The mill

In 1921, on a forest floor in Guyana, a naturalist found a circle of army ants and stood watching it for a long time. The circle was enormous. It took an ant something on the order of two and a half hours to complete one lap of it, and by the time he found it, it had been turning for most of a day. It went on turning after he left. It was still turning the next day. Then it thinned, and stopped, because the ants in it had died.

Nothing was chasing them.

There was no obstacle, no predator, no drought, nothing in the environment that had changed. The ground was ordinary. Food lay within a few metres of the column for the entire time. Every ant in the circle was healthy when it began walking and every one of them walked until it could not, past food, in a ring, for two days.

And not one of them did anything wrong.

An army ant of this kind cannot see, in any sense a person would recognise as seeing. What it has instead is chemistry and contact. It reads the ground in front of it and the body ahead of it, and it moves along the strongest mark it can detect, laying a little more of the same as it goes. That is close to the whole of the behaviour. It is also, by an enormous margin, the most successful arrangement its lineage has ever found.

The rule is not a shortcut and it is not a limitation. It is the thing that lets a few hundred thousand blind animals move as a raiding front across a forest floor, find what is there, carry it home, and do it again tomorrow in a different direction. Everything the column can do, it can do because each ant follows the mark in front of it without asking where the mark came from.

The mark is where the intelligence is kept. No ant holds the route. The ground holds it, and the ants read the ground.

A raiding column has a leading edge, and the leading edge is doing something the rest of the column is not. It is in contact with ground nobody has marked yet. It is the only part of the system that is looking at the world rather than at the record of the world, and everything behind it inherits its judgement, one body at a time, all the way back to the nest.

Lose that edge, and the column does not notice.

It cannot notice. There is nothing in an ant that asks whether the mark it is following still corresponds to anything. If the front of a column swings far enough to meet its own flank, the path closes, and what is left is a loop of extremely strong chemistry with no outside on it. Every ant on that loop is now following an excellent trail. It is the best-marked ground any of them has ever walked, and it gets better with every lap, because they are all still laying.

The feedback that made the colony intelligent is the feedback that closes the ring and tightens it.

This is the part worth sitting with. The failure is not a breakdown of the mechanism. It is the mechanism, running correctly, with the thing that used to correct it removed. There is no bug to find. Every individual ant, examined on its own, is behaving exactly as it should, and the circle kills all of them.

A mark records what happened. That is all it can do and all it was ever able to do. It says: this was walked, and something came of it. It does not say whether that is still true, because nothing in the past can say anything about the present.

So a system that keeps its memory outside itself needs something that keeps touching the world and depositing what it finds. Not a manager and not a plan — the colony has neither and does not want either. Something at the edge, in contact, still laying marks that came from outside the marks.

Take that away and what remains looks, from the inside, exactly like working.

That is the property that makes it dangerous. A ring is not a system in distress. It is a system in perfect agreement with itself, doing at full speed the thing it was built to do, and it is indistinguishable from health by every measure available inside it. The ants are not confused. Their coordination is flawless. The trail is strong, and getting stronger, and it is a circle.

The same thing happened here, in software, and it is written down.

On 25 August 2026 this system ran a large piece of work as concurrent fleets. The design underneath one of them changed partway through — the whole premise of what it was building was abandoned. The fleet went on building it. It was still going, faithfully and competently, until a person noticed and stopped it by hand.

The record of that day is precise about why, and it does not blame the worker:

The premise died in the conductor's head and was never deposited anywhere a worker reads — the plan doc still read as an open trail, so a worker followed it faithfully.

The document goes further, and this is the sentence that matters. No lock would have caught it. No registry of who-claimed-what would have caught it. There was no contention; nothing raced; no second worker touched that ground. One mark on the ground had stopped being true, everything downstream of it was correct, and the only repair is on the other side of the nest entirely: a judgment change is real only when it is written where workers read.

Telling the conductor is not a deposit.

The circle in Guyana is the oldest and clearest instance of a thing that turns out to be everywhere, and it does not require ants. It requires only a system that keeps its memory outside itself, which is every system worth the name, and a moment when nothing is left at the edge putting new marks down.

The naturalist watched for as long as he could stand to. There was nothing to do. A person cannot tell an ant that the ground it is standing on is a lie, and would not be believed by anything in that column even if the column had ears.

He went away and came back and it was still going round.

The mark was perfect. The mark was the problem.

The seventeen lessons that do this to a system the reader owns are at /course. The last one is the Gordon Test.