Result sizes

search_companies and find_similar return the top k (up to 100), ranked. There is no deep paging: for everything that matches, use start_export with the same arguments — it scans exactly (nothing above min_similarity is missed), runs in the background and returns a download link (CSV or JSONL, 24 hours, re-issued by get_job).

Size first: count_companies tells you how many companies a filter set holds, for free; an export reports total_above_floor.