We do not have a continuous system, the data is a mix between our own crawled generations and the dataset published by Dave Caruso (https://github.com/paperdave). With our crawler, we were able to get about 100k generations per day.