max_cluster_size effecting the GlobalSearch and DynamicGlobalSearch

[gutihernandez]

Context

During global and dynamic-global searches, I've observed that communities whose total unique entity count exceeds max_cluster_size are missing from the query results. This appears to be a filtering issue that occurs at query time, not during the indexing phase.

Indexing Phase

• The hierarchical Leiden algorithm creates sub-communities when a community exceeds max_cluster_size
• However, at the end of indexing, we still end up with some communities whose cluster_size (total unique related entities) is larger than max_cluster_size
• These larger communities are successfully created and stored

Query Phase (Global/Dynamic-Global Search)

• These larger communities are mysteriously missing at the time of Querying Phase…
• Currently forced to rely only on sub-communities
• Unable to locate where this filtering is happening in the code

What I Want to Achieve

• Include ALL communities in global and dynamic-global search results, regardless of their size
• Not be limited to only sub-communities when the parent community exceeds max_cluster_size

Specific Questions

1. Where in the codebase is this filtering happening during query time?

• Specifically during global and dynamic-global searches
• Need to understand why larger communities are filtered out even though they exist after indexing

2. Is this intended behavior?

• If intentional, what's the rationale for excluding larger communities during queries?
• If not intentional, how can we modify the query behavior to include these communities?

Investigation Status

• Confirmed that large communities exist after indexing
• Confirmed that filtering happens during query time, not indexing
• Unable to locate the specific code where this filtering occurs (for the querying phase)
• Specifically looking at global and dynamic-global search implementation

Any help in locating where this size-based filtering is happening during the query phase would be greatly appreciated. Has anyone encountered similar behaviour or knows where in the codebase this filtering might be implemented?