Tip: Check fallback_used before trusting a RagResult

Recommendation

After calling RagPipeline.run() or run_and_generate(), check RagResult.fallback_used before presenting the result. When fallback_used is True, no grounding context was found in the corpus, and the pipeline sent a fallback prompt that explicitly tells the LLM not to invent APIs, workflow names, or CLI commands.

Why it matters: A confident-looking augmented_prompt and a non-zero confidence score can still come from a fallback path — the only reliable signal is this field.

Tradeoff

Reading fallback_used adds a conditional branch to your response handling. If you skip it, you lose the ability to distinguish a well-grounded answer from an honest "I don't know" — and those two cases often need different UI treatment or downstream logging.

Example

answer, result = pipeline.run_and_generate(query, provider="openai")

if result.fallback_used:
    # No corpus match — surface a warning or lower-confidence indicator
    warn_user("Answer is not grounded in the corpus.")
else:
    # Safe to display citations from result.citation
    display(answer, result.citation)

Source files

• src/attune_rag/pipeline.py
• src/attune_rag/__init__.py

Tags: pipeline, orchestration, rag, result