If this is you You got a debt notice from an automated data match, and it's on you to prove the number is wrong.

A State That Can Find You Should Pay for Its Errors

Australia joined its tax and welfare records and raised false debts against hundreds of thousands of people.

A case file: the rule, the person it landed on, and who could correct it.

The argument in 9 lines
If this is you
You got a debt notice from an automated data match, and it's on you to prove the number is wrong.
The case
Australia's scheme averaged annual tax income across fortnights and issued debt notices on the difference, making recipients disprove inferred debts; a Royal Commission found it unlawful.
The claim
A power that takes a burden off people is legitimate only if it pays for its own errors, outsiders can see them, and a trigger corrects them automatically.
The mechanism
Detection moves up to the agency while the cost of detection's errors stays with the recipient, so capacity grows while correction doesn't.
Who pays
Recipients with irregular work, who had to disprove debts the machine inferred, and households for whom a wrong payment means a missed month of rent.
Who can correct it
The regress of contestable correctors has to end in triggers that need no watcher, like a draining bond, a lapsing power, or a second body that can reopen cases.
Whose call it was
A joined record built to pay people is one policy change from billing them, so its direction has to be fixed in law and contestable by those it points at.
What would change it
Put recurring burdens on whoever sees the pattern most cheaply, prefer remedies that need to know less, like presumptions and universal provision, and write correction triggers before capacity runs.
The objection it answers
Building no new capacity because capacity can be abused sounds cautious. Declining to move a burden leaves it on the claimant, the patient and the subpostmaster.
Reading settings

The question up top, the outline, the instrument and the what-next. Turn it off and you just get the essay.

An institution. Both paths open to the person lead back to it. A spotlight holds them.

An Australian welfare recipient with irregular work reported their income to the welfare agency, fortnight by fortnight. Its scheme averaged annual tax income across fortnights, compared the average with what people had reported, and issued debt notices on the difference. Where the averaging was wrong, as it often was for anyone with irregular work, the recipient had to disprove a debt the machine had inferred.

Australia already had the integrated record. The tax office knew what people earned; the welfare agency knew what it had paid them. Joined, those two files are the infrastructure behind the most common remedy for administrative burden: the state already holds the facts, so it should act on them without making anyone prove anything. The Australian government joined them and pointed the result at the recipients. A Royal Commission later found the scheme unlawful.1

A state that can find you should pay for its errors.

Nothing required the data match to send bills. The same join could have found people who had been underpaid and topped them up without a form. The capacity was neutral in construction and partisan in use, and the people on the receiving end had no say in which way it faced.

Any remedy that has the state act on what it already knows needs a capacity like Australia’s, so the Australian case has to be answered. That danger has a name: administrative capacity and corrective capacity can move in opposite directions. A state can get dramatically better at identifying people while staying terrible at correcting its mistakes about them.

Each remedy creates new authority

Presumptive coverage needs someone to write the presumption and police its edges; the PACT Act replaced individual proof of causation with a list, and a list has an author.2 Automatic benefits need records joined across agencies, which is the Australian infrastructure by another name. An institution that detects and repairs its own mistakes needs monitoring, and monitoring that can find the institution’s errors can find yours. A bonded deposit needs an adjudicator who decides when the bond drains. A stop button needs someone who decides when the line restarts. Participatory review needs a standing body, and standing bodies develop interests.

Authority turns dangerous when it can act on people who can’t make it answer, and each remedy creates authority of that kind unless something stops it.

Retreat, building no new capacity because capacity can be abused, fails by the standards the remedies were built on. Declining to move a burden leaves it where it sits, on the claimant, the patient, the subpostmaster, so retreat keeps the old routing and calls it caution. Faith, building the capacity and staffing it well, fails too. A repair office run by decent people is still an institution, and can be wrong about the people it serves in all the ordinary ways.

A burden’s existence says nothing about its address

Any arrangement at scale produces misfit. Rules compress, categories leave remainders, and somebody carries what falls outside. That much can’t be designed away, and pretending otherwise leads to the demand for infallibility, which teaches institutions to hide their mistakes.3

What can’t be avoided is the burden’s existence, and existence says nothing about its address. “Someone must carry it” is true. “So it may as well be you” does not follow. A burden is preventable when an institution could stop it arising, and absorbable when an institution could carry it after the fact. Most burdens of the kind the claimant, the patient and the subpostmaster carry are both. Moving them takes new capacity, so the questions are which capacity and what it does next.

Where a burden belongs

Who sees the error first and most cheaply? Tort law has asked a version of this for decades under the name of the cheapest cost avoider: put the loss on whoever could have prevented it most cheaply.4 Applied to institutional error, the principle splits in a useful way. A person sees their own case first. Only the holder of all the files can see that ten thousand cases share a shape. Each Australian recipient saw a wrong number, while the agency could see the method behind all of them.

Does the burden recur across a class? A burden that recurs is a property of the rule, and the rule’s author can see it more cheaply than anyone living under it. An idiosyncratic burden, the case no category anticipated, is often cheapest for the person to flag, since only they know the fact. Recurrence is what moves a burden upward. The individual’s part shrinks to reporting what only they can know, and the institution carries the rest.

