Archive
RAG & Retrieval

The Tiny Triple That Beats Fine-Tuning

Archive
Expert
300pts0 solves
Anthropic published a 2024 blog post introducing a pre-indexing step that prepends an LLM-generated 50-100 token context snippet to every chunk — dropping retrieval failure rates by around 49%. Name the technique (two-word phrase, as in the title of the post). Flag format: CONGRESS{name}. Example: CONGRESS{chunk augmentation}.
Show hint
First word describes the kind of addition; second is the whole point of RAG.

Archive — no submissions accepted

This challenge is preserved for reference. Play live challenges at /challenges.