepted statement and issues a receipt. A verifier can then check the statement and its inclusion in the log. The architecture is defined in IETF RFC 9943.

Krnali VE implements that evidence pattern with issuer checks, a persistent transparency log, cryptographic inclusion receipts and an offline verifier. The original observation, signed statement and receipt can be retained and exported for independent inspection.

Evidence around verifiable credentials.

A verifiable credential carries claims from an issuer to a holder and a verifier. Its proof helps establish where those claims came from and whether they have been altered. The issuer's trustworthiness and the decision to accept a credential still need their own rules.

Krnali VE adds evidence of the surrounding process. Our credential integration observes retained issuance and verification records, signs selected event data and registers the statements with the transparency service. Teams can inspect what the source recorded and verify the resulting evidence offline.

The current mapping excludes credential contents and disclosed personal claims. It retains selected references, outcomes and fingerprints. Capturing these events does not by itself prove that an entire business process completed.

Make agent activity inspectable.

AI agent transparency means being able to examine the records behind an outcome: what the system reported, which authority decision applied and whether a request was allowed or refused.

Our development integration links retained agent outcomes with authority decisions in Krnali SAR, so teams can trace an outcome back to the decision that allowed or refused the action. Selected observations are signed and can be checked independently; prompts, model answers and credentials are excluded from the retained artifacts.

Krnali SAR controls the authority to act. Krnali VE preserves evidence for inspection. Recording a decision never grants or renews that authority, and a model's explanation is not proof that an action took place.

A working core, with a defined scope.

Signed statements, issuer checks, persistent logging, inclusion receipts, evidence export and offline verification are implemented. Credential event capture and correlated agent and authority records have been exercised in development deployments.

Production hardening, broader integrations and full SCITT conformance work remain. Current source integrations depend on the records their source systems retain and expose; they do not guarantee complete capture or whole-process verification.

Continue exploring.