Measurement
Count from the catalog, not from the page
A screen reported seven products selling below cost. The catalog held twenty-six. The bug was not in the calculation, it was in what the count was computed over.
A count of what, exactly
The figure was derived from the rows the screen had fetched, which was one page of results. It was an accurate count of a page and a badly wrong count of a business.
This class of defect is easy to ship because nothing errors. The number renders, it looks plausible, and it is only caught when somebody compares it against a query written independently.
Summaries belong to the model
Any headline figure should come from the model, not from whatever a screen has loaded. If a screen can work out a total by itself, eventually it will, and it will be wrong in exactly this way.
The rule we now apply is simple: screens render aggregates, they do not derive them.
Want this against your own numbers?
A scoping call works out which measures are worth baselining and whether your systems can actually be read.
Let’s chat