data-product-reviewer

Score the dataset against the DAUTNIVS attributes (Discoverable, Addressable, Understandable, Trustworthy, Natively accessible, Interoperable, Valuable, Secure) on a 0-2 scale using evidence like catalog entries, contracts, and SLO docs. Totals of 14-16 with no zero in core attributes qualify for publication; below 10 it remains just a dataset.

DAUTNIVS is an eight-attribute usability rubric extending FAIR principles: Discoverable, Addressable, Understandable, Trustworthy, Natively accessible, Interoperable, Valuable, and Secure. Each attribute is scored 0 (absent), 1 (partial or manual), or 2 (present and self-serve) based on artifacts, not assurances.

Run an agent-consumability probe: check whether a programmatic client can resolve the dataset's address to metadata, schema, sample data, and access policy without human intervention. A product only navigable by humans scores at most 1 on Discoverable and Understandable.

Pull the catalog or marketplace entry, the data contract and SLO document from Git with CI history, responses from the product's discovery and SLO APIs, storage snapshot history, and consumer records such as experiment trackers pinning checkpoints to dataset versions. Self-reported claims without artifacts score low.

Only datasets intended for cross-team or agent consumption should enter review. Exploratory research data shared within one team should be excluded so the review process does not become gatekeeping theater for early-stage work.

A zero score on Secure blocks publication regardless of total score, and any zero in a core attribute caps the verdict at beta or non-product. Scores below 10 mean the asset remains a dataset, and unresolved hard-stop policies prevent even beta publication.