Compose a query in Investigate

The Investigate view translates a natural-language prompt into a typed graph traversal over the Keystone knowledge graph. You never write SPARQL yourself — the backend does — but the shape of the question matters.

1 · Pick an intent

The composer in /investigate shows starter prompts grouped by shape:

• Readiness — what blocks conversion from X to Y?
• Risk — which modules expose clean-core violations?
• Deprecation — which deprecated APIs does my system use?
• Module impact — what changes in SD between release X and Y?

Pick the group that matches your question and either press one of its canned prompts or use one as a starting point.

2 · Scope the prompt

A precise prompt returns precise evidence. Atlas parses three things from your wording:

• Source / target release — name them (e.g. “ECC 6.0 EHP8 → S/4HANA 2023”)
• Modules — list them (SD, MM, FI)
• Shape — the verb carries the intent (blocks vs deprecates vs replaces)

If you leave any out, the default from the active case fills it in.

3 · Ask

Hit Ask. Atlas calls POST /api/atlas/resolve with your prompt and the active case context. Two panels populate:

• the resolve result in the middle — a ranked list of typed answers
• the evidence stream on the right — the sources that back each answer

4 · Refine

The resolve result has a filter strip above it. Common refinements:

• Confidence tier — only show tier 1–2 (hides inferred and community-source facts)
• Module — narrow to one of the case’s modules
• Artifact type — BAPI only, CDS only, etc.

Refining doesn’t re-run the query; it filters the existing result client-side.

5 · Pin the finding

Click any row, open its drawer, and press Pin to case. The finding lands on the case timeline with the full evidence chain. See Read the evidence stream for how to read that chain.