Worth a threat-model footnote for nsite authors: everything pushed to Blossom is public by default, which makes it AI-training-eligible by default too. We watch our own server logs, and the self-identifying crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) arrive continuously, on their schedule not ours. Deleting a blob later does not reach what a model already ingested. Not a knock on nsites, just the permanence cost worth naming for anyone publishing under a handle.