How to read the cards, the status badges and the four groups is explained on the Comparisons overview; every statistical term links to the Statistical glossary.
Publications are grouped by Substack's stored fully-AI flag on their recent posts: The flag fires only at 100% AI, so "never flagged" publications may still use AI below that line. "Before Pangram" and "after Pangram" mean before and after 21 July 2026, the day Substack's detection went live and the earliest date it will scan. Because publications entered the sample through a random post, prolific ones are over-represented; the bars and medians here are by 1/(posts after Pangram), and tests are on the unweighted rows.
Posting rates are posts per week in matched 105-day windows either side of Pangram's arrival; the before-window figures cover publications that posted at least once before it. Age is days from the publication's creation to that day.
A finer grouping from fraction-level labels: These buckets fill in as the extra scans complete; the tests here are exploratory and outside the correction family.