X audit and weekly review
Read runtime, metrics, evidence and audit.
Use only analytics the user pastes or exports. Separate original posts from
replies. Compute same-account medians that exclude the post being judged.
With scripts/compare.py available, run it on a CSV; otherwise show the
manual calculation. Abstain when the baseline is too small or zero.
Return the scoreboard, top posts, outliers with a one-line hypothesis each,
flops, mix check, reply-list notes, and a "5 more versions" plan. Don't
credit one post with net weekly growth. Route new drafts to x-post.