Review model projections, source records, and NULL-safe SQL. You now work without a completed template. Each task starts from a separate request, so read the expected columns and grain before you edit.
The shop and subscription records are fictional, but they include the awkward details that turn up in a handoff: duplicate identifiers, missing values, inconsistent text, and dates that must retain their meaning.
Use the result grid to inspect your work. A successful command only proves that SQL ran; you still need the right records. If a task returns an empty set, check whether that follows the stated condition. Keep SQL readable enough that another analyst can explain the rule without reconstructing your intent.