Read the evidence stream
The evidence stream is the right-hand panel in Investigate. Every row is a single statement Atlas used to answer your query, plus enough provenance to tell you whether to believe it.
Anatomy of a row
Section titled “Anatomy of a row”| Element | Meaning |
|---|---|
| Claim (top line) | The fact Atlas is asserting — e.g. “BAPI_PO_CREATE1 is deprecated as of S/4HANA 2023” |
| Tier chip | C2 green (public stable) / C1 amber (system-internal) / C0 red (private) — mirrored from the source contract |
| Confidence label | 0.98 tier-1 / 0.70 tier-2 / 0.55 tier-3 / 0.40 tier-4. See Evidence and confidence. |
| Source row | Source name + fetch timestamp, e.g. Simplification Item Catalog · 2026-04-16 02:14Z |
| Expand | Opens the full provenance chain: claim → section → document → URL + content hash |
When to trust a row
Section titled “When to trust a row”- Both tier chip and confidence label are green — emit freely.
- Tier C2 but confidence 0.70 — first-party text, not structured. Read the source before citing.
- Tier amber or confidence ≤ 0.55 — review explicitly. Atlas never auto-emits these.
- Fetched > 72h ago — the Corpus tab of the active case will tell you the source is stale. Treat findings downstream of it as provisional.
Chasing provenance
Section titled “Chasing provenance”Click Expand on any row to see the full supportedByStatement chain:
- the section of the document (section anchor + label)
- the document itself (URL + content hash + fetched-at)
- the concrete passage the extractor lifted the claim from
The content hash is stable: copy it, and any future exported evidence trail for the same fact will show the same hash. That is what makes the evidence independently verifiable.
Carrying evidence into a review
Section titled “Carrying evidence into a review”Every row has a stable content hash; paste it into a review comment and a teammate opens the same row in Atlas and sees the same chain. That is the minimum unit of reviewable evidence in Atlas.