Example: Custom Retrieval

This example compares dense-only retrieval against hybrid retrieval on the same query, and shows how the resulting RetrievalResult objects differ.

from ragaxis.retrieval import DenseRetriever, HybridRetriever

dense = DenseRetriever(store=store)
hybrid = HybridRetriever(store=store, rrf_k=60)

query = "How is chunk provenance tracked?"

dense_result = dense.query(query)
hybrid_result = hybrid.query(query)

print("dense:", [c.position for c in dense_result.chunks])
print("hybrid:", [c.position for c in hybrid_result.chunks])

Hybrid retrieval combines dense and sparse rankings using Reciprocal Rank Fusion. The rrf_k parameter controls how much weight lower-ranked results retain in the fused list; the default of 60 matches the value used throughout RAG Axis documentation and tests.

Both dense_result and hybrid_result carry a confidence signal. If either retriever returns no candidates above the configured threshold, the result includes ConfidenceUnknown rather than an empty list with no explanation, per invariant I3.