What actually repairs it

Part III of three. A market where the buyer cannot check what he is told prices everyone at the level of the worst, and the honest supplier pays for it. So what fixes that? A good deal is known — including a large body of evidence about what has failed, and one experimental result that contradicts the answer I expected to give.

Two sentences of recap, for anyone arriving here first. When a buyer cannot verify a claim at the moment he is told it, he pays the price of an average supplier, which means real quality goes unpaid for and the good sellers withdraw — a mechanism George Akerlof described in 1970 and won a Nobel for. Its victims are not the buyers, who simply pay less and take their chances; they are the honest suppliers, and above all the new ones, who inherit a reputation they did nothing to earn.

I should say plainly that I make my living in this trade, so a conclusion that the answer is verification is a conclusion that suits me. Take what follows on its evidence or not at all. I ran garment factories in Bangladesh for seven years and I am now on the other side of the same asymmetry, receiving offers I cannot check.

First, notice that nobody measures it

Before asking what repairs the problem, it is worth seeing how little anybody has tried to.

The largest annual survey of what American fashion companies weigh when choosing a sourcing country asks them to rate each country on nine criteria: speed to market, sourcing cost, flexibility, minimum order quantity, vertical integration, and the risks of labour and social compliance, environmental compliance, geopolitics and import barriers.

There is no criterion for reliability. None for trust, credibility, quality consistency or delivering on time. Not this year, and not in the three editions before it — I checked the words. And it is not an oversight: the designers revised the list twice in three years, adding four new criteria, and never added one. Note too that the nearest thing on the list, "speed to market", is defined as lead-time length, not lead-time adherence. Duration is measured. Keeping your word is not.

The best-known consultancy survey of apparel procurement chiefs has the same shape: no entry for reliability in its ranked list of sourcing considerations — nor, oddly, for price — although the prose of the same report says three-quarters of respondents prioritise suppliers on reliability and performance. The buyers know it matters. Their instrument does not ask. And when somebody does ask, it comes first: an inspection company surveying six hundred and fifty businesses about their 2024 found that "keeping to production and shipping schedules" was the hardest operational problem in their supply chains, named by 43 per cent — ahead of managing costs at 34, and rising to 55 per cent among European businesses.

And for Bangladesh specifically, I could not find a measured on-time-delivery rate anywhere. Not from the ILO, not from the World Bank, not from the factory-improvement programmes that have operated there for years. The World Bank explains why in print — apparel lead-time data "are difficult to obtain at the country level because the definition often differs" — and then falls back, in its own study, on asking buyers for their impressions of which countries are reliable.

One correction before this goes too far, because a stronger claim is available and it is not true. When buyers are asked what holds them back from Bangladesh, they do not say they distrust it. They name social and environmental compliance risk, and the risk of future trade barriers — and they rate Bangladesh's price its single greatest strength. Nobody is refusing Bangladesh for dishonesty.

The narrower claim is the one that survives, and it is strange enough. Reliability in Bangladeshi garment sourcing is worth something like ten to fifteen per cent on a supplier's markup — it is measurably priced — and essentially nobody measures it. A quality that is valuable and unobservable at the moment of decision is not an unfortunate gap in the statistics. It is Akerlof's condition exactly, and the whole trade is maintaining it together.

What the literature says to build

Akerlof did not stop at the diagnosis. His last section opens: "Numerous institutions arise to counteract the effects of quality uncertainty." He names four. Guarantees, whose real function is to move the risk onto the party who actually knows. Brand names, which both indicate quality and give the customer a means of retaliation, since he can stop buying. Chains, which he notes cluster on interurban highways and serve people who are not local: a chain does not sell excellence, it sells predictability to somebody who cannot judge for himself. And licensing and certification — doctors, lawyers, barbers, diplomas.

Do they work? Sometimes, measurably. When Los Angeles County made restaurants display hygiene grade cards in the window in 1998, inspection scores rose, customers became sensitive to hygiene, and hospitalisations for foodborne illness fell — and the researchers showed the health gain was not merely diners switching restaurants. The restaurants got cleaner. In a controlled experiment on an online marketplace, a seller's established identity was worth 8.1 per cent more than the identical goods sold under a new name he also controlled.

