Nine Days From Zaranj

The capital of Nimruz Province was the first provincial capital the Afghan government lost since 2001 — a dusty administrative town on the Iranian border, remote, thinly garrisoned, of no particular strategic weight. It fell with very little fighting. Officials withdrew, some across the border. In Kabul the loss was logged the way such losses had been logged for twenty years: an isolated setback in a hard province, a resourcing problem, a thing to be corrected next quarter.

Nine days later the government did not exist.

Between the two dates: Sheberghan on the 7th. Kunduz, Sar-e Pol, and Taloqan on the 8th. Pul-e Khumri and Aybak next. Farah, Faizabad, Ghazni. Then Kandahar and Herat within hours of each other, Lashkar Gah, Mazar-i-Sharif on the 14th, Jalalabad and Kabul on the 15th. Nearly every provincial capital in the country, inside nine days, to a force that had spent two decades unable to hold a single city.

Every one of those losses had a local explanation, and every local explanation was accurate. The garrison was undermanned. The air support never came. The elders had already negotiated. The commander was owed four months' pay. Any single item on that list is a management failure. The list itself is a phase transition.

What kills a system is not the quantity of its failures but the correlation between them. Command structures aggregate quantities — districts held, casualties taken, units at strength — and a correlation is not a quantity. It is a property of the whole, visible only from outside the whole.

The comfortable conclusion from there is that nobody saw the list, because reading lists was nobody's job. That conclusion is false, and false in the flattering direction. People whose job was exactly this said exactly this, for years. Three days after Kabul fell, the special inspector general who had spent a decade auditing the force put it flatly: "All the signs have been there. I mean, we've been shining a light on it in multiple reports going back to when I started in 2012 about changing metrics, about ghosts, ghost soldiers who didn't exist, about poor logistics." Dozens of reports. Hundreds of recommendations. Congressional testimony. The reports arrived; they were discounted. That is not blindness. Blindness is forgivable and unfixable — an institution with an interest in not believing its own instruments is neither.

And the instruments were compromised at the source. The Afghan forces carried ghost soldiers: payroll and ration entries for men who did not exist, sustained for years, at a scale that had auditors warning that hundreds of millions of dollars in annual U.S.-funded salaries rested on personnel data nobody could verify. "Units at strength" was never a measurement. It was a claim someone was paid to make. That is the part worth carrying, because it is not aggregation loss. A reporting chain does not passively drop correlation on the way up; it reports whatever protects the reporter, and correlation is the first casualty precisely because it is the only quantity that implicates everyone at once. Every node had an incentive to describe its own sector as recoverable. Sum those incentives and the chain is not a lossy instrument. It is an adversarial one, working exactly as its people were rewarded for making it work. The system counts its parts, finds the count tolerable, and walks off the cliff with a current inventory in hand — and the inventory was fiction long before anybody added it up.

The pattern has a long service record. Sedan broke on May 13, 1940 and was treated in Paris as a sector problem requiring reserves; France signed an armistice six weeks later. Ban Me Thuot fell in March 1975 and was read in Saigon as a highlands defeat worth trading space for time; Saigon fell in fifty-one days. Mosul went in June 2014 and two Iraqi divisions evaporated behind it. In each case the first city was reported as insufficient reinforcement. In each case the actual variable was that the soldiers had already privately concluded the thing was over, and were waiting to find out who else knew.

In Afghanistan there was no need to infer the moment. It was published. On February 29, 2020, in Doha, the United States signed a withdrawal agreement with the Taliban to which the Afghan government was not a party. The premise did not erode invisibly; it was revoked in public, on a date, in a signed document. Every Afghan officer learned that day that his patron had negotiated an exit without him, and had done it with the people shooting at him. Everything after was settlement of accounts. The Taliban spent the following eighteen months brokering surrenders through village elders — most of the summer's "battles" were contracts signed weeks earlier and executed on cue. When the maintenance contractors left in mid-2021 and the aircraft that resupplied isolated outposts stopped flying, the last argument for holding a position went with them. The institution had eighteen months to read what it had signed, and spent them declining.

The instructive control case is next door in time. Najibullah's government survived the Soviet withdrawal in February 1989 for three years. It did not fall when the troops left. It fell in April 1992, months after Moscow stopped sending money. Foreign soldiers were never the load-bearing element in either regime. The flow of resources was, and the belief that the flow would continue was, and belief is the fastest of all of them to go.

So the honest lesson is not that anyone should have predicted Kabul on August 6, and it is not that the system lacked a category for what it was seeing. It had the category. It had the reports. What it lacked was any place where information was safe to be accurate. Its instruments could see a lost city; they could not afford to see a lost premise, because a lost premise implicates whoever is holding the instrument. Outside observers — journalists in Kandahar, Afghans buying plane tickets — could see it fine. Legibility ran backwards, and not by accident: the people with the least data had the clearest picture because they were the only ones with nothing to lose by looking straight at it.

A system whose parts each optimize their own appearance cannot produce a true picture of the whole, however much data it collects. Five years on, then, the useful question is not why Kabul fell, and not even whether anyone is reading the list. It is which currently-running system is being handed its Zaranj this month — and whether a single person inside it is rewarded for saying out loud that the list is the story.