One Province a Day

Between those two dates, something in the neighborhood of thirty provincial capitals changed hands. Sheberghan on the seventh. Kunduz, Taloqan, and Sar-e Pol on the eighth. Then Pul-e Khumri, Faizabad, Aybak, Farah. Ghazni, Herat, and Kandahar inside about thirty-six hours. Lashkar Gah. Mazar-i-Sharif, where the northern warlords had promised a last stand and instead crossed into Uzbekistan. Jalalabad, which fell on the same morning Kabul did.

Three a day, more or less, for ten days — against a force that on paper numbered somewhere north of three hundred thousand, had absorbed roughly ninety billion American dollars, and had been trained for two decades by the most expensively equipped military establishment in human history.

One province a day is how it got narrated while it was happening, and the phrase stuck because it was the fastest rate anyone watching was prepared to imagine. It was off by a factor of three. Keep hold of that, because it is the small version of the finding: even the account of the collapse undercounted the collapse, and the undercount ran in the direction of what the observers already believed.

The immediate commentary treated the speed as a shock — an intelligence failure, a stunning collapse, a surprise. It was none of those things. It was a disclosure. The structure came apart at the speed it did because there was very little structure to come apart, and everyone with direct exposure to the file had been saying so, in writing, for years.

i · the metrics measured the donor

The first thing to understand about the Afghan National Defense and Security Forces is that its most-cited number was substantially fictional.

The Special Inspector General for Afghanistan Reconstruction — a US government office, funded by Congress, producing reports nobody was obliged to read — spent years documenting that the payroll counts could not be verified. Soldiers who had deserted stayed on the rolls. Soldiers who had never existed drew salaries into accounts controlled by commanders. The "ghost soldier" phenomenon was not a rumor from the field; it was a recurring line item in official American oversight reporting, published on schedule, and functionally ignored, because the number it corrected was the number the entire enterprise used to demonstrate progress.

This is the mechanism, and it is more general than Afghanistan. When an outside power funds an institution, the metrics that get built are the ones the funder can generate and verify: dollars disbursed, recruits processed, rifles delivered, brigades stood up, districts nominally under government control. Every one of those measures an input the donor supplied. None of them measures the thing that actually determines whether an army fights — whether the men in it believe the state they serve will exist next year and is worth the risk in the meantime.

The easy version of what happened next is that legitimacy has no unit, so it went unmeasured, and what goes unmeasured does not exist administratively. That version is too generous. It was measured, repeatedly, by the office Congress funded to measure it. SIGAR said the payroll was fiction. It said the ministries could not sustain themselves unassisted. It said the force was not viable without external support, and it said so in documents with dates on them, indexed and public. The information was not missing. It was present, official, on schedule, and completely inert.

Which makes this an incentive failure rather than a measurement failure — the worse of the two findings, because measurement failures can be fixed by measuring better.

The dependency ran deeper than payroll. The Afghan air force — the actual battlefield differentiator, the thing that let isolated outposts survive being surrounded — was maintained by foreign contractors. When the contractors left, the maintenance stopped, and the aircraft went unserviceable on a timeline measured in weeks. The logistics system that moved ammunition and food to remote checkpoints was built on top of American coordination. Remove the coordination and the checkpoints get hungry, then get quiet, then get taken.

An institution that only functions while an external party is holding it up is not an institution. It is a service contract. And its real constituency — the party it was accountable to, the party whose satisfaction determined its budget — was never the Afghan population. It was Washington.

So was the measurement system. That is the piece that keeps getting mistaken for an oversight. The dashboard was not an error anybody committed; it was a product, built to the specification of the party paying for it, and it discharged its function flawlessly — it produced evidence of progress for the constituency that required evidence of progress in order to keep authorizing the funds. SIGAR's contrary reporting was never suppressed. It didn't need to be. It was structurally unusable, because no channel existed by which an accurate finding about Afghan legitimacy could move an American budget line. An information system aimed at its funder rather than at its terrain will go on telling the truth and go on changing nothing, indefinitely, at full volume.

ii · they bought the network instead of fighting it

The second thing to understand is that most of those capitals did not fall in combat. They were handed over.

The Taliban had spent the preceding year doing something more sophisticated than fighting: negotiating. District by district, commander by commander, they made arrangements — surrender the post, keep your weapons or don't, keep your life, go home. In many places the local deal was concluded well before any fighting could have started, and the "battle" for a provincial capital consisted of a convoy driving in.

