GUIDE, read 13 Nov 2026
A certified label is one person's sign-off on one date, and AI-ready is a state that must hold now
A certificate is a record of a decision, not a measurement
When a steward certifies a table, the catalog stores a status and, at best, a name and a date. The status says someone with authority looked at the asset and approved it. It does not say the table loaded this morning or that its tests passed last night.
Readiness is a different kind of statement. It describes the asset now: owned, documented, passing, current, in use, allowed. A certificate can be true on the day it is set and false three weeks later, with no change to the label itself.
Certification covers one of ten conditions and says nothing about the rest
The checklist treats certified as one condition among ten. At the moment of sign-off a reviewer may also have looked at the owner and the description, but the label does not record that, and four of the other nine can change without anyone touching the label.
- Quality tests pass: a test that passed at sign-off can fail on the next run.
- Fresh: a late load makes a certified table stale the same day.
- Active use: a table can be certified and then replaced by a newer one without being deprecated.
- Access policy allows AI use: an approval to use the table in reports is not an approval for an agent.
- Named owner: the certifier can leave the company while the certificate stays.
Seven ranked tools set the label by rule, seven leave it to a person
The Readiness Coverage Matrix reads each tool's documentation for who sets the certified status. The split on 13 Nov 2026 is below. A rule can set the label without a person marking each asset, which is the first step to keeping it current. It is not enough on its own, as the next section shows.
| Who sets the certified label | Ranked tools |
|---|---|
| Checked by rule | Euno, Atlan, Ataccama, OpenMetadata, Secoda, Databricks Unity Catalog, OvalEdge |
| Set by a person | Informatica, Qlik Talend Cloud, Precisely, Collibra, Alation, DataHub, Snowflake Horizon |
| None found on 13 Nov 2026 | Monte Carlo, dbt, Actian |
A rule that sets the label reads only the conditions it was given
Databricks documents a rule that sets system.certification_status to certified when the owner is a named principal, the table has a description, and it had more than 100 read queries in 30 days. That is three conditions: owner, documented and usage. The documented example reads no quality test and no freshness check, so a certified table under that rule can still be stale.
Secoda documents a rule that tags a resource Verified once it has a complete description and an owner, and a second automation that acts on its Data Quality Score. OvalEdge says a certification policy can be applied by rule and can re-run when the glossary terms, business description or lineage of an object change. Collate says a scheduled workflow sets Gold, Silver, Bronze or no certification on dashboards. Atlan says a playbook action sets the certificate to Verified, Draft, Deprecated or No certificate, and the pages read describe no link from its quality results to that certificate.
On the person side, Precisely documents certification as a request and approval workflow in which a data steward submits the request. Alation describes its Endorsement flag as set by users on an object. Snowflake's Horizon Context example has a data steward apply the CERTIFICATION_STATUS tag by hand. DataHub documents rules that apply tags, terms and owners, and a tool that sets a lifecycle stage (proposed, approved, deprecated) that a person or agent applies.
Neither route withdraws the label when the data breaks, unless it is built to
A person who certifies once and a rule that sets the label once have the same weakness: both are silent afterward. OvalEdge states the re-run triggers, and none found on 13 Nov 2026 for an automatic step that withdraws Certified when a later quality or lineage condition fails. For Snowflake the pages read show none found on 13 Nov 2026 of a rule that sets or withdraws its certification tag from quality, usage or lineage signals. Databricks says its tag automations assign or remove governed tags as data and metadata change, the closest documented withdrawal by rule in the set. The question to put to any vendor is what removes the label, not what sets it.
Add an expiry and a quality gate to every certificate
A certificate that cannot expire is a claim about the past. Treat a certified label as ready only while all of the following hold, and have the agent or the catalog check them at read time.
- The certificate carries a date, and the date is inside the review period your team set, such as 90 days.
- The latest quality run on the asset passed.
- The asset loaded inside its freshness window.
- The owner named on the asset still answers.
- The asset is not marked deprecated and still shows recent queries.
Break a certified table on purpose and see what the label does
Pick one certified table in a test project. Fail one of its tests, delay its load past the window, and clear its owner field. Read the label through the same route the agent uses after each change. If the label still says certified after all three, the label is a certificate and not a readiness state, and the agent needs a second signal beside it.
Where Euno is behind
Euno does not measure data quality itself: it reads health from tools such as Monte Carlo and Elementary, no filter on its labels for agents was documented on 13 Nov 2026, and it publishes no dollar prices.
Sources
- Databricks docs: automate tag assignment, read 13 Nov 2026
- Secoda docs: automation use cases, read 13 Nov 2026
- OvalEdge docs: certify objects, read 13 Nov 2026
- Collate docs: dashboard certification workflow, read 13 Nov 2026
- Atlan docs: playbooks, read 13 Nov 2026
- Precisely docs: certification workflow, read 13 Nov 2026
- Alation docs: trust flags, read 13 Nov 2026
- Snowflake docs: Horizon Context, read 13 Nov 2026
- DataHub docs: MCP server, read 13 Nov 2026
Last reviewed 13 Nov 2026.