How many people are in there. And a number that stands up.

Anyone can put a counter on a door. There are beam sensors for a hundred euros, there is analytics in almost every modern camera, and there are spreadsheets with a number somebody wrote down. They all give you a number.

The problem comes the day that number has to leave your building. An inspection asks when an occupancy figure has to be justified. The landlord asks when the rent is calculated on footfall. The advertiser paying for impressions asks. The committee asks when it has to decide how much staff to put on a Saturday. And a judge asks, if something went wrong and how many people were there has to be reconstructed.

At that point 'a system we have counts it' isn't enough. You need to be able to say who has checked that the system counts correctly — and that somebody must not be us.

The Centro Español de Metrología is the body that in Spain decides whether an instrument measures correctly. It is the same one that approves a butcher's scales, a petrol pump or a traffic radar: when you buy a kilo of something, the CEM is behind it being a kilo.

Type approval doesn't mean the system never gets it wrong. It means an independent body measured how much it gets wrong, under what conditions, and put it in writing with its stamp. That is exactly what you need when the number goes into an official file: not a promise from the manufacturer, but the measurement of a third party that doesn't sell the product.

The precision recorded in that type approval is 98 to 99%. It is not a figure of ours, and that is why it can be stated: it is the one measured by whoever granted the approval, and it is in the document.

And watch the scope, because that is where the catch usually is: what is type-approved is people counting done the way the protocol says, with the camera where it has to be and with the field of view it demands. Mounted any other way it counts just as well or as badly, but it is no longer type-approved. That has to be said before installing, not after.

That it is the most precise on the market. We don't know: to claim that we would have had to measure every other one under the same conditions, and we haven't. What we do know, and can be checked, is that it is the only European one type-approved to do this autonomously.

That it is absolutely effective. There is no such thing. A system that looks at images gets things wrong, and the value of this one is precisely that somebody from outside measured how much.

That it works the same however it is mounted. It doesn't: the type approval covers one specific mounting, and if your door doesn't allow that mounting, we tell you before selling you anything.

And that it is good for everything. This counts people. It doesn't identify them, doesn't recognise them and doesn't know who they are — by design, as explained further down.

The difference between counting and counting properly is not losing sight of anyone. A system that detects people in separate frames counts twice whoever stops in the doorway and doesn't count whoever goes past hidden behind somebody else.

Here each person is followed as one continuous trail from the moment they enter the counting area until they leave, so crossing the line twice doesn't add up to two. In a crowd, which is where every counter breaks, that is the only thing holding the number up.

And it discards what isn't a person. A shopping trolley, a pushchair, a box, an umbrella, somebody's shadow at seven in the evening. All of them move and all of them make a cheap counter count too many.

The camera is installed at an overhead angle, looking down. From there what you see of a person is the top of the head and the shoulders — not the face, not the features, nothing that lets you tell one person from another.

That isn't a setting that can be switched off: it is where the camera is mounted. A system that counts from above cannot identify anyone even if it wanted to, and that is the difference between meeting the RGPD by design and meeting it by promise.

It counts people; it doesn't know who they are, doesn't recognise them from one visit to the next and stores nothing that would let anyone do so later. When somebody asks whether this is facial recognition, the answer is that it cannot be.

Current, maximum and average occupancy. Entries and exits per entrance. Total movements. And the figure almost nobody has: where people actually come in, which is almost never where you thought.

The statistics update every minute, so they are useful for deciding now —opening another door, sending somebody to a queue— and not only for next month's report.

And because the criteria don't change, the historical series is worth something. Comparing this November with the last one is a real comparison and not two estimates set side by side.

With most IP cameras on the market. Exactly which ones at your installation depends on your installation: where they are mounted, what angle they have and what can be seen from there. That is checked before anything is promised.

We don't publish a list of compatible brands, because a list of brands tells anyone with a different one that theirs doesn't fit, and usually that isn't true. The useful question isn't what brand you have: it is where that camera is and what can be seen from it.

And if the camera that is there doesn't allow the mounting the type approval demands, we say so. It can still count, but then it isn't a type-approved number — and if what you needed was the paperwork, better to know beforehand.

Before talking about price there is one thing to look at: where the camera watching that entrance is and what can be seen from it. The type approval demands a specific mounting, and at some doors that mounting doesn't fit.

If it fits, this goes in on top of what is already there and the number starts coming out from day one. If it doesn't fit, there are two routes: move the camera —sometimes half an hour, sometimes a building job— or count without type approval, which is fine for running the place but not for an official file.

All three routes are valid answers. The one that isn't is finding out on the day of the inspection.