Metadata-only is a feature

REDWOOD CITY, CALIFORNIA — Row sampling sounds helpful until security review starts. A vendor asks for access to result rows so they can “profile quality.” The room goes quiet. Legal wants a DPA. Security wants a network diagram. The deal that was supposed to catch stale dashboards becomes a project about moving customer data into someone else’s cloud.

VectorData does not take that path. The product is opinionated: table timestamps, dbt artifacts, job statistics. Not result rows. Not columns of customer data for profiling. The warehouse stays the warehouse.

That constraint makes the software better. It forces the product to use signals that actually exist in metadata — freshness SLAs, run status, bytes billed, blast radius across models. If a vendor needs your rows to tell you a table is stale, the interesting question is why INFORMATION_SCHEMA was not enough.

What VectorData can see

Update times. Job stats. Bytes billed. dbt run_results and manifests. Enough to know that fct_orders did not land, that a child skipped, that a slot of warehouse spend spiked with a bad scan. Enough to rank what to fix first. Not enough to reconstruct a customer, an order line, or an event payload.

‹›

“The best warehouse reliability tools refuse to sample your rows. That is not a limitation. It is how you stay in the deal.”

Ricky Jiménez Sparks, CEO of Vectornosis

What a review actually asks

Security does not ask whether your charts are pretty. It asks where the rows go, who can query them, whether the vendor can write, and what happens to artifacts when the contract ends. VectorData’s answers are short on purpose. Rows do not go anywhere. The service account is read-only. The product cannot usefully write. Metadata is enough to know that a table is late, a run failed, or a job billed like a mistake.

That is why we would rather lose a deal that requires sampling than win it by taking a copy of the warehouse. The second path turns a freshness product into a data-residency project. The first path keeps the evaluation about whether Watchdog is the queue the team will actually use.

How VectorData uses what it can see

Timestamps become freshness. Run results become failures and skipped children. Job stats become cost spikes. Together they become a ranked incident. None of that requires an order line. If a vendor’s quality story starts with profiling columns of customer data, they are building a different product — one that security will eventually have to unwind.

Read /security. Send the questionnaire to hello@vectornosis.com. The meeting should confirm the architecture, not discover it.

Prefer a dedicated service account with the narrowest roles that make monitoring work. Access is read-only. The full breakdown lives on /security. If you are in review, email hello@vectornosis.com and we will walk the questionnaire.