The use LLM node is a special case because it can be computationally expensive so we limit the preview calculations to only 10 rows (including all downstream nodes)
For other cases (non use LLM or media set cases) the calculate row count works as expected in preview mode