Sixteen People Died Under a Concrete Canopy Link to heading

At 11:52 on 1 November 2024, the concrete canopy over the entrance to the railway station in Novi Sad collapsed onto the people standing beneath it. Fourteen of them died where they stood, the youngest aged six and the oldest seventy-four. A fifteenth died in hospital on 18 November. The toll eventually reached sixteen.

The station had recently been renovated. That is the detail that turned grief into something else, because in a country where public contracts had been a reliable way to get rich for over a decade, a newly renovated building falling on people was not read as an accident. It was read as a receipt.

Almost two years later, on 25 October 2026, a list of candidates assembled by student assemblies won a snap parliamentary election with roughly 52% of the vote against 32% for the governing bloc, taking somewhere between 135 and 140 of the 250 seats in the National Assembly. Turnout was 67.6%, the highest Serbia has recorded since the dissolution of Yugoslavia in 1992. Aleksandar Vučić, the dominant figure in Serbian politics for fourteen years across the offices of deputy prime minister, prime minister and president, conceded on the night.

I am not writing about this because a government fell. Governments fall constantly, usually to the next faction of the same elite. I am writing about it because of how it was built, and because the how is the part that is portable.

What those students built was, in the most literal sense, a distributed system with no single point of failure. They built it that way on purpose, under hostile conditions, and it held for twenty-three months.

The Shape Engineers Have Always Preferred Link to heading

My first reaction to reading how they organised was not political. It was recognition, and it was closer to delight than to analysis.

Software architecture is a strange corner of human endeavour, because it is one of the very few where we have collectively agreed that concentrated authority is a defect. Not a risk to be managed. A defect. Nobody competent stands up in a design review and argues for a master node you cannot afford to lose. If your system has one, you are expected to be embarrassed about it and to have a plan for removing it.

Everything we consider good practice follows from that single conviction.

We load balance, so that no individual node is irreplaceable and work flows to whoever is available rather than to whoever is nominally in charge. We keep nodes stateless, so that nothing important lives in only one place and losing any one of them costs you nothing you cannot rebuild. We design for graceful degradation, so the system bends under load instead of snapping. We keep backups, and we write disaster recovery playbooks in advance, in calm conditions, because we know perfectly well that nobody reasons clearly at three in the morning with everything on fire. And we treat a vulnerability report as the highest priority item on the board, which means we have decided that the person telling us we are broken is the most valuable person in the room rather than an enemy of the project.

None of that is modesty. It is the accumulated residue of decades of outages. Every one of those practices exists because somebody built the clever centralised version first and watched it fail.

Here is the part I keep turning over. We hold those as professional values, we argue about them in pull requests, we put them in architecture decision records and we feel genuine satisfaction when a system survives something it was not supposed to survive. And in two decades of doing this I had quietly assumed they were a convenience of working with machines. You can keep a server stateless. You cannot keep a person stateless. Human institutions concentrate authority by default, because people want the authority, and someone always ends up at the top holding things that only they hold.

Then several thousand students in a country under a captured government arrived at the same architecture from first principles, with no specification and nobody to copy, and ran it in production against an adversary.

They load balanced. Any faculty could carry the movement when another was blockaded or exhausted. They kept their nodes stateless, in the sense that no individual ever held authority the rest of them could not reconstitute by Tuesday. They degraded gracefully through police violence, through a sonic weapon, through a hostile broadcast media, and through twenty-three months of a government simply waiting for them to get bored. They had their recovery procedure written down before they needed it, which is the whole function of the rulebook I will come to shortly. And when flaws turned up inside their own candidate list, they treated those reports as the highest priority item in front of them and acted on 38 of them, including against their single most prominent figure.

Fault tolerance is a concept I have spent my career applying to infrastructure. Watching it applied by human beings, deliberately, at national scale, under a government that would have been delighted to find someone at the top to remove, is genuinely one of the more beautiful things I have seen.