And what fails, which is the more useful half

The failures have a structure, and the structure is always the same: who pays the person doing the checking.

A two-year randomised experiment on environmental audits in Gujarat found that under the ordinary arrangement — the audited firm picks and pays its own auditor — reports were systematically filed just below the legal limit when true emissions were higher. Not random error. Readings pinned neatly inside the line, which is the signature of a purchased pass. When the payment structure changed, auditors reported honestly and the plants actually cut emissions. The same finding turns up in supply-chain auditing: across some seventeen thousand supplier audits at nearly six thousand suppliers, auditors reported fewer violations when the audit was paid for by the supplier being audited, and fewer when the same auditor had visited before.

There is also a limit that has nothing to do with anyone's honesty. An auditor described his own position to a researcher like this: "you have no powers of search so you cannot open a locked drawer… you can look at a record that says something but you wouldn't be able to go and find out whether it's actually true." That is not a complaint about factories. It is a description of an instrument being asked to do something it was never built to do.

And the hardest result of all: in the Chinese dairy case, the firms that government inspectors examined and pronounced clean did no better in export markets than firms nobody inspected. The certificate existed. Buyers did not believe it, so it bought nothing.

The correction I did not expect

At this point I have to report something that cuts against where this was going.

The cleanest experimental test of what actually fixes a market like this ran nine hundred and thirty-six participants through it, and its conclusion is blunt: liability has a crucial effect; verifiability, at best a minor one. Letting sellers build reputations "has little influence". Competition drives prices down and does not make the market work better "as long as liability is violated."

Read against everything above, that is not a contradiction so much as a correction of emphasis. Being able to check a claim is necessary. It is not sufficient. What makes checking bite is that something follows from it — and that is the thread running through every case here that ended well. Los Angeles did not encourage restaurants to self-report; it required a letter in the window. The Gujarat auditors told the truth when the payment structure made truth the thing they were paid for.

The one that worked, and it is in Bangladesh

After 2013 a binding agreement put independent engineers into Bangladeshi garment factories on exactly those terms: the inspectors are not chosen or paid by the factories they inspect, and what they find is published factory by factory. As of 30 June 2026 it covers 1,762 factories and 2.8 million workers for 245 brands, and reports 75,198 inspections, 288,915 safety findings, and an average remediation progress rate of 81 per cent.

But the inspections are not what made it different, because every scheme inspects. What made it different was a clause: disputes could be taken to binding arbitration, enforceable in a national court under the New York Convention. It was not decorative. It was used twice, in 2016, and settled — one of the two for US$2.3 million, most of it spent remediating more than a hundred and fifty factories. A voluntary body set up alongside it, by other brands, in the same country, in the same year, described itself in its own founding documents as an association formed to further its members' "common business interests", and bound its members to nothing but their fees.

Whatever anyone thinks of how that came about, it is the largest working instance in the world of the thing this argument has been circling, and it is in Bangladesh. Which makes the gap easier to see, not harder. On Bangladesh's public register 1,443 factories — more than four in ten — have no third-party inspection listed at all, and among the smallest they are very nearly all of them. The instrument exists, it works when it has teeth, and it has never reached down to the tier that most needs a way of being believed.

Nor is any of this charity. Measured across more than two thousand manufacturing establishments in thirty-six countries, suppliers who achieved compliance with a buyer's social standards saw their annual orders from that buyer rise about four per cent — an effect the researchers found was driven largely by apparel. Being checkable is worth money. It is simply worth less than it costs to a factory with ninety people.

The tier where none of this reaches

Take a factory that has won its own order and does not own its dyeing. It has a cutting section, sewing machines and a boiler. The yarn, the knitting, the dyeing and the finishing all happen in other companies' buildings. So the question that decides whether it can keep its word is never sewing. It is fabric — and it is worth walking one order through.

Knitting. The weight of a knit — its grams per square metre — is set principally by the stitch length, and stitch length is set on the machine. A 2023 study in Tekstilec describes the state of the art without embarrassment: the absence of a system able to monitor yarn-feed uniformity "has led to the use of experimental methods which are dependent on skilled operators", and when something drifts, "the equalisation is done by the operator using the trial-and-error method, which consequently increases the risk of human error." A 2002 paper in the Textile Research Journal found that "the actual stitch length in the fabric measured after knitting is always less than the stitch length required." The first number in the specification is produced by a person's judgement on somebody else's machine.

