The census on the demo, 1:09. Recorded in Estate › Atlas on Meridian Simulation, a simulated retailer. Nothing is mocked up. Recorded under the working name, Ontos. Captions are in the player.

What to look at

Most tools show you what they found. The harder question is what they missed. A dashboard built on a warehouse that half the business routes around is confidently wrong, and nothing on it says so. The Estate Atlas starts from the other end: it names the blind spots.

The census

The Atlas is an inventory of your estate, and nothing more. It never proposes a strategy; it counts, with the evidence behind each count.

The Estate Atlas on the demo. Where to start: core.sales_by_channel is queried 2 times and nothing certified answers from it yet. 5 systems, 4 read and 1 partly read; 4 connections, all scanned; 146 assets (51 tables and views, 36 procedures, 32 query patterns; 4 hidden from answers); 184 links between assets (102 reads, 37 calls, 25 writes; 119 read from SQL, 46 declared by the source, 19 from imported manifests); 100% traced to a source, 68 with strong evidence.
On the demo, Estate › Atlas. Five systems, four connections, the evidence behind every line between them, and where to start.
  • Systems are declared by a person. Terazu never invents one.
  • Assets are the tables, views, reports and models it knows about. When a source table disappears, its record is kept and marked, not deleted.
  • Evidence coverage is the share of assets with at least one piece of evidence behind them.
  • How we know splits every line between assets by where it came from: declared by the source, read from SQL, or taken from an imported manifest.

Dark territory, named

Under the counts, “What Ontos can’t see yet” lists the blind spots: systems with no connection, systems known only because another tool’s export mentions them, connections never scanned. Below them, each connection lists the captures it hasn’t run.

What Terazu can't see yet, on the demo: 1 blind spot, 2 captures to run. Not connected: Meridian Finance (Cognos), known only from an imported manifest, never scanned. Not read yet: the point-of-sale Oracle database, its query log not captured, with a Read query history button, and no table profiles yet, which Oracle can't give.
On the demo, Estate › Atlas. One blind spot, with its reason, and the captures one connection hasn’t run: the point-of-sale database’s query log and its table profiles. Query history is read only where a person turns it on.

A blind spot isn’t a failure. It’s the difference between “we checked, and there’s nothing” and “we never looked”, written where everyone can see it.

What each system is missing

Each system has a card with its state (read, partly read, connected but not read yet, or not connected), the owner your people declared, and what Terazu holds for it. The captures it’s missing sit under “What Ontos can’t see yet”, grouped by connection: where the source supports it, one click runs the capture, after stating what it will run on your system. Query history is different: it’s read only on a connection where a person has turned it on in Connections. Until then the Atlas says it’s off, not that it wasn’t captured; on the demo it’s on for the point-of-sale database and not read yet, so the Atlas offers to read it.

System cards on the demo, each with its owner and how it is known. Meridian Analytics Warehouse: warehouse, read, owned by the Data warehouse team and Priya Sharma (Data Steward), 89 assets (43 tables, 32 query patterns, 12 procedures, 2 views), evidence on 100%, 49 relations, 34 declared foreign keys, and the lines inside it by schema. Meridian Data Platform (Airflow): orchestrator, read, owned by the Data platform team, 8 assets, 6 Airflow tasks. Meridian Finance (Cognos): partly read, owned by Finance BI and Tom Okafor (Finance steward), 23 assets (13 model measures, 6 model objects, 3 reports, 1 semantic model), no connection, known only from an imported manifest. Meridian IDMC: read, owned by the Integration team, 19 assets, 17 IDMC tasks. Meridian Merchandising & POS: Oracle, read, owned by Store systems and Dev Patel (Store operations manager), 7 assets (6 tables, 1 package), its query log and table profiles not captured yet.
On the demo, Estate › Atlas › Systems. Each card says how the system is known and who owns it: the warehouse, Airflow, Informatica and the point-of-sale system are read through the agent, and the Cognos model is known only from its files. Every system has a declared owner.

Who is going around the contract?

Where query history is on, bypass evidence compares it with what has been certified: tables people query that no certified definition covers, and certified tables nobody queries. Where it’s off, the Atlas says demand there is unknown, not zero. It reports findings, never tickets, and never names the people behind the queries.

What it doesn’t do

  • It can only name what’s declared or referenced somewhere. A system nobody mentions is simply invisible.
  • Evidence coverage says evidence exists, not that it’s correct.
  • Two connections to one warehouse are counted twice, and the card says so.
Keep reading The next best step Bring your Power BI model Earned context