What changed in the platform

A write-off document from the store export was landing in sales history alongside receipts. Demand is now assembled from stock movements with an explicit operation type, and write-offs are excluded.

Why: The error was quiet: the order looked justified while the network kept rebuying what it had just discarded. After the fix, ordering in two stores fell 16% and 7% at the same availability.

Items that failed to reach a minimum number of sales in the period never entered the calculation at all. The threshold is gone: goods on the shelf are computed even when they sell one at a time.

Why: In a pharmacy chain, cosmetics and orthopaedics silently vanished from the report — whole groups where decisions were made with no data at all. Removing the threshold nearly doubled the number of items covered.

Supplier files in different formats are merged into a single table: item, price, availability, lead time. Rows that failed to parse are shown as their own list.

Why: A distributor holds dozens of price lists, compared by eye in a spreadsheet. The point is less the automated comparison than the fact that an unparsed row no longer disappears in silence.

A client now has a separate key to its own data marts, stock and orders. The network owner issues and revokes it, and it grants no write access of any kind.

Why: A network's data must come out without a “we will export it on request” conversation. Writing to the accounting system stayed with whoever owns it: we deliberately expose no write API.

Previously a supplier received one order covering its whole range. A narrow group — a single product line, say — can now go to its own channel, with its own schedule and recipient.

Why: Different people at one supplier handle different orders. A single shared file meant one of them was not reading it, and a short-delivery complaint ran into “I never received it”.

If the data snapshot holds no rows for a warehouse, the send stops: the owner gets an explanation instead of a file, and the supplier gets nothing at all.

Why: An order once went out as an empty file: nothing was delivered, yet the system considered the send successful. An empty file looks like normal operation — the worst kind of failure.

Three hours after sending, the system re-checks: the file was not empty, the recipient exists, a reply is there. The result shows in the order history, not only in a log.

Why: Some channels had been “working” for months while the link to the supplier's chat was dead. A silent failure is worse than a miscalculation: a miscalculation is visible, a send that never happened is not.

A supplier answers right in the chat: accepted, partial, out of stock, plus a free-text note. The reply lands in the order history and is visible to the owner in the web app.

Why: Confirmations lived in chat threads, so a discrepancy at receiving had nothing to argue against. A supplier's promise now carries a time and an author.