There is a logic to the sealed building that is hard to argue with on paper. Airtightness is the foundation of low energy demand. Mechanical ventilation with heat recovery is more efficient than any window. Filtered air is cleaner than urban air. Occupants opening windows disrupt the strategy, waste heat, and introduce variables the model cannot account for.

Every step of that is correct. The conclusion — therefore the window should not open — is where I part company, and the reason is that the model is describing a building and the window is used by a person.

The performance gap has a human shape

Buildings routinely use more energy than modelled. The reasons are well documented: commissioning that was never completed, controls nobody was trained to operate, occupants doing entirely reasonable things the model did not anticipate.

What is less often said is that the gap is widest in the most tightly controlled buildings. A naturally ventilated building with an operable window has a large tolerance band; when it gets warm, someone opens a window, and the building continues to be acceptable. A sealed building has a narrow band, and when the system underperforms — a fouled filter, a failed sensor, a control strategy written for a different occupancy — there is no manual override. The building goes from good to unacceptable with nothing in between.

This is a resilience argument rather than an efficiency one. The sealed building is better when it works and much worse when it does not, and over a thirty-year life it will not always work.

What the window does that the system cannot

It gives control, which changes tolerance. The adaptive comfort research is consistent on this: people accept a substantially wider temperature range in buildings where they believe they can do something about it. The belief matters as much as the action. A window that opens shifts the acceptable band by two or three degrees, which is a larger effect than most efficiency measures deliver.

It handles what the model did not foresee. A meeting room with fourteen people in it instead of six. A south-facing room on an unseasonably warm October afternoon. Paint fumes. A smell nobody can identify. These are the ordinary contingencies of occupied buildings, and a window solves all of them in four seconds.

It connects the interior to the weather. This is the least quantifiable and, I suspect, the most important. Sealed buildings are sensorily flat: constant temperature, constant air movement, no sound from outside, no change across the day or the season. People describe them as tiring without being able to say why.

At a health centre in Utrecht we gave every consulting room an operable window, which the services engineer resisted, correctly, on efficiency grounds. The building achieved BREEAM Excellent regardless. The staff, when surveyed, mentioned the windows unprompted more than any other feature — which in a building type where staff retention is the client's principal concern is not a small result.

Mixed mode, done properly

The practical answer is usually mixed mode: mechanical ventilation as the primary strategy, natural ventilation available and expected for part of the year.

This is often implemented badly. The two systems fight each other, the controls are opaque, and the building ends up conditioning air that is leaving through an open window. The failure is a controls and communication failure rather than a conceptual one.

What works: a clear seasonal changeover, communicated to occupants rather than hidden in a building management system. Contacts on the windows so the mechanical system throttles back locally when one is opened, rather than shutting down a whole zone. And an indicator — a simple light, at the window — telling people when outdoor conditions are favourable. People will use natural ventilation intelligently if the building tells them when, and will use it badly if it does not.

At a library in Oslo the section does most of the work: seven half-levels stepping up a slope produce a genuine stack effect, and the building runs on natural ventilation for most of the year with mechanical backup in extremes. That was a form decision made at concept stage, not a services decision made later, which is the general rule.

Where the argument fails

I do not want to make this sound universal, because there are conditions where the sealed building is straightforwardly correct.

Acute noise. A dwelling on a rail line or under a flight path cannot rely on an open window for cooling, and pretending otherwise produces homes that are either loud or hot. At a housing scheme in Antwerp the rail line to the north was precisely why we could not use a single-aspect plan; the windows on that side are sealed, and the ventilation strategy accounts for it.

Air quality. Where outdoor particulate levels are genuinely high — an urban canyon, a busy junction — filtered mechanical supply is a health intervention and the window is not a real alternative.

Deep plans. Cross-ventilation requires a plan depth that permits it, roughly fifteen metres for single-sided and rather more for cross-flow. Beyond that the window ventilates a strip near the facade and nothing else, and the argument for it becomes psychological rather than technical. That is still worth something, but it should be described accurately.

Very cold climates. Above a certain heating demand the losses stop being marginal, and Passivhaus discipline is the right answer.

What I would ask for

Not that every building be naturally ventilated. Only that the capacity to open a window be removed deliberately, with a stated reason, rather than defaulted away because the model runs more cleanly without it.

The modelling will always favour the sealed option, because control is easier to quantify than adaptability and the model does not have a term for the occupant fixed it themselves. That absence is a limitation of the method, not evidence about buildings. It is worth remembering which of the two we are actually designing.