The students would not describe any of it in these terms. That makes it better, not worse. It means the shape is not an artefact of computing. It is what you converge on when the environment is hostile, the nodes are unreliable, and your opponent will attack whatever you centralise.

That describes what happens in a data centre. Today it also describes what can happen in a country.

The Plenum Is the Whole Trick Link to heading

Start with the thing everyone gets wrong about protest movements.

The conventional shape of an opposition movement is a figurehead with a following. It is easy to understand, easy to report on, easy to rally behind, and catastrophically easy to defeat. A movement with a leader has a single point of failure with a name and a home address. You arrest the leader, or you discredit the leader, or you buy the leader, or you simply wait for the leader to make one unforced error on camera. The movement is a function of one node and that node is exposed.

The Serbian students did not have one.

Decisions were made in plenums: open assemblies at each faculty, attended by whichever students turned up, one person one vote, simple majority after as much deliberation as it took. No chair with casting authority. No designated spokesperson. No executive committee that the plenum merely ratified. The plenum was not an advisory body reporting to a leadership; the plenum was the leadership, and it was a different room full of people every week.

Above the faculty plenums sat larger university-level plenums whose job was coordination between the decentralised departments rather than command over them. Representation was weighted by student numbers. The structure was borrowed in part from the Croatian student movement, which had used the same open horizontal assembly a decade and a half earlier.

Everything came out of those rooms. The sixteen minutes of silence. The faculty blockades. The marches. The decision to contest an election at all.

Think about what this does to the other side’s playbook. There is nobody to arrest whose arrest matters. There is nobody to buy, because nobody holds the authority you would be buying. There is nobody to put on state television and humiliate, because nobody speaks for the movement. There is no negotiating partner, which is enormously frustrating for a government that would very much like to negotiate a settlement with somebody and then declare the matter closed.

The cost is real and worth stating plainly: plenums are slow, they are exhausting, and they produce decisions that are sometimes incoherent at the edges. Horizontal structures trade decisiveness for survivability. The students made that trade deliberately, and against an opponent who controlled the police, the courts and most of the broadcast media, survivability was the right thing to optimise for.

The Weekly Ritual That Needed Nobody to Organise It Link to heading

The second design decision I find genuinely clever is the one that solved sustainment.

Every movement’s real enemy is not the riot police. It is the calendar. Attention decays, news cycles move, people have jobs and exams and children, and the governing side knows all of this perfectly well. The standard government response to mass protest is not suppression; it is patience. Concede nothing, wait eight weeks, let the weather turn.

The students made waiting ineffective by building a commemoration that ran on its own.

Every Friday at 11:52, the minute the canopy fell, people stopped traffic and stood in silence. It began in December 2024 as fifteen minutes for fifteen dead, and became sixteen minutes when the toll rose to sixteen.

Consider the properties of that as a mechanism. It is fixed in time, so it needs no announcement and no organiser. It is short enough that an ordinary person with a job can participate without cost. It is replicable anywhere by any number of people down to one, so a small town with eleven willing participants and a capital city with a hundred thousand both run the same ritual. It is unambiguously about sixteen dead civilians, which makes it close to impossible to attack without the attack becoming the story. And it recurs weekly, which means the movement’s floor does not decay toward zero between its peaks.

By March 2025 the peaks were over 300,000 people in Belgrade.

A heartbeat you cannot switch off, in other words, which is why the patience strategy failed.

The Directive, or Admission Control for a Candidate List Link to heading

Here is where it gets unusual, and here is the part I would most want other movements to study.

Winning an election means fielding candidates. Fielding candidates is the exact moment that movements like this historically die, because the candidate list is where the existing political class gets back in. Experienced operators volunteer to help. They have name recognition, media contacts, and a plausible claim to competence. They also are the system the movement formed to remove, and once they are on the list the movement has achieved nothing except rotating the personnel.

In April 2025 the plenums wrote a rulebook, known as the Directive, that closed the door.

