The silent 500-row limit

A finance department, a treasury tool. A consolidated total off by several hundred thousand euros. Everything is green, and there's no error message.

A wrong figure that looks right

A gap of several hundred thousand euros should jump out at you. Well, no.

The screen is clean. The totals show up. Nothing blinks, nothing warns.

That's what makes this kind of error dangerous. An absurd figure, you correct right away. A plausible figure, you believe.

The cause: 500 rows, not one more

The platform the tool ran on limited its queries to 500 rows. Silently. Beyond that, it read nothing more, and didn't say so.

The scope had more rows. So the tool was adding up part of the data, believing it was adding up everything.

A total that adds up 500 rows out of 700 looks normal. It shows up like the others. It's just incomplete. And nothing on the screen lets you guess it.

The most troubling part is that the tool made no calculation error. Every addition was right. It was the list of rows to add up that was cut short.

Ten minutes to fix, the rest to be sure

To be honest, the technical fix took ten minutes. The real work was elsewhere. It comes down to three steps.

Reproduce the gap to the cent. Before saying "it's fixed", you have to prove that the cause you found explains the whole gap. Not part of it. All of it. As long as the gap isn't reproduced to the cent, a second cause may be hiding behind the first. To the cent isn't perfectionism. A gap that matches exactly proves the cause is the right one, and the only one. A gap that almost matches means there's something left to find.

Audit every screen exposed to the same limit. If one screen fell into the trap, others could too. Every screen exposed to the same limit was reviewed.

Add a safeguard. The tool now flags when the volume approaches the limit. The limit can no longer act silently.

Without these three steps, fixing one screen can leave the same error lying dormant in another. And you find it the day it costs you dearly.

The question we forget to ask

When you receive a new tool, you think to check that it calculates correctly. The formulas, the rounding, the currencies.

You think less often of the simplest question: does the tool see all the data?

A correct calculation on incomplete data gives a wrong result. And that result sets off no alarm. It goes into the dashboards, the summaries and the decisions, with the serious look of an exact figure.

The limit itself isn't the scandal. A platform has limits, that's normal. The problem is a limit that acts without warning.

What this means for you

Your management tools show totals. You read them, you present them, you make decisions with them. Ask them one question before anything else.

• Does the tool see all the data? Not "does it calculate correctly". Does it see everything.
• On a known case, does the total match to the cent? Take a scope whose exact figure you know, and compare.
• What happens when the volume grows? A tool that sees everything today can exceed a limit tomorrow, without warning.
• Who would warn you? If the answer is "nobody", a safeguard is missing.

A gap reproduced to the cent is a gap understood. Until then, the investigation isn't over, even if the new total looks right.

Another silent error, in an import this time: "Import complete", zero rows imported. And what a treasury tool must guarantee above all, one version of the figures: the Excel consolidation that ends in a six-month debate. Diagnostics and troubleshooting of this kind are described on the page Two days to put a figure on the gain, and other projects are in the case studies.

The right question isn't "is the calculation correct?". It's "does the tool see all the data?".