That worked because what the twenty-year project had actually produced was not a national army but a patronage network wearing one's uniform. Loyalty ran upward through personal relationships to commanders, and through commanders to a capital that distributed money and position. It did not run to an abstraction called the Republic. So the Taliban did not have to defeat the network. They had to make it a better offer, and the offer was straightforward: the money is leaving, we are staying, and we will still be here when your family needs to live somewhere.

Read the individual decisions and they are not cowardice. A soldier at a checkpoint in Faryab in the summer of 2021, whose salary had been irregular for months, whose commander had already been on the phone with the other side, who had watched the Americans hand over Bagram in July without informing the Afghan commander until after they had gone, and who understood with complete clarity that no air support was coming and no relief column existed — that man made a correct assessment. Dying to defend a payroll that had stopped arriving, on behalf of a government that had been formally excluded from the negotiations determining its own future, is not brave. It's a category error.

Which is the part that gets skipped in the "they wouldn't fight" framing. They wouldn't fight because the political question had already been settled, and it had been settled in Doha in February 2020, when the United States negotiated its withdrawal directly with the Taliban and left the Afghan government off the paper entirely. That document told every armed man in the country which side had a future. Everything after it was scheduling.

It is also the answer to the objection that ought to be run rather than ducked, because it is the first one a hostile reader reaches for. The Taliban were externally sustained too — sanctuary in Pakistan, money out of the Gulf, a narcotics economy — and they were no Weberian army either, but a network of local commanders with their own patronage and their own side arrangements. By a pure funding argument they should have been every bit as hollow as the force they replaced. They were not. So the provenance of the money is not the mechanism.

What separated the two networks was that Taliban commanders had reason to believe the structure they belonged to would still exist next year, and after February 2020 that belief was underwritten in writing by the departing superpower. Money buys men for exactly as long as it arrives. The expectation of permanence buys them past that point, and it is the one input in the whole program that cannot be procured — only demonstrated, and only by staying.

iii · the precedents, since everyone called it unprecedented

Saigon, 1975. The Army of the Republic of Vietnam was, on paper, among the largest militaries on earth — equipped, trained, and funded by the same benefactor, according to the same input metrics. Congressional aid was cut. Ban Me Thuot fell on March 10. Saigon fell on April 30. Fifty-one days, and the collapse ran faster than the maps could be redrawn, and the postmortems said stunning then too.

Mosul, 2014. Two Iraqi divisions — call it thirty thousand men, rebuilt at a cost around twenty-five billion American dollars — abandoned Iraq's second-largest city to an Islamic State force a fraction of their size, leaving the equipment behind for the incoming side to inherit. Same input metrics. Same sectarian patronage structure underneath the org chart. Same result, and the same word in the headlines.

Tehran, 1979. An imperial army of roughly four hundred thousand, the American client force par excellence, declared neutrality on February 11 and evaporated. Havana, 1958. The White armies of the Russian Civil War, propped on Allied supply lines that eventually stopped. The colonial auxiliary formations of two centuries, which reliably held exactly as long as the metropole's money and never one payday longer.

The template is stable enough to write down. An external power funds a security force to hold a political settlement the local population did not produce. The force is measured in inputs the funder controls. The funder mistakes the inputs for the institution. The funding ends. The force reverts to what it always was — a network of individuals with families and options, now optimizing under new conditions. And the dissolution is faster than the construction by two orders of magnitude, because the construction was mostly disbursement and the dissolution is just people going home.

Nothing about this requires anyone to be treacherous or incompetent, which is why it keeps happening to competent people. It requires only that the thing holding the structure up was never inside the structure.

The honest version of the lesson is unglamorous and will not be learned: legitimacy cannot be procured. It is not a line item, it does not respond to funding, and no amount of training throughput substitutes for it. A state that exists because someone else is paying for it will last exactly as long as the payments, and the interval between the last payment and the collapse is not a measure of how well the state was built. It is a measure of how long it takes to drive to the capital.

Twenty years, two trillion-odd dollars, and ten days. The ratio is the whole finding. Somebody will run the experiment again, and the reports explaining precisely why it failed will be public, indexed, and free — which was never the constraint, because accurate information does not travel along the axis of truth. It travels along the axis of who is paying, and it stops where the money stops. That is the same sentence as the one about the army. It was always the same sentence.