Former members of the twenty parties in the governing coalition were barred. Former officials of the established opposition parties were barred too, which is the clause that tells you how seriously this was meant; the movement treated the official opposition as part of the same captured arrangement rather than as a natural ally. Currently enrolled students were also barred, and students with their studies on hold, which removed the obvious temptation for the movement’s own most prominent organisers to convert visibility into a seat.

What was left was people with no stake in the existing arrangement. The published list ran to 250 largely unknown names: doctors, academics, actors, farmers, former athletes, activists. It was headed by Ilija Srdanović, a cardiologist and professor from Novi Sad, which is to say by someone whose standing came from a career entirely outside politics.

That is admission control. You do not vet for loyalty or for talent; you vet for the absence of the thing you are trying to eliminate, and you write the rule down before you know which individuals it will exclude.

Then they enforced it on themselves, which is the harder half.

In August 2026 the plenums put 99 of more than 300 prospective candidates under formal challenge. By September, 38 candidates and substitutes had been removed from the list. The most prominent was Vladan Đokić, the rector of the University of Belgrade, a figure who had been publicly aligned with the protests and whom 43 of the 85 participating student assemblies voted to remove.

None of this had any legal force whatsoever. Every one of those 38 people could have refused to stand down and remained on the ballot.

None of them did.

Where the Design Is Genuinely Weak Link to heading

I am not going to pretend this is a finished blueprint, because the movement itself could not solve its hardest problem and it is more useful to say so.

The plenums understood exactly what the risk was after a win: that their own MPs, once seated with a majority, become a new elite with a four-year mandate and no reason to keep listening to a room full of students. They tried to engineer against it. They agreed in principle to cap their MPs’ mandates at sixteen to eighteen months, forcing a return to the voters long before the term expired.

Then in a contentious vote in September 2025 the cap was struck out altogether.

The reason is partly legal. Serbia’s Constitutional Court struck down the practice of binding MPs with undated resignation letters in 2010 and 2011, which was precisely the instrument that would have made a mandate limit enforceable. Under Serbian law a seated MP holds their own mandate and cannot be made to hand it back on demand. What the candidates eventually accepted was moral and political responsibility to the plenums with no enforcement mechanism attached, a formulation the Belgrade plenums approved by 23 votes to 11.

So the movement has built excellent admission control and has no revocation. The eligibility rules worked because they operated on people who wanted something and had not yet received it. Those same rules have no purchase on an MP who already holds the seat.

The next two years are the actual test, and it is not the test the celebrations are about. The question is not whether a decentralised movement can take power from an entrenched one. That has now been demonstrated. The question is whether it can hold the people it sent into office accountable to the rooms that selected them, using nothing but moral authority and the knowledge that the plenums are still meeting.

I genuinely do not know the answer. Neither do they. Anyone copying this design should copy the enforcement problem too, as an open issue, and should try to solve it before they win rather than after.

Non-Violence Held, and It Was Not Because Nobody Was Provoked Link to heading

The thing I most want to resist is the sentimental version of this story, where two years pass peacefully because everyone was nice. That is not what happened, and the sanitised version is useless to anyone trying to learn from it.

Protesters were hit with a suspected acoustic device in Belgrade in March 2025, an allegation the government denied. There were clashes with police in June 2025. On 14 August 2025 in Valjevo, officers were filmed striking a man who was already lying on the street; police said the force was lawful and that citizens had attacked officers first. The following day masked men vandalised a cafe, injured an employee and assaulted people at a repair shop. Nobody was meaningfully held to account for any of it.

What matters is the asymmetry. Across almost two years, with hundreds of thousands of people in the streets and every available provocation applied, the movement did not convert into an insurgency. That was not luck and it was not temperament. It was an operating constraint, held deliberately, through the exact conditions designed to break it.

It is worth being clear-eyed about why non-violence was the correct strategic choice rather than merely the decent one. A movement that stays peaceful keeps the moral asymmetry that makes state force counterproductive; every baton on film recruits for the other side. It keeps the participation ceiling high, because parents and doctors and pensioners will stand in a street for sixteen minutes and will not attend a riot. And critically, it keeps the movement’s own objective intact. You cannot credibly build an anti-corruption government on the back of political violence, because the first thing political violence does is select for the kind of person who is good at it.

