How chain roll-up avoids forked definitions

Once you pass a dozen stores, HQ’s most familiar pain is this: add the numbers each store filed, and they match neither any store’s own number nor finance. The fork usually happens at three nodes on the tree—store code, item code, the sales sentence—not at chart color. Roll-up wants the same branches, not each store’s private sheet glued into a forest.

Freeze three master-data books first: stores (whether closed, franchise, and warehouse-only count), SKUs (repack, free goods, one item many names), and staff or associates if commission rolls up. Missing those three, visualization adds error by store; see Without master data, visualization lies. The sales sentence must be unique: returns in or out, stored value, delivery fee, internal purchase. Write it in an HQ document. Store managers only see a slice. They cannot change the formula locally. After one definition, HQ watching outlier stores means something; see After one definition, weekly meetings argue less and After stores roll up, how HQ watches outlier stores.

When POS, inventory, and delivery platforms each publish a set, the fork multiplies again. Standardize APIs and order numbers. Do not let each store reconcile itself. See When every system calculates its own numbers, what breaks. If quality is poor, wash these three codes and the sales sentence first, then talk more charts; see Data quality is poor. Build the dashboard first, or clean first. If franchise may use a “local promo definition,” list it separately on the tree. Do not add it to owned stores on one total and pretend they are one. Closed, remodel-closed, and warehouse-only sites need a status. Silent disappear-and-return from the total makes year-over-year lie. The business-day cut (do the two hours after midnight belong to yesterday or today) is also HQ’s call. Mix night stores and morning markets on one calendar day and the fork looks like a system defect.

HQ has one yardstick. Slices can have permissions

What a single store sees should be that store cut by the same formula—not the store calculating locally and uploading. Upload mode forks again in month two. When you spot-check, do not only tie the total. Pull three receipts from three stores, from POS into HQ detail. A total that happens to match while receipts do not is a more dangerous fake-clean. When DaXi builds chain roll-up, we first stop private formulas in store-manager weekly reports and replace them with an HQ slice plus an exception list. To name the column that will not add up now, submit store count and POS system on the Data service page. When systems keep growing, APIs first, charts later; see More systems connecting: standardize APIs or draw charts first.

Do not give a new store a “local report template” first. Issue codes, grant slice permission, teach the exception list. Once a template goes out, the fork grows from week-one sales.