Every home-automation system eventually offers to tell you when to open a window. Most of them answer the wrong question, confidently, using a sensor everybody has. It is a small tragedy of instrumentation: the sensor is honest, the number is real, and the advice is backwards.
Relative humidity is the most quoted and least useful number in a home. It changes when the temperature changes, without a single molecule of water going anywhere. Asking it whether to open a window is like asking a thermometer whether you should wear a coat, having first put the thermometer in the oven.
Relative humidity cannot answer this
The obvious approach compares indoor and outdoor relative humidity: if it is drier out there, open up. This does not work, because relative humidity is relative to temperature. Air at 20 °C and 60 % holds considerably more water than air at 10 °C and 60 %. Two identical percentages can mean very different amounts of water.
Measured here on one build day: outdoor dew point 19.0 °C against indoor 18.1. Opening the windows would have added moisture, while the relative humidity readings suggested the opposite. (A 0.9 °C margin between two consumer sensors sits close to their combined noise, so a reading this tight decides nothing on its own; repeat it before trusting it. The example stays because the direction of the error is the lesson, not the decimals.)
Dew point is a measure of absolute moisture. Lower outside than inside means airing out dries the place; higher means it wets it, whatever the percentages say. It is derived from temperature and humidity by a standard formula, so it costs nothing but the arithmetic.
The question people actually have
Dew point is correct but unsatisfying at the window. What you want to know is: if I open this, what will it be like in here afterwards?
So the useful derived value is the indoor relative humidity that the outside air will produce once it is inside and has reached indoor temperature. Same formula, run backwards. It is a number you can put a threshold on, because it is expressed in the units of the thing you are trying to avoid.
Three vetoes and one override
Three separate factors can each independently say no: particulates outside, moisture, and heat. Any one of them vetoes the advice to open.
But carbon dioxide above roughly 1000 ppm overrides all three. Stale air beats a little humidity, and the ranking is a deliberate judgement rather than an emergent property of some scoring function. One ceiling was added after review: when outdoor particulates are at health-hazard levels (wildfire smoke, an inversion episode), the CO₂ override itself yields.
Stale air beats humidity; it does not beat smoke. An override needs its own override, because ranking mistakes at the top of the chain are the expensive ones. Beyond that, keep the overrides explicit and few; a weighted sum of five factors produces an answer nobody can argue with because nobody can follow it.
The outdoor sensor is optional on purpose
Outdoor particulate readings come from a source that fails regularly. The tempting design makes the advice depend on it. That produces a system which goes quiet precisely when a sensor dies, and silence is indistinguishable from "no advice needed".
So the advisor works without it, and says the outdoor particulate level is unknown. That now includes the ceiling above: with the outdoor feed down, the CO₂ override cannot be capped, so it stays active and the advice carries the admission that the ceiling is not protecting anyone today. Degraded and honest beats absent and silent, all the way up the priority chain.
When two sources disagree, publish both
There are two outdoor particulate sources for one location: a commercial modelled service and the median of a dozen citizen-run sensors within a few kilometres. On build day they read 15.5 and 7.0, a factor of two apart. Both are plausible; the cheap sensors drift with humidity, the model is a model.
Averaging them would produce a number that is wrong in a new and untraceable way. Instead both are exposed as attributes, and the divergence is treated as its own signal: a sustained gap beyond roughly two-fold means one of them is in trouble and worth investigating.
The wind rule that surprises people
There is also an alert for "wind is getting up and a window is open". The non-obvious part is that the threshold for a tilted window is lower than for a wide-open one, 35 km/h against 55.
Intuitively a barely-open window seems safer. The mechanical story (a tilted sash hanging on its lower hinges, taking the gust as a lever) is plausible and, as a reviewer rightly noted, unproven in either direction. The thresholds do not rest on it. They rest on the behavioural half, which nobody disputes: a wide-open window is the one you remember. The tilted one is the one you forget, on the one side of the building the wind is hitting.
The dew point, by contrast, is boring and correct, which is the best combination a sensor can offer.