Best Sellers pages are a useful starting point because the ranking itself is the data, but they update often enough that one scrape is only a snapshot. Storing each run with a timestamp and comparing position changes makes the dataset much more actionable. Locale matters too, since category rankings can differ by marketplace. I work with AllyHub, and our Amazon Niche Finder is another way to generate candidates for manual validation: https://allyhub.com/tools/amazon-niche-finder. What collection interval has worked best without creating noisy data?