Margin sends your prose to Anthropic's API for analysis. Anthropic does not train on API traffic. Margin stores nothing: no database, no logs of your text. Close the tab and it's gone.
Three parallel reads: a structural read (what each paragraph is doing), a pacing map (tension, density, and valence), and a voice check (statistics on your baseline, with a short note only where the numbers cross a threshold). None of this rewrites your work.