Hi, I am currently having a issue with thresholding while using ColBERT as a reranker. I notice that according to the scoring logic here (https://github.com/qdrant/qdrant/blob/master/lib/segment/src/vector_storage/query_scorer/mod.rs#L34-L54), the scores should always be in range [0, min(len(query_vectors), (doc_vectors))].

However, in practice, I always get a score with at least 15, no matter how short my query string would be. Is this due to a padding (or something similar) logic in the implementation? I tried to use the semantic search site but could not found related code.

Same issue could be found here: qdrant/fastembed#383 and stanford-futuredata/ColBERT#374