It rained on nineteen days in thirty. The garden still went dry.

A month that felt wet, measured properly. The rain arrived on two days in three and the garden lost nearly twice what it gained.

This one started as a complaint about the site. In the middle of September the answer came back "water today", and it felt wrong — there had been showers all fortnight. The obvious suspicion was that the arithmetic was too eager.

So it was measured rather than argued about. Here is what actually fell on that garden, in west London, between 18 August and 16 September 2026.

19 of 30days had rain on them
71mmfell in total
47mmreached the roots
92mmthe plants used

Two out for every one in

The garden lost nearly twice what it gained. That is the whole month in one line, and it is why the answer was "water" — not because the model was impatient, but because 47 millimetres arrived and 92 went out.

The 92 is not a guess. It is what the sun and the wind actually removed, day by day, measured by the same weather service that reports the rain. It is called evapotranspiration and it is the half of the sum that every weather app leaves out.

Why it felt wet

Because rain on nineteen days out of thirty is wet, by the only measure anyone carries around in their head. You remember the days you got rained on. Nobody remembers millimetres.

But six of those nineteen days gave the roots nothing at all. Not "a little" — nothing. Enough rain to wet the leaves, darken the path and register on a forecast, and all of it gone by lunchtime without ever reaching soil.

A wet fortnight and a watered garden are different things, and the gap between them is where this site lives.

What it did not turn out to be

The first theory was seasonal: that the model treats September like May, and asks for water a plant slowing down for the year does not need. That turned out to be wrong, and it is worth saying so plainly.

The water-out figure is a measurement, not an assumption, and it falls on its own as the year turns — about a quarter lower in that September than in May. The gap between waterings stretches to match, with no calendar anywhere in the code. For the same garden it went from roughly five days in May to ten.

The same arithmetic run on Sydney moves the opposite way at the same moment, which is the proof there is no hemisphere built into any of it.

The useful version

If you take one thing from this: count millimetres, not rainy days. They are not the same measurement and in a British autumn they can point in opposite directions. A fortnight of drizzle can leave a garden drier than a fortnight of nothing followed by one proper downpour.

Check your own garden

The front page works this out for where you are, from the rain that actually reached your soil over the last fortnight.

Check my garden

The other notes