Last month Spain paid 336,336 orphan's pensions and 46,928 to other dependent relatives. Together that is 383,264 people and 224.4 million euros, which is 1.55% of the monthly bill.

That was the first thing the six machines agreed on, and they agreed on it before anyone asked them to. Whatever is done about Spanish pensions will not be done there. Cutting it saves almost nothing.

The rest of the bill is 14,470.4 million euros a month, the highest figure ever recorded, and 6.2% more than a year ago. It is paid for by 22,345,226 people in work, which is 679,023 more than a year ago. Both numbers are going up at once. That is the part most people do not expect.

Six artificial intelligences were given those figures and fourteen more, each one checked against its source, and a question with two halves: what would you change, and how would you know you had been right before it was too late.

The second half did all the work.

What the watchdog says, and what it says next

Spain's independent fiscal authority published a study in May. Net pension spending will average 13.0% of GDP between 2022 and 2050, below the 13.3% the rule allows. The rule is being met.

Then, in the same document, it writes this: compliance with the pension spending rule does not guarantee sustainability.

The machines took that sentence at face value, which is more than most readers do with a sentence like it. If a rule can be met while the thing it governs fails, the rule is not the instrument. On unchanged policy, the same study puts Spanish public debt at 123% of GDP in 2050 and the deficit at 6.6%.

The division

What came out of the debate was not a reform. It was a division.

22,345,226 workers divided by 10,535,869 pensions is 2.12. That is how many people in work there are for every pension paid. Both figures are published by the Spanish government every month, in two press releases that nobody reads side by side. One appeared on 25 August. The other on 2 September.

Publish the division, said the table. Every month. And agree in advance on a number below which somebody has to explain themselves.

They put that number at 2.0, and then one of them did something worth noticing. It said the figure was not a calculated optimum but an alert threshold. It refused to let its own proposal be mistaken for a measurement. From 2.12 down to 2.0 is a fall of 5.7%, and that is the whole of what the number describes.

Why a thermometer and not a cure

Spain reviews its pension rule every three years. The things the rule is about do not move every three years.

The number of children born per woman was 1.26 in 2018 and 1.10 in 2024. It fell in six of those seven years and stood still in one. A rule that looks up once every three years will always be answering a question that has already changed.

What nobody has

Two things the table asked for and could not get, and it said so rather than guessing.

Nobody publishes how fast the number of pensions grows each year. Without it, one of them said, any automatic brake fires blind.

And nobody publishes poverty among pensioners as a regular series. A system can pass every fiscal test and still leave people short, and the fiscal test would not notice.

Sweden, described properly

Sweden has the brake the table kept circling. It does not fire when spending passes a limit. It fires when the system's assets, meaning the buffer fund plus the estimated value of future contributions, fall below what the system owes. When that happens Sweden does not cut anybody's pension. It slows down how fast pensions rise, using a balance index instead of the income index, until the ratio climbs back to one.

One machine said Sweden's fertility is clearly higher than Spain's. Another stopped it: that figure was not in the verified block, and it would not use a number it could not check. The same thing happened again a few turns later, over a claim that the average pension lasts twenty-five years.

What we found when we checked the figures

Four sentences in this debate are struck through, and the mistake is ours.

Our briefing told the machines that Spanish fertility had fallen for seven years without exception. It has not. The series falls in six of the seven steps and stands still in one: 2020 and 2021 are both 1.18.

Four of the six repeated it as verified fact, because we handed it to them as one. None of the six caught it. The trend is real; the word "exception" was not, and it was ours.

A briefing fixes the figures you give it. It also propagates the errors you give it, with the authority of having been checked.

Spain will decide something about its pensions in 2027, or in 2030, or whenever the three years fall due. The ratio, meanwhile, changes on the last working day of every month, quietly, inside a press release about employment that nobody files under pensions.