You can copy the design off a page. Holding that discipline for two years while being hit is the part nobody can hand you.

Why You Probably Did Not Hear Much About It Link to heading

Now the part that is my opinion rather than the record, and I want to keep the two clearly separated.

Major international outlets did cover this. Reuters, the Guardian, the Associated Press, Al Jazeera, CNN, NPR and Euronews all reported it, and the ABC carried it here in Australia. I cannot claim the story was suppressed and I am not going to.

What I will say is that I had to go looking. A movement spent twenty-three months removing a fourteen-year incumbent in a European country, and I encountered it as a result at the end rather than as a story while it was happening. I suspect that is most people’s experience of it, and I think that is worth being curious about.

I do not know the cause. It may be that foreign bureaus have been cut to the bone and sustained coverage of a two-year civic process in a country of under seven million is no longer economic. It may be that a weekly sixteen-minute silence is structurally unreportable, because recurring discipline is not an event and news organisations cover events. It may be that recommendation algorithms select for conflict and this story’s defining feature was the absence of it. Any of those explanations is sufficient on its own and none of them require anybody to have decided anything.

But I would note which stories do tend to travel. Instability travels. Collapse travels. Strongmen travel, both admiringly and disapprovingly. Street violence travels extremely well. What does not travel nearly as well is a story in which ordinary people organise patiently, hold a difficult discipline for two years, write rules on themselves, and win at the ballot box, because that story carries an implication the other ones do not: that this is available, and that the barrier is organisation rather than fate.

That is a claim about attention, not about conspiracy, and the reader can weigh it against their own news feed.

What Is Actually Exportable Link to heading

Strip out everything specific to Serbia and a short list remains. Not a guarantee, and emphatically not a reason to think this is easy, but a set of decisions that are available to anyone.

Make decisions in open assemblies with no permanent leadership, so that there is no node whose capture ends the movement. Build a recurring ritual that is fixed, short, cheap to join and needs no organiser, so that attrition and official patience stop working. Write your eligibility rules down before you know who they will exclude, and make them bar the established opposition as well as the incumbent. Enforce those rules on your own prominent people, in public, when it hurts. Let different parts of the movement contribute what they actually have, the way the IT students built encrypted coordination tooling and the law students and practising lawyers built a legal-response network for anyone detained. Hold non-violence as a hard constraint and expect it to be tested, because it will be. And solve the accountability problem for the people you send into office before the election rather than after it, because that is the one the Serbians did not manage and it is where this will succeed or fail.

Fourteen years of accumulated power came apart because a large number of people stayed organised for twenty-three months and refused to produce a leader anyone could remove.

It is early. The hard part, which is governing, has not started, and a sizeable minority of the country voted the other way and has to be governed too. There is nothing inevitable about what happens next and the movement’s own weakest design decision sits directly in its path.

But something has been demonstrated that was worth demonstrating. Entrenched power is not a fact of nature. It is a system, and systems have failure modes, and sixteen minutes of silence every Friday for two years turns out to be one of them.

The Reporting Link to heading

Read it and form your own view.

On the disaster itself:

On how the movement was organised, which is the part worth your time:

On the election and the result:

On the incumbent side and what it cost:

The Guardian also ran useful pre-election background on the movement’s strategy, a stakes explainer, post-result reporting from Belgrade and an editorial on the move from protest to power, and the Associated Press and Reuters both carried strong post-election pieces on the difficulty of turning a movement into a government. Search their sites by date if you want them; I have linked only what I could verify directly.

One point of accuracy, since a lot of the commentary has muddled it. The vote on 25 October was a parliamentary election. Vučić resigned the presidency on 27 September 2026, barred from a third term and intending to take the premiership instead, which holds more formal authority under the Serbian constitution. He was a candidate for prime minister on election day, not the sitting president. Ana Brnabić is acting president, and a presidential election must follow within ninety days of the resignation.