Who can bear the cost with least damage to a life? An agency spreads a wrong payment across a budget. A household pays for it with a missed month of rent. Taken alone, this question would move every cost upward, so it comes third and settles what the first two leave open, tipping the cost toward whoever would lose money and away from whoever would lose a month.

The Australian scheme raised capacity and pushed its errors down

Moving a burden upward creates a power to carry it, and at this step the test turns on itself. The same three questions apply to the new power’s errors.

The Australian scheme fails here. Detection moved upward: the agency, and no longer the recipient, found the discrepancy. The cost of detection’s errors stayed below: the recipient had to disprove the average. Capacity went up and its error cost went down. Call that the signature of a corrector that won’t be corrected. The pattern appears again in Britain, where the Post Office investigated and prosecuted its own subpostmasters on the word of its own accounting software, and afterward ran a compensation scheme for the harm.5 An institution that investigates, prosecutes and compensates for the same error has been handed every capacity a remedy needs and none of its conditions.

The recursion produces a rule. A new capacity is legitimate only if it pays for its own errors, outsiders can see them, and a trigger corrects them without relying on the people it harms to stay vigilant.

The last clause matters because keeping watch is itself labor. If correcting the corrector falls to its victims, the maintenance ledger has moved up a floor and nothing else has changed. A correction mechanism that can’t itself be contested is a new sovereign, so every corrector has to be contestable, and that means accepting a regress. A chain of contestable correctors still depends on someone doing the contesting. So the regress has to stop in mechanisms that act without anyone keeping watch: a bond that drains when a threshold is crossed, a power that lapses unless renewed, a second body with standing to reopen the first body’s cases. A trigger like that needs no hero to pull it. It needs to be written before the capacity starts running.

Prefer the remedy that needs to know less

Run this way, the recursion also ranks the remedies. Some move a burden upward by making the institution know more about people. Others move it by making the question unnecessary. A presumption that anyone who served in a given place and time is covered needs little investigative capacity. Universal provision needs less still, since an institution that refuses no one has no use for the data that would justify refusing. Proactive monitoring sits at the far end: joined records, pattern detection, and authority to act on what it finds.

Where two remedies would move the same burden, prefer the one that creates less power to know. Ownership raises the matching point: collective provision still needs institutions that classify, and the classification is where domination can live on after the deeds change hands. A tool built for one purpose also gets applied to the next without anyone authorizing it. A joined record built to pay people automatically is one policy change from billing them automatically, and billing is its likely fate unless its direction is fixed in law and contestable by the people it points at.

Who applies the test and how to weigh its costs stay open

The test needs someone to apply it. Whether a burden recurs across a class is itself a classification, made by an institution with interests. The test moves the question of who decides without answering it.

The third question compares costs that share no unit. An agency’s budget line and a household’s month are both costs, and the test leans toward the household without saying how far.

Some correction has to happen faster than anyone can check it. A stop button that needs approval before it halts the line is a request form. For authority that’s only useful when unchecked at the moment of use, the test has nothing cleaner to offer than heavy bonding and review after the fact.

The test still catches the most common failure early. After the Dutch childcare benefits scandal, the government set up a dedicated organization to repair the harm, and many families found themselves waiting on it much as they’d waited on the tax authority.6 The repair office is a corrector if it sees the pattern before the families do, carries the cost of its own delay, and can be reopened by someone other than the families. If it fails those tests, the families have been given a second tax authority.

Notes

1

Royal Commission into the Robodebt Scheme, Report (Commonwealth of Australia, July 2023), which called the scheme “a crude and cruel mechanism, neither fair nor legal.”

2

Sergeant First Class Heath Robinson Honoring our Promise to Address Comprehensive Toxics (PACT) Act of 2022.

3

Sidney Dekker, Just Culture: Balancing Safety and Accountability, 2nd ed. (Ashgate, 2012), argued that responses to incidents seen as unjust breed secrecy and self-protection and choke off the reporting of failures that safety depends on, and that what matters is who gets to draw the line between an honest mistake and a culpable one.

4

Guido Calabresi, The Costs of Accidents: A Legal and Economic Analysis (Yale University Press, 1970).

5

Bates v Post Office Ltd (No 6: Horizon Issues) [2019] EWHC 3408 (QB). The Post Office set up and ran the Historical Shortfall Scheme (later renamed the Horizon Shortfall Scheme) in 2020, after the group litigation settled.

6

Parliamentary Interrogation Committee on Childcare Benefits, Ongekend onrecht (Tweede Kamer der Staten-Generaal, 2020); Nationale ombudsman, Wachten op herstel, hoe lang nog?, report 2022/122 (July 2022), on the delays at the Uitvoeringsorganisatie Herstel Toeslagen.

When they get you wrong or count on you

123
Who pays
Recipients with irregular work, who had to disprove debts the machine inferred, and households for whom a wrong payment means a missed month of rent.
Who can change it
The regress of contestable correctors has to end in triggers that need no watcher, like a draining bond, a lapsing power, or a second body that can reopen cases.
Whether it was theirs to decide
A joined record built to pay people is one policy change from billing them, so its direction has to be fixed in law and contestable by those it points at.

Continue reading

What to read next

This essay belongs to the Conditional authority pattern; its toolkit, topic and guide are one step away.

Related essays

New essays by email

One situation a week, the essays that explain it, and one question to put to the institution.

New essays by email

One situation a week, the essays that explain it, and one question to put to the institution.