Dyeing, where it is actually decided. The usual reassurance is the lab dip — a small sample dyed to the colour and approved by the buyer. A 2006 study in Coloration Technology asked whether a pilot machine can be set to reproduce bulk conditions, and the answer deserves quoting exactly: some parameters can be, "whereas others cannot. It was impossible, for example, to simultaneously achieve appropriate values for the nozzle flow velocity and the dyebath circulation time."

That is not a complaint about anyone's dyehouse. It is a finding that the small trial and the large run are not the same physical event. The trade knew it long before the paper: a 2011 review records that because non-reproducibility was so common, "the dyers subtracted 10-15% from the quantities given in the laboratory recipe before proceeding to mass production" — a standing correction for a gap everyone worked around. Nor does it settle with practice. A 2022 study put it flatly: "achieving the same shade every time is still a very challenging process. The fabric was dyed with the same recipe after some duration for the same shade but it showed more variation in the shade than the previously dyed fabric."

And there is a reason a small order is worse. Industrial dyeing machines are large — one maker advertises 200 to 550 kg a tube and up to four tonnes a batch — and their published performance assumes a full load. The European Commission's reference document on textile processing states the consequence: machines are often underloaded, which "often occurs in commission companies where a high production flexibility is required to serve variable lot sizes according to customer's demands", and below roughly sixty per cent of nominal capacity "the real liquor ratio will differ greatly from the nominal liquor ratio". That is chemistry, not bookkeeping. Salt is dosed per litre, so the ratio decides what the fibre actually sees — the same document works the example, a tonne of fabric taking 1,200 kg of salt on one machine type and 300 on another — and adds the sentence that closes the loop: "the lower the salt concentration, the more sensitive the system becomes to any change in parameters that influence exhaustion."

So a small lot is not merely dearer per kilo. It is run in a machine it cannot fill, at a real liquor ratio nobody quoted, in a bath that is more sensitive precisely because the lot is small.

Finishing, where the width and weight are meant to come right. A paper published this year on cotton/elastane knits tracked them moving stage by stage — weight factors from 0.95 to 1.08, width factors from 1.00 to 1.10 — and described its own contribution as converting "traditional empirical process control into a numerical and repeatable predictive method". Which tells you what the normal method is.

The setting is also thermal, and that matters more in Bangladesh this year than usual. Heat-setting an elastane blend runs in a gas-fired stenter inside a narrow window — Cotton Incorporated puts it at 182 to 196 °C, above which the fabric loses power as the elastane degrades. In August 2026 that gas is not reliably there. Trade reporting describes dyeing and finishing as the worst-affected step: against the fifteen PSI said to be needed for uninterrupted running, pressure "has come down to 1 to 3 PSI, in many cases almost zero." A national daily quotes one mill dyeing two tonnes a day against a normal twelve to fourteen; in one textile town more than a hundred factories stopped. And the clause that matters here: erratic gas pressure "is causing heavy losses in dyeing operations, where interrupted production can ruin entire fabric batches."

And none of it is happening in the factory that took the order. It is happening in a queue, and the queue is short of counters: Bangladesh's textile millowners' association counts about 345 dyeing-printing-finishing members against several thousand garment factories. In 2022 the managing director of one dyeing group put his order book plainly: "We are overbooked with orders from local garment exporters for next four months."

Where sending work out is the norm, the cost of standing in that queue is measured. A study of three hundred knitwear exporters in Tirupur found 267 of them doing job work — and ninety per cent of them missing their schedule, half by ten days or more. Its explanation is one sentence: most of the process took place outside the factory, and the unit doing the final stage was "responsible for both quality and delay."

Priority inside somebody else's mill is real and rationed. A study of the same cluster found that finishing units "under the owner's control have priority access", and that a firm enjoys "a definite priority" where the dyehouse is run by family or friends — privileges that firms outside those networks "do not enjoy". A Turkish mill states its own split on its website without any embarrassment: "50% of our dyehouse capacity is set aside for fabrics we produce ourselves; the remaining 50% is allocated to the fabrics of the customers we serve on commission." None of that is misconduct. It is what any business does with capacity it owns, and a European reader should recognise it immediately.

