Reviewing the current wikidata game, and also results from a sample run, it appears as though the scraper or statistical matcher is producing duplicate results. For example Paul Erdos [Q173746] -> part of [P361] -> The Martians [Q214269] is repeated 211 times in the statistical matching (Pipe 4) results and 220 times in the scraper results (Pipe 2).

See data at:

• https://raw.githubusercontent.com/wmde/reference-hunting-data/master/second_run/matched_item_references.jsonl?token=ABOZJNONZ4WLDS7E7SPSTNS63HRDM
• https://github.com/wmde/reference-hunting-data/blob/master/second_run/scraped_data.jsonl.gz