In Lucene, scores can differ between partitions even if the document fields match exactly. The reason is that Lucene’s scoring depends not only on term frequency (TF) and inverse document frequency (IDF) within each partition but also on how big and diverse the index is.
For example:
IDF differences → If “John” or “Smith” appears more frequently in Partition 0 than in Partition 1, Lucene will assign a slightly different IDF value, which changes the score.
Normalization → Field length normalization and document norms may differ across indexes, which affects scoring.
Independent statistics → Each partition is scored in isolation, so two identical matches won’t necessarily get the same numeric score.
If you want consistent scores across partitions, you’d need to:
Use a MultiSearcher / IndexReader that merges statistics across indexes.
Or normalize the scores manually after retrieval if you’re always querying partitions separately.
The important part is: scores are relative within a single index, not absolute across different ones. As long as the top matches are identical, the small score difference doesn’t usually matter.
By the way, I recently wrote about how Lucene scoring concepts feel similar to how different worlds are weighted in Anime Witcher — worth checking if you like technical + fantasy blends.