Most teams book more than sales through the invoice system. Stock that expired, broke, went out as a sample, went to a marketing giveaway, or was simply corrected off the books is often recorded as a sale to a house account at zero price. It is a practical habit, and your ERP is happy with it.
The problem is what the forecast sees. Every one of those units looks like a customer buying something. Write 20 units off a slow mover and the engine reads a 20 unit spike, then orders to cover it.
Go to Settings → Data management to say which accounts count.
The list is sorted by volume, not by name, so the accounts moving the most stock are on the first screen. House accounts usually stand out two ways: a large number of units, and little or no revenue against them.
Check the revenue column before you tick anything. An account with real money against it is usually a real sale being handled oddly rather than a write-off, and excluding it would hide genuine demand.
Matching ignores case and spacing. If your ERP returns the same account as both HRFI and hrfi, one entry covers both.
A change applies automatically to days synced from then on. Past history keeps its old split until you rebuild it.
Use Rebuild demand history after changing anything. It works through one month at a time and shows progress, so it is safe to leave running and safe to start again if you close the tab. Until you rebuild, the page warns you that the forecast still reflects the old numbers.
The rebuild always covers the last three years, whichever window the list happens to be showing. That is deliberate: the forecast reads a full year of demand for its rate, and reaches further back still for seasonality, so rebuilding only what is on screen would leave stale figures inside the range the engine actually uses.
Anyone with access to Settings can read the list. Only a company admin can change what counts, because it changes what the forecast believes and therefore what Stokk orders.