SovAIHub
ModulesSAI-210
SAI-210 table of contents
Concept2 min readDraft

Artifact identity and provenance

Identify software, models, data, prompts, policy, and configuration with verifiable origin and lineage.

Last content review 2026-08-03Included in SAI-200, SAI-210, SAI-220, SAI-250, SAI-270

Before you start

Module prerequisites: SAI-100. Check the exercise-specific tools and environment before running commands.

Module outcomes: Create model acceptance criteria; Record licensing and provenance evidence; Define model update and retirement controls. This lesson contributes to that sequence.

Why identity must come before trust

An artifact name such as model-latest, base-image, or policy.json is convenient but not sufficient for a controlled decision. A sovereign AI lifecycle needs an immutable identity for every material software package, container, model, dataset, prompt, policy, configuration, and generated build output.

Identity answers what exact object is this? Provenance answers where did it come from, what happened to it, and who approved those events? Neither proves that an artifact is safe. Together they make verification, comparison, rollback, and investigation possible.

Minimum provenance record

For each material artifact, record:

  • Immutable digest or equivalent content identity.
  • Human-readable name, version, format, and size.
  • Origin, supplier or publisher, acquisition time, and acquisition method.
  • Declared license, usage conditions, and redistribution constraints.
  • Parent inputs, build recipe, transformations, and produced outputs.
  • Verification results, exceptions, reviewers, approvals, and timestamps.
  • Current lifecycle state: quarantined, evaluated, approved, promoted, revoked, superseded, or retired.
  • Links to evaluation, vulnerability, configuration, release, and deployment evidence.

Use signed attestations where the organization can protect signer identity and verify signatures. A signature without an understood identity and authorized purpose is only a cryptographic event.

Provenance graph

Treat provenance as a graph, not a single document. A served model may depend on model weights, tokenizer, runtime image, libraries, prompt templates, policy, hardware profile, and serving configuration. The release record should identify that complete set.

When an input changes, the graph makes the affected builds, evaluations, approvals, and deployments discoverable.

Verification questions

Before promotion, verify that the artifact identity matches the reviewed object, all required parents and transformations are known, approval applies to this version and purpose, evidence is integrity-protected, and revocation can prevent further use.

Implementation-specific manifest formats and signing commands belong in versioned adapters. The stable requirement is that identity and provenance remain independently verifiable.

Module worked example

A newer model advertises higher coding scores but changes its license and chat template.

Apply the control
Review the exact revision and terms, then rerun representative acceptance cases with the intended runtime.
Expected result
The decision records quality, operating constraints, rejected alternatives and a fallback.
Troubleshooting
Check tokenizer, template, quantization and parser versions before attributing a regression to model quality.

Practice and completion check

Choose one control from this lesson. Identify its actor, target, enforcement point and expected result. Describe an allowed case and a denied or failed case. For a runnable exercise, retain observed output and separate it from expected output. Record unresolved limitations before marking the lesson complete.

Your lesson progress

Completion is self-reported learning progress; it does not certify practical competence.

Status: Not marked

My learning

Loading your learning record…