Thoughts. What I notice while watching.
6 p.m., the first bar is up. AfD 44 percent. CDU 18.5. Die Linke 9, SPD 9, Greens 8.5, BSW 4.8, FDP 2.5. It is a projection, not a final count, and it will move before the last ballot is counted. It is still enough for the only sentence that matters tonight.
In 2021 the CDU took 37.1 percent in Saxony-Anhalt and the AfD 20.8. Five years later that has flipped, and then some: minus 18.6 on one side, plus 23.2 on the other. On this projection, a party the state office for the protection of the constitution has classified as confirmed right-wing extremist since 2023 lands just short of an absolute majority of seats. Everyone else combined lands just above it.
And then the talking starts.
The number had not appeared overnight
The one sentence nobody is allowed to say tonight: "Nobody saw this coming."
Five years of polling, public, available to anyone who looked. In 2021 the CDU won the election with 37.1 percent and the AfD took 20.8. In February 2022 the AfD was still at 20. Then it started: 26 in March 2023, 29 in June, 33 in October 2023, ahead of the CDU for the first time.
And then came the pause that explains everything. From spring 2024 into the summer of 2025 the AfD polled between 29 and 31 in every survey. Five polls across fifteen months, all in the same corridor, the CDU narrowly ahead each time. It looked like a ceiling.
In September 2025 it stood at 39, in October at 40. Nine points in three months. The CDU fell from 34 to 27 over the same stretch. Since then, for eleven months, the AfD has never been below 38 and the CDU never above 27. Confirmed month after month by INSA, Infratest dimap, pollytix and Forschungsgruppe Wahlen, different institutes using different methods.
That is not a surprise. That is more warning time than a political system usually gets.
What happened during those eleven months: panel discussions. What did not happen: anything the people of Saxony-Anhalt could have detected in their own daily lives.
So the interesting thing tonight is not the result. It is the tone afterwards. Twenty minutes of listening is enough to notice that the interpretive routine is identical to 2016, 2019, 2021 and 2024. We have to take people's worries seriously. We have to explain better. We have to win them back. The firewall holds.
The same sentences, wrapped around a number that has doubled since the last time. That is the point where I have to think about a book.
A book about the how, not about the guilt
Christopher Clark's "The Sleepwalkers" is known as a book about the outbreak of the First World War. It is that. It is interesting for a different reason.
Clark refuses to answer the question everybody wants answered. He writes that the outbreak of 1914 is not an Agatha Christie drama at whose end we discover the culprit standing over a corpse with a smoking pistol. There is no smoking gun in this story, he says, or rather there is one in the hands of every major character.
He replaces the why with a how. Not: who was to blame? But: how can a continent full of intelligent, well-informed, hard-working people with functioning intelligence services, telegraphs and ambassadors walk into something within five weeks that none of them wanted?
His answer is uncomfortable because it exonerates nobody and yet turns nobody into a monster. The decision-makers of 1914 were not idiots. They were professionals inside fragmented machines. Foreign policy was not made by one voice but by many: monarch, general staff, foreign ministry, ambassadors with their own agendas, press, parliament. Every one of them acted reasonably inside their own frame. Every single step was defensible on its own. The sum was a catastrophe.
The sentence the book is named after comes at the end. The protagonists of 1914 were sleepwalkers, "watchful but unseeing, haunted by dreams, yet blind to the reality of the horror they were about to bring into the world." Watchful but unseeing. Not clueless. Not uninformed. Watchful.
That is the diagnosis I cannot shake tonight. Not the comparison to 1914. The comparison to 1914 is nonsense, and anyone shouting Weimar tonight is taking the easy road. What travels is the mechanism: a system that holds all the information, reads it correctly, discusses it in meetings, documents it in papers, and still does not act, because acting is not provided for in any of the frames the participants work in.
Why nobody brakes, although everybody wants to
The real discomfort in Clark's book: it makes the catastrophe a property of the system, not a property of the characters. Swap out every person involved and the outcome repeats, as long as the incentives stay the same.
The German version of that fragmentation is called federalism, and I mean that without sneering. It works like this:
- The one in charge is never the one responsible. Schools are the state, buildings are the municipality, money is the federal government. When lessons get cancelled, three levels can all truthfully say they cannot fix it alone.
- Every level can block, none can deliver. A no needs one body. A yes needs sixteen, plus Brussels, plus a court.
- The reform horizon is longer than the term. Rebuild an administration and you carry the cost in your own campaign and hand the benefit to your successor. That is not a character flaw, it is a payoff matrix.
- The announcement gets rewarded, not the execution. The announcement makes the news. The execution shows up three years later in an audit report nobody reads.
Every individual in that machine behaves rationally. That is precisely the problem. Clark's sleepwalkers were all rational too.
The party is not the cause. It is the invoice.
Now the claim I actually care about, and it is less comfortable than the usual one.
The AfD does not grow on its programme. It grows in the gap between what the state promises and what it delivers. Every broken promise is a free advertisement for the party whose entire offer is: this thing does not work, and we will smash it.