The World Bank wrote the conclusion in 2016, describing this exact chain: apparel factories must send unfinished fabric out to be dyed and finished, and "it is not uncommon for factories to discover the fabrics are damaged, meaning that the factory must replace the fabric, take the loss on the initial purchase, and ship the product late." Then the last clause, which is this whole argument in eight words: "the apparel supplier will be viewed as unreliable."

One correction, because I had it wrong myself until I checked. The very smallest units are not in this queue at all. A 2018 Bangladeshi case study of a two-line subcontracting factory found it sourcing "all of its raw materials such as fabrics, thread, accessories and chemicals from the domestic market" — buying cloth already dyed, from the wholesale trade in Old Dhaka. It escapes the queue by never entering it, and pays for the escape by taking whatever colour the merchant already has. The firm that actually stands in the queue is the one in between: large enough to specify a colour, too small to own the vessel that dyes it.

What that does to the argument

Now put that beside the model this piece has been running on.

Akerlof assumes the seller knows the quality and the buyer does not — and every remedy that follows is a device for making the seller reveal, or answer for, what he already knows. Guarantees, brands, certification, liability: all of them work on a man who is holding something back.

At this tier he is not holding anything back, because he does not have it. He cannot tell you the bulk will match the lab dip, because the literature says the small machine cannot reproduce the large one. He cannot tell you the shade will repeat, because the same recipe does not reliably repeat. He cannot tell you the fabric will arrive on time, because it is in a queue in a building he does not own, behind work with a priority he cannot buy. And this month he cannot tell you the gas will be there.

That is not asymmetric information. It is an absence of information, and it changes what honesty is capable of. An honest man in that chain still cannot tell you what will happen. He can tell you what he intends, what he has done before, and who he buys from. He cannot tell you the thing you actually want to know.

So the asymmetry has not gone. It has moved. What the buyer cannot distinguish is no longer the good factory from the bad one — it is a promise backed by control from a promise backed by hope. Both arrive in the same words. Of two men who say "yes, on time", one is describing his own production plan and the other is describing his intentions, and nothing in the message says which is which.

Which also disposes of the certificate at this tier, and it should be said plainly. A certificate says a mill can be audited. It does not say whether your eight hundred kilos will run before their eight tonnes.

And it is why verification — the thing this piece has spent two thousand words defending — is not sufficient here. You can verify a factory completely and honestly, its machines and books and people and record, and have verified almost nothing about the outcome, because the outcome is decided in three other companies. Verification tells you what a factory is. It does not tell you what a queue will do.

The obvious answer, and why nobody on earth has it

If the constraint is priority rather than capacity, and priority is bought with committed volume, the answer looks obvious: small firms should combine and buy it together. A shared dyehouse. A booked slot. I went looking for that, fully expecting to point at one.

In the districts where it should exist, it does not. Not in Tirupur, Surat, Ludhiana, Panipat or Erode; not in Bursa or Denizli; not in Faisalabad; not in Vietnam; not in Bangladesh. Clusters share effluent treatment, which is genuinely collective and well documented. They share estates, land, gas and water. Wet processing itself is everywhere bilateral — one buyer, one dyer, one lot at a time. Capacity reservation contracts are ordinary instruments in pharmaceuticals and semiconductors; in textile dyeing they do not exist. Searches in three languages for a co-operatively owned dyehouse return nothing at all — the phrase has no established meaning, which is usually what it looks like when a thing has no instances.

Prato is the case that settles it, because Prato has every advantage. Seven centuries of textile trade, firms within walking distance of each other, one language, functioning courts, and a district-wide electronic network built in the 1980s. Its famous middleman, the impannatore, is defined in the Italian national encyclopaedia as a cloth manufacturer "who does not have a factory of their own" — he buys production phases on his own account. The standard model of how the district actually works describes him attaching a dyeworks by proximity and momentary availability: "the middleman looks around for suitable firms in order to build a production chain." There is no booking in it, and no queue to hold a place in, because subcontracting there is non-exclusive — the dye firm serves many customers and owes a slot to none of them. Prato's own answer to fragmented dyeing, when it finally came, was not to share dyehouses but to merge them: Gruppo Colle was formed in 2003 out of three separate companies. The largest dyeing district in China did the same thing, cutting from 212 printing-and-dyeing firms to 109. Both made the small buyer's queue shorter by making the queue-holders fewer, which is not the same as making it easier to join.

