Proposal: add optional audit manifest example

Proposal

Would this examples repo be open to a tiny audit/provenance manifest example next to one of the simple MLCubes, such as hello_world or matmul?

Related proposals:

• Core MLCube optional run audit manifest: https://github.com/mlcommons/mlcube/issues/367
• Cookiecutter scaffolded manifest example: https://github.com/mlcommons/mlcube_cookiecutter/issues/10

The examples repo seems like a low-risk place to demonstrate the idea without changing MLCube runtime behavior, schemas, or generated templates.

Suggested example file

For example:

hello_world/
  audit_manifest.example.json

Example contents:

{
  "schema_version": "mlcube.run_audit.v1",
  "example": "hello_world",
  "mlcube_task": "hello_world",
  "runner": "docker",
  "inputs": ["workspace/parameters"],
  "outputs": ["workspace/output.txt"],
  "claim_status": "diagnostic",
  "redaction_status": "safe_for_public_log",
  "notes": "Example sidecar manifest for recording reproducible, audit-safe MLCube run metadata."
}

Why this may help

• gives users a concrete pattern for recording run provenance and result status
• keeps the manifest optional and separate from runtime behavior
• helps distinguish diagnostic/internal runs from public/reportable artifacts
• encourages audit-safe metadata without storing raw secrets, private paths, tokens, or sensitive arguments

Possible first PR scope

If maintainers are interested, I can prepare a small PR that:

• adds audit_manifest.example.json to hello_world or matmul
• adds one short README note explaining that the manifest is optional
• does not change code, runner behavior, or benchmark outputs

This is motivated by AANA work around audit-safe AI evaluation artifacts, but the contribution would be generic to MLCube examples and would not require AANA as a dependency.

Review the hello_world and matmul example directories, then choose one for audit_manifest.example.json as proposed. Add the optional manifest and one short README note, preserving runtime behavior and benchmark outputs; done means the example documents audit-safe metadata without secrets or sensitive paths.