<p>One screen said 16 stalled deals. Another said 19. Both were right.</p>

S

ame business fact, two different numbers, same product, same day.

It wasn't the model. It was architecture.

Some cards were computing totals in the frontend from whatever records happened to be loaded on screen. Other surfaces were asking the backend for the full count. Both were counting correctly. They were counting different populations.

Pagination made it worse. A frontend component could see a single page of results and present that as the total.

The fix was unglamorous: move the metrics that matter to authoritative backend counts, so every surface reads from one source instead of computing its own version of reality. Then regression and end-to-end tests across the dashboard, so drift gets caught automatically instead of by whoever happens to notice.

What I took from it is bigger than the bug.

When people make decisions off your output, a feature being individually correct isn't enough. It also has to be consistent across surfaces, explicit about what it's counting, aware of state, and traceable back to the source. Otherwise the user has to decide which of your numbers to believe — and once they're doing that, they've stopped trusting all of them.

So the question I've started asking, and I'm curious whether others do:

Have you checked whether the same business fact produces the same answer everywhere in your product? Not whether each feature is correct on its own. Whether they agree with each other.

That's turned out to matter far more than I expected.