Skip to content

Run your first investigation

This tutorial exists so a skeptical reader can verify Atlas works end-to-end in about ten minutes. You will open the Studio, ask a natural-language question about an S/4HANA conversion, and read an answer that Atlas backs with structured evidence from the knowledge graph. No setup, no CLI — just the browser.

Anatomy of /investigate 1 · COMPOSER Natural-language prompt + intent "What blocks ECC 6.0 → S/4HANA conversion readiness?" POST /api/atlas/resolve 2 · RESOLVE RESULT Ranked blockers with typed answer rows BAPI_PO_CREATE1 tier 1 · 0.98 BAPI_SALESORDER_CREATE tier 1 · 0.98 I_LegacySalesOrder tier 3 · 0.55 Custom ZCL_ORDER_BOOK tier 2 · 0.70 Filters above the list narrow by tier · module · artifact type. Click any row to open its drawer. 3 · EVIDENCE STREAM Per answer row, its sources Simplification Item Catalog SI-4321 · 2026-04-16 api.sap.com API_PURCHASE_ORDER_SRV Released-CDS list S/4HANA 2023 FPS02 Customer extract my-prod · ZPROG_ORDER content hash + fetched-at on every row Composer asks. Resolve returns typed rows. Evidence stream carries the sources, one-to-one.
Anatomy of the Investigate view — composer, resolve result, evidence stream.

An answer to “what blocks ECC 6.0 → S/4HANA conversion readiness?” that carries its sources with it: which deprecated BAPIs are involved, which migration guides flagged them, and which modules in your corpus are implicated.

The first screen after sign-in is the home dashboard. It shows four KPI tiles — active cases, open blockers, corpus freshness, estimate accuracy — plus a “create your first case” shortcut. Keep this tab open; come back here between tasks.

Click Investigate in the left nav (or press G then I). The view has three zones: a query composer at the top, a resolve result panel in the middle, and an evidence stream on the right.

The composer ships with canned prompts grouped by shape — readiness, risk, deprecation, module-impact. Pick “What blocks ECC 6.0 → S/4HANA conversion readiness?” from the readiness group and press Ask.

Atlas calls POST /api/atlas/resolve, translates the natural-language intent into a SPARQL traversal, and returns two things:

  • a ranked list of blockers in the middle panel — deprecations, missing released APIs, custom code hotspots
  • an evidence stream on the right — every blocker linked to the source document, migration guide, or catalog entry that Atlas read

Click any blocker in the middle panel. A drawer slides in from the right with:

  • the canonical graph node (e.g. BAPI_PO_CREATE1)
  • its replacement, with a confidence tier badge
  • the source documents that contributed — Simplification Item Catalog, api.sap.com, SAP help portal pages — each with a last-fetched timestamp

This is the same provenance Atlas carries everywhere: claim on top, sourcedFrom chain below, tier label on the side. See Evidence and confidence.

Click Pin to a case in the drawer header. A modal asks which case should own this finding; pick “Create new case” and give it a one-line title. The drawer confirms, the case gets the evidence attached to its timeline, and you land on the new case’s object view.

From here every investigation you run is tracked against that case — the classic Atlas flow: ask, anchor, iterate.