Describe what's wrong

I attempted to make a table with the usearch index and query it via the cosineDistance function. This is as specified in the docs on earch indexes which were merged in #53447 a couple days ago.

Querying the index only resulted in std::exception. Code: 1001, type: std::runtime_error, e.what() = Magic header mismatch - the file isn't an index (version 23.8.1.2501 (official build)) rather than resulting rows.

I am working on this as a part of an effort to switch off of Qdrant for arguflow.ai

The easiest way to reproduce this is to use the jupyter notebook on this by arguflow.ai - the README and first cell of the notebook contain steps to repro.

Does it reproduce on recent release?

Yes, 23.8.1.2501

I enabled crash reporting, so hopefully a Sentry report went through when this failed.

How to reproduce

1. Use 23.8.1.2501
2. Make a table with a usearch index like INDEX usearch_embedding_vector embedding_vector TYPE usearch('cosineDistance') GRANULARITY 100
3. Attempt to query the table with cosineDistance like SELECT point_id, cosineDistance(embedding_vector, {first_vector}) AS score FROM qdrant_benchmark_10 ORDER BY score ASC LIMIT 10
4. Sample data is in the jupyter notebook on this by arguflow.ai

Expected behavior

Rows should get returned

Error message and/or stacktrace

std::exception. Code: 1001, type: std::runtime_error, e.what() = Magic header mismatch - the file isn't an index (version 23.8.1.2501 (official build))

Additional context

I am hopeful that usearch is more accurate than annoy . I'd be happy to contribute if someone adds a rough outline on what it would take to fix this. Fwiw, the same error occurs with L2distance as shown in this notebook.