Mara needs a support triage recommendation from a replayed January order feed. Reconcile received rows with unique transactions, report refund shares at channel grain, and finish with the eligible channel that has the highest observed share. The checked final query is the recommendation.
Mara, Lumen's operations lead, received two versions of a January order report. One used every imported row; the other counted order IDs. She needs to understand the difference before using either report to prioritize support work.
The source in practice.review_feed contains replayed rows with the same order_id and unchanged details. Your first task is to measure that duplication. Then build a unique order relation and use it consistently. A refund share needs refunded orders as its numerator and all unique orders in the same channel as its denominator.
The final query identifies the highest observed refund share among channels with at least five unique orders. Return its counts too. This is a support triage comparison within the supplied feed, not evidence that a channel caused refunds. A different mix of customers or order dates could explain the difference.