vector_similarity_search introduces risk embedding

Source: Pattern·Target: Failure Mode

Summary

Vector indexes must be rebuilt when embedding model versions change; stale vectors from old models mixed with new produce retrieval quality degradation.

Evidence

·embedding_drift lists vector_similarity_search in affects_patterns; old and new model vectors occupy incompatible spaces (failure_modes/embedding_drift.yaml)
·vector_similarity_search operational_considerations: keep the embedding model version fixed; different versions require full reindex (patterns/vector_similarity_search.yaml)

Evidence grounding

Grounded, 2 supporting items

Changing the embedding model without full reindex silently corrupts similarity results.