Your embedding model outputs unnormalized vectors whose magnitudes vary widely. A teammate wants to switch the index from cosine similarity to dot product because it is cheaper to compute. What is the consequence?

1. A No change in ranking, since cosine similarity and dot product always produce identical orderings.
2. B Ranking improves automatically, since dot product ignores vector magnitude entirely in scoring.
3. C Queries fail at runtime, since vector indexes reject dot product on vectors that are unnormalized.
4. D Ranking changes, since dot product favors large-magnitude vectors that cosine normalizes away.