Source provenance
Keep source identity and retrieval time attached to every normalized result.
- Provider resultURL + native text
- Provenanceprovider + sourceUrl + retrievedAt
- Your applicationKeep attribution when rendering
The attempt trail describes the call. Per-result provenance describes the source, its provider, and when it was retrieved.
On this page
Provenance fields
| Field | Meaning |
|---|---|
provider | Adapter identity for this result. |
sourceUrl | The source link assigned by the adapter. |
retrievedAt | SDK retrieval time in ISO 8601, not source publication time. |
requestId? | Provider request identifier when the adapter has one. |
Retain provenance through your pipeline
When storing sources or passing them to an answer-generating step, keep their provenance beside the text. Preserve the original link even when your application derives a citation label. Retrieval time is useful for traceability but does not prove a page is current or substantiate a factual claim.
Keep native relevance in context
Exa and Tavily may provide numeric scores. Context's string relevance is not converted into a fabricated number, and TinyFish position is not a score. A missing value should stay absent. Do not compare scores across providers without a separately validated calibration method.