AI Agents / Retrieval and evidence for agents
AI Agents: Document identity
Assign stable document identifiers with revision and source location.
Environment and inputs: Retrieval-augmented agent workflows with lexical search, embeddings and optional reranking.
When to use it
Separate the question of finding the right evidence from the question of generating a good answer. Save the retrieved passage identifiers so a wrong result can be traced to ingestion, selection or interpretation rather than guessed from the final prose.
Practical example
Illustrative chunk metadata. Retain this identity through ranking and citation rendering.
{
"document_id": "runbook-17",
"revision": "r3",
"section": "Recovery checks",
"chunk_id": "runbook-17-r3-04",
"source_location": "page 4"
}Check the result
Reimport an updated document and verify old citations still resolve to their recorded revision.
Save and compare results
Save the test input, configuration and result. Fix the failing boundary, repeat the same test and compare it with the saved working configuration.
Scope and limits
These exercises have not been tested on your systems. Check feature support and configuration names against the installed release and linked documentation.