Recall, accuracy, and query difficulty

Recall, accuracy, and query difficulty explain how we measure whether approximate neighbors are good enough, why some queries are harder than others, and what adaptive exploration can recover under stress.
Created: Updated: 2 min read

These are the topics in this section, each on its own page with a stable path you can bookmark or share.

What topics are covered in this section?

Recall, accuracy, and query difficulty explain how we measure whether approximate neighbors are good enough, why some queries are harder than others, and what adaptive exploration and certification try to fix.

How is ANN quality measured?

Recall@k and recall@1 compare retrieved IDs to ground-truth neighbors. Precision versus recall in ANN contexts is easy to confuse with information-retrieval usage – these pages keep the definitions straight. Query difficulty, poor entry regions, missing bridges, under-exploration, and named search failure modes explain non-uniform errors. Adaptive ef and query-adaptive exploration try to spend more effort only where needed. Accuracy certification and graph spanners point toward per-query trust beyond average recall.

Without this vocabulary, tuning becomes chasing a single headline percentage.

How do these ideas guide Weaviate operations?

Weaviate exposes distance and certainty metadata and lets you raise ef or strengthen build parameters when recall is insufficient – always validated on representative queries, including filtered and hard cases. Dynamic ef ties exploration to requested limits. Certification as a shipped product feature remains frontier research; empirically high recall on your golden set is the practical bar. Failure-mode hubs continue the debugging story when metrics move.

Accuracy language also anchors Part VIII benchmarking methodology.

Open recall@k, query difficulty, and under-exploration pages, then Part III’s variance chapter and Part VIII’s methodology. For formal guarantees, see accuracy-certified search in the frontier glossary.

This section is about measuring and explaining ANN correctness. Next, read “What is recall@k?” or Part VIII’s benchmarking chapter, then adaptive-ef terms when easy and hard queries need different budgets.