Clusters changed for assigned text after reducing outliers

Hi,

I was calling

topic_model = BERTopic()
topics, probs = topic_model.fit_transform(docs)
new_topics = topic_model.reduce_outliers(docs, topics)

to reduce outliers according this page. But I found some of the assigned clusters (before reducing outliers) of the text are changed to a different cluster after reducing outliers.

Shouldn't the already assigned clusters will not change after reducing outliers according to "You can pass outlier and non-outlier documents together since it will only try to reduce outlier documents and label them to a non-outlier topic."?

Is this an expected phenomenon due to that I misunderstand anywhere or there is something wrong inside it?

Thank you for helping me out!

Reproduce the reported sequence using BERTopic.fit_transform followed by reduce_outliers(docs, topics), then compare the original and returned topic assignments for non-outlier documents. Read the reduce_outliers behavior and the referenced documentation first; done means determining whether non-outlier labels change and providing a focused explanation or reproducible discrepancy for maintainers.