There is exactly one large working instance of shared dyeing on earth that I could verify, and its shape tells you why. In Gaoyang County in Hebei, the programme's own name for its method is the "five concentrations", and capacity is the first of them, ahead of effluent. Thirteen dyeing firms were moved into a single park carrying all of the county's dyeing capacity; the operator opened access to small and micro enterprises across the county and installed fifteen-kilogram sampling vessels so that a trial dye would cost hundreds of yuan rather than thousands. It now serves more than 1,800 weaving firms.

Read what that cost. Every dyehouse in the county was moved by the state, so nobody could defect; and the operator carries a duty to serve, so access is not negotiated. It is not small buyers clubbing together to buy priority. It is the concentration of supply, plus an obligation — and it took a government to do either.

What the market does instead, everywhere else, is put a merchant in the way. In Old Dhaka the fabric wholesalers hold dyed cloth so that the smallest units never enter the queue at all; someone takes title, carries the inventory risk, and charges for it. Aggregation does happen. It happens after the fact, on a trader's balance sheet, and the trader is paid for it — which is a more sympathetic account of the middleman than this trade usually offers, and I think a truer one. His margin is the price of not standing in the line.

And it should be said plainly, because the cheap reading of everything above is that this is a Bangladeshi failure. It is not. It is unsolved in Tuscany.

Seventeen years

Austria's answer is the one worked example that ran the whole distance. It rewrote its wine law to demand, in its own marketing board's German words, "eine lückenlose Prüfung der Weinbestände" — a gapless testing of the wine stocks — so that a quality wine now carries a state test number and must pass both a chemical analysis and a blind sensory panel before it may be sold as what it says it is. (The board's English page renders that as testing "at every single step in the wine production process", which claims more than the German does; where the two differ I have used the German, since that is the original.) Then, in 1986, it founded a national body whose stated job was to rebuild the reputation, because a collective injury cannot be repaired by individuals: no honest grower could have bought his way out alone.

It worked and it took a generation. Austria exported 64 million litres in 2024, and four fifths of that volume and 94 per cent of the value left the country in bottles rather than in tanks. A country that had been selling anonymous bulk came back selling wine with a number on it.

So the repair is never the promise and never the document. Where it worked, two things were true, and the second is the one everybody forgets: somebody looked who was not paid by the party being looked at — and something followed from what they found. A state number on a bottle. A letter in a window. A clause that can be taken to arbitration and enforced in a court. Where the looking happened without the consequence, the paperwork accumulated and nothing moved: the auditor with no power to open a drawer, and the government clearance that bought its holders nothing at all.

But the piece has to end one step short of where I wanted it to. Everything above works where somebody knows and will not say. At the bottom of this trade the problem is a different one — nobody knows, because no single party controls enough of the chain to know — and there the record is that it has not been repaired anywhere, and never by honesty or by co-operation. It was repaired once, in one Chinese county, by moving every dyehouse into one park and imposing a duty to serve. Everywhere else the answer has been to buy the dyehouses, or to pay a merchant to stand in the queue for you.

And one last thing, which is the most useful of all. The penalty falls on a whole origin only for as long as the buyer cannot tell its members apart. Where he can name who did what, he names them, and stops. Distinguishability is the variable — not virtue, not price, not compliance. That is a hard thing to build and a cheap thing to say, and I would rather end on the hard version.

Because what the honest factory is asking for is smaller than it sounds. It is not asking to be trusted. It is asking to be checked — and finding that nobody has built the thing that would do the checking at a price it can pay. And at the bottom of the trade it is asking for something harder still, which no certificate and no contract has ever supplied: to be able to know, itself, what it is promising.

— A.K. · TB Textile Sourcing · Slovenia, EU
(European who ran knit factories in Bangladesh for seven years, and is now on the buying side of the same asymmetry. Every figure above is published; sources are named in the text.)