Flow
Which numbers can raise an order
1Sales need
Applies to every item the store may stock.
2Display minimum
Now only for required items with no set floor.
3Set floor
Required items only.
4Group number
Counts only when there is no skip reason.
5Allowed status
Raises nothing on its own; just puts the item in the running.
Allowed items in the ordering system were getting orders: a yes on a spreadsheet had become a reason to buy. One small rule was applied to items it was never meant for.
The four statuses
Each product and store pair has a status. Required means the store must carry it. Trial means it's on the shelf with a cap. Blocked means it's left out. Allowed means the store may carry it, and sales decide how much. Allowed should never add any quantity by itself. It just says the item is in the running.
Where the extra order came from
The system picks a target level by taking the largest of a few numbers. One is how much sales say the store needs. One is a display minimum, the least amount that makes a shelf look full. One is a set floor that only required items have. The display minimum was used for every row the store could carry. So an allowed item with no stock and no sales still got a target, and anything over what's on hand turns into a suggested order.
When you take the biggest of several numbers, each one looks harmless. The thing to check is which items each number was meant for.
Why a big import is the worst time
The import for a new vendor reads a sheet with a yes or no per store and writes allowed or blocked rows. That makes a lot of allowed items at once, and I'd guess most had never sold in those stores. Those are the rows the display minimum inflated. The import's own checklist even counted a report full of order lines as success, which is the wrong thing to reward.
Only required items get a floor now
The display minimum now only fills in for a required item that has no set floor. Everything else follows sales. Trial items seem to have lost that floor too, from my reading of the change, and no test covers that yet.
A new test sets up an allowed item with no stock and no sales. It checks that the target and the order are both zero and that the row is still marked for a person to review. The first version had the wrong expected display minimum, and a quick follow-up fixed it.
The same habit in group planning
Another planner handles groups of items a buyer can swap for each other. For each group it gives a number and, sometimes, a reason to ignore that number, like "no stock and no sales." Three readers only looked at the number, so a group the planner had said to skip still looked like work. All three now use the same check.
A group counts as ready only if it has no skip reason and a positive number.
The test confirms a skipped group shows no demand and adds nothing to either count, even though its raw number is still there.
I only have the code and tests to go on. I can't tell from them whether any carts were built from these phantom lines.