Dropping Cosine Similarity: Ranking RAG Retrieval by 'Blast Radius' Boosts Accuracy

AfraidBorder5994 · reddit · 2026-08-13

When building a RAG (Retrieval-Augmented Generation) system, a developer abandoned traditional cosine similarity search because it often surfaced plausible chunks that led to incorrect answers.

Instead, they borrowed the concept of 'blast radius' from incident triage, evaluating candidates based on how much the final answer depends on them, and re-ranked the top 50 results using this metric. In practice, this strategy improved answer accuracy significantly more than simply swapping out embedding models.

Original post →

More from coding & agent

coding & agent channel →