The Rule Works. She Doesn't Fit.

A benefits category handles nine thousand claims a year, and handles them well: fast decisions, consistent outcomes and an audit trail a regulator can walk end to end. But the story here is one woman whose income shows up in seasonal lumps the eligibility window can’t see. Her employer keeps no records the portal will take, and her file has now been opened and closed four times without anyone at the agency doing anything wrong.

Nothing about her is fraudulent, and nothing about her fits. She carries the remainder, meaning the part of reality a rule’s picture of the world has no room for.

In 2020 the English Court of Appeal heard four single mothers whose pay the eligibility window kept misreading. Universal Credit counts earnings in monthly windows. When a weekend or bank holiday moved payday forward, two salaries could land in one window and none in the next, which cut one month’s award and cost them the work allowance in the other. The court accepted that counting actual receipts suited the great majority of claimants, and held that refusing an exception for a group that might number 85,000 was irrational.1

Compression is the whole point of a rule

Nobody writes a rule to fail. You write a rule so it can be applied nine thousand times without nine thousand judgment calls. Eligibility categories squash circumstances into boxes. Database schemas squash lives into fields. Job descriptions squash work into titles. Metrics squash outcomes into a number. A committee squashes an argument into text that has to mean one thing by morning. Each one holds for most cases and drops the rest. A representation rich enough to hold every case would have to be renegotiated case by case, and then there would be no rule.

The remainder isn’t evidence of bad design. A careful rule and a careless rule both produce one. The careful rule produces less and tracks where the leftover cases go. The mismatch is structural. The only choice is where the mismatch lands, and whether anyone says so out loud before it does.

Diagram

Nine thousand claims go through, and one woman does not

View source

%% title: Nine thousand claims go through, and one woman does not
%% caption: The eligibility window decides nine thousand claims fast and books them as resolved. The woman whose income arrives in seasonal lumps falls outside it and does the reconciling herself, unpaid, until her file closes for the fourth time.
flowchart TD
  classDef actor stroke-width:1.6px;
  classDef system stroke-width:1.2px;
  classDef gate stroke-width:1.6px;
  classDef burden stroke-width:2px;
  classDef failure stroke-width:2.2px,stroke-dasharray:6 3;
  classDef repair stroke-width:2.4px,font-weight:bold;

  P["Benefit claims filed this year"]:::system --> C{"Eligibility window<br/>Does the income fit?"}:::gate
  C -- "fits the window (9,000 claims)" --> H["Decided fast<br/>Audit trail a regulator can walk"]:::system
  H --> L["Agency Ledger: '100% Resolved'"]:::repair
  C -- "does not fit (1 woman)" --> R["The remainder<br/>Seasonal lumps • No records the portal takes"]:::burden
  R --> W(["She reconciles it, unpaid"]):::actor
  W --> D["Same facts, new clerk<br/>Document hunts • Daughter on hold"]:::burden
  D --> F(["File closed a fourth time<br/>Right unrealized"]):::failure

Something has to move: the rule, the organization, or the person

When a rule meets a case it can’t hold, something gives. Sometimes the rule gives through an exception, an appeal or a redesign. The organization may give instead, in staff time, a discretionary budget or a case review, or the software may make up the difference. And sometimes nothing gives, and the person does.

She waits. Reapplies. Explains the same facts to a different clerk. Hunts for documents nobody naturally keeps. Learns the agency’s categories well enough to translate her own life into them. Does composure at the counter. Asks the one decent employee to bend something. Recruits a daughter to sit on hold. Builds a workaround and maintains it, unpaid. Takes the worse service because the right one is three appeals away.

Every one of those is labor, paid in time, risk, cognition and feeling, