Validate and publish context
After governance, organize asset semantics, standards, lineage, and reviewed documents into context that AI can cite. This phase validates knowledge, scope, and the query agent's ability to explain assets. It does not automatically deliver underlying business data.
Enter Context Management
Complete space governance and verification, then open Context Management. Resolve pending proposals, failed tasks, and business disputes so queries use confirmed results.
Define the context scope
Context can include asset business names and descriptions, standards, lineage and impact relationships, metrics, reports, reviewed documents, dictionaries, and policies. Select materials based on intended questions and exclude unrelated or outdated content. Publishing descriptions and knowledge does not authorize release of underlying data.
Review the Context quality report
The report identifies whether materials support queries across asset counts, semantics, standards, lineage, and classification. It is neither a model answer accuracy measure nor a quality rule execution report.
| Dimension | Question | Remediation |
|---|---|---|
| Asset count | Are required objects present, with unnecessary objects excluded? | Adjust space scope |
| Chinese business semantic completeness | Can users understand table and field meanings? | Enrich and review business metadata |
| Standards coverage | Are key fields associated with correct standards? | Review standards mapping and validation |
| Lineage coverage | Can sources and flows be explained? | Check metadata lineage or SQLink results |
| Classification and security levels | Are sensitivity and levels clear before external use? | Complete classification and security review |
There is no universal passing score in this guide. Combine the report with task and organizational requirements. Completeness statistics cannot establish that every definition is correct.
Validate the query agent
The query phase uses a space instance of a Business Query Agent template, with responsibilities distinct from the governance agent. Check its template, model, and query tools.
- Ask about a governed table's purpose; verify its business name, description, and ownership.
- Ask which standard applies to a field; verify the standard and actual association.
- Ask where an object comes from; verify the lineage path.
- Ask a question unsupported by the space; verify that missing evidence is stated.
If answers contain plausible definitions without evidence, correct context or tools and repeat validation.
Choose a publishing format
| Format | Consumer | Capability |
|---|---|---|
| Space Agent | People or integrated AI applications | Natural-language queries and explanations using authorized context |
| Space MCP | External AI agents | Standard tools for assets, semantics, lineage, and knowledge |
Both use the space's governance results, but verify invocation and authorization separately. Space MCP serves external consumers; an external MCP configured in DIG provides tools to DIG.
In Publish, review the available settings, intended consumers, and scope. Use actual client connection details from the page rather than screenshot addresses.
Verify after publication
- Authorized consumers can access the capability.
- Responses remain within permitted space context.
- Answers trace to actual assets or reviewed files.
- Unauthorized and unsupported questions do not produce fabricated certainty.
- Recheck context quality and answers after asset, rule, or file changes.
Troubleshooting
The report still shows gaps. Check whether affected objects were in scope, proposals were approved, and DAM results actually changed. Execution success does not complete every dimension.
Answers are irrelevant. Check query template, selected context, and files. Ensure the instance is a query agent for the correct space.
External agents cannot use MCP. Check publication, access scope, client details, and network reachability. Use the space's published endpoint.