Nurture sequences rot quietly: a step that worked at launch slowly dies and nobody notices for a quarter. This recipe audits every step of the sequence against its own baseline, flags underperformers with the exact rule and numbers behind each flag, and pairs each with a diagnosed cause — subject fatigue, CTA mismatch, length, timing, or message drift. It then drafts a refreshed variant per flagged step (same job in the arc, new angle targeting the diagnosis), brand-voice passed, and stages the whole update plan as a before/after package. Applying changes stays a human decision: refreshed copy lands as staged drafts, and live steps never change silently.