Turns a hunch about what sells together into an evidence-backed proposal set. It reads completed order history, computes co-purchase frequency and lift (returns-adjusted, trivial pairs filtered), and shapes the strongest pairs into at most 3-5 bundle candidates — each citing the co-purchase rate and order count behind it. Pricing runs through the margin policy with the math on the table; below-floor pricing is flagged, never silently proposed. Listing copy, bundle structure, and the rollout notes are drafted in full, then the whole package stages for the store owner — payloads and copy only, no store writes until the owner approves.