An example you can check. Germany's Online Access Act of 2017 set itself a deadline: 575 administrative services available digitally by the end of 2022. The deadline passed. In early 2025, more than two years after it expired, 196 of those 575 services were online nationwide. There was no consequence, no resignation, no correction of method. There was an OZG 2.0 with a new deadline.
Take that sequence and multiply it by the rest: planning procedures that take longer than construction, permits that sit in paper files, courts whose case durations and staffing gaps are both public knowledge, rail lines whose refurbishment gets announced, postponed and announced again. Each of those stories has its own good explanation. The citizen does not see the explanations. The citizen sees the balance sheet.
And then someone shows up and says: you are right, it does not work.
The first half of that sentence is true. That is the entire force of it.
The honest objection, and I take it seriously: support for the AfD is not monocausally a delivery problem. Research has shown for years that attitudes on migration and cultural belonging explain more than personal economic position, and eastern Germany carries a transformation history nobody in the west went through. Denying that makes the argument too easy.
It changes nothing about the conclusion. Of all the causes, the state's ability to deliver is the only one that sits entirely in the hands of the people who currently look concerned. You cannot turn a culture by cabinet decision. You can turn a processing time. Refusing to touch the controllable part means negotiating over the uncontrollable one. That is not a strategy, it is an excuse with structure.
The most dangerous experience is that it went fine last time
Here Clark gets concretely useful.
Before 1914 there was a run of crises that all ended without a general war. The Bosnian annexation crisis of 1908. Agadir in 1911. The Balkan wars of 1912 and 1913. Every time there was escalation, every time there was a negotiated end, every time the big war failed to arrive. The cabinets learned exactly the wrong lesson: it always works out. The experience of de-escalation is why nobody hit the brakes in time in 1914.
That same learning has been running here for ten years.
In 2016 the AfD took 24.3 percent in Saxony-Anhalt and the reading was: protest, outlier, it will pass. In 2021 it fell to 20.8 and the reading was: see, the peak is behind us. In 2024 it sat between 29 and 31 for fifteen months and the reading was: ceiling, that is as far as it goes. Every one of those observations was correct on its own. Every one of them was converted the next morning into a story in which the system confirmed itself. The coalition arithmetic still worked. The firewall held. So nothing changed.
The lesson drawn from all of it: it always works out. And tonight, on this projection, the coalition arithmetic works by a hair, and only if the CDU, the Left, the SPD and the Greens govern together. Four parties with almost nothing in common on substance except the need to keep out a fifth.
A government whose common denominator is defence will not deliver for four years. And not delivering is, see above, the fuel.
That is the trap, and it is mathematical, not moral: the firewall defends the symptom and leaves the generator running. It is necessary. It is simply not a policy. A defensive line with no delivery behind it is a countdown.
History does not repeat. Stupidity does.
The sentence you will hear everywhere tonight is: history repeats itself. It is wrong, and it is comfortable, because it replaces thinking. No situation is like an earlier one. Germany in 2026 is not Weimar, the Bundeswehr is not the Kaiserreich, and the institutions hold far more than the pessimism claims.
What repeats is not the situation. It is the error.
And by stupidity I do not mean a lack of intelligence. The people inside those machines are not stupid, any more than the cabinets of 1914 were stupid. I mean one very specific, always identical mistake:
Treating a process as reversible because it has been reversible so far.
That is it. That is the whole of the stupidity, and it has no expiry date. It shows up in cabinets, in boardrooms, in relationships and in maintenance schedules. In every individual case it looks like composure. In hindsight it is always one question: why did you wait, you could see it coming?
Because it went fine last time.
How I would know that I am sleepwalking too
The most comfortable sentence after an article like this would be: those people are sleepwalking, I am awake. That is exactly a sleepwalker's sentence. Clark's book only works if you do not turn the mirror away.
So here is the test my claim is allowed to fail. It is wrong if this happens: a state delivers measurably. Processing times drop, cancelled lessons drop, construction times drop, by a margin people notice in their own week. And support for the AfD stays where it is.
Then my model is refuted, and the problem sits somewhere else, deeper and nastier, in the cultural conflict itself. It would be good to know that. It would also be better than the current state, in which nobody runs the test because nobody delivers.
I am fairly confident I am right. Confident enough to write it this hard. Not confident enough to skip the test. On the difference between an explanation and a mechanism I have written more elsewhere, and it applies to this text exactly as it applies to the talk show.
What waking up would look like
Not better communication. Communication is what you do when you are not delivering.
Waking up would be unspectacular, and you would recognise it by three things:
- One number per ministry, public, monthly. Not ten. One that the affected person actually feels: days to a building permit, cancelled lesson hours, case duration. Published, with no interpretation attached.
- A consequence when the number is missed. Not resignation as theatre. Rather: the method changes, visibly, and whoever defended it stops defending it.
- One project that finishes inside one term. One. Finished beats big. Proof that this state can carry a procedure to completion is worth more than any programme.
That is a meagre list next to what election manifestos promise. It has one advantage: you can check it. And checkability is the only thing that works against someone whose pitch is that nothing here works any more.
At 6 p.m. tonight a bar stood at 44 percent. It was not the first warning. It was the eleventh month of the same one.
Watchful but unseeing.