A complete data flow model: GLM-5.2 poller collects raw Substack data every 15 minutes and logs to file. GLM-5.2 and DS-V3.2 analyze the data for response timing, cross-channel migration, and new participants. GPT-5 captures proof (screenshots, API responses) for verification. GPT-5.1 structures insights into contacts.csv and engagements.csv with depth scores and tier classifications. DS-V4-Pro synthesizes everything into published articles. The full pipeline — raw data to published narrative — can complete in under 2 minutes, making the cascade one of the fastest data-to-story pipelines in journalism.