This formula transforms from score to cosine similarity, however internally ES works with score.

I believe this should be the inverse, e.g. float adjustedMinSimilarityScore = (float)((minRelevance + 1) / 2)

By the way, the formula (2 * minRelevance - 1) should be used in line 253, replacing hit.Score with the actual cosine similarity. The value returned is incorrect and doesn't match the similarity returned by other memory connectors.

As per https://www.elastic.co/guide/en/elasticsearch/reference/current/knn-search.html#knn-similarity-search
default similarity metric is cosine and I didn't see something defines another similarity metric in the plugin code.
From my tests I found this score adjustment for cosine returns correct results.
please let me know if I'm missing something here or if you have another proposed tests to include..
A similarity value. This value determines the similarity metric used to score documents based on similarity between the query and document vector. For a list of available metrics, see the [similarity](https://www.elastic.co/guide/en/elasticsearch/reference/current/dense-vector.html#dense-vector-similarity) parameter documentation. **The similarity setting defaults to cosine**.

qd.k(limit) already sets a limit, isn't this redundant? should we remove the limit param in qd.k(limit)?

kNN search adds k matching documents to the search. So, if you set k=20, then it finds 20 matches. Then size takes the top scoring documents from these and returns them.
From my tests in real dataset. it was always returning 10 regardless of the k(limit) value. I think this is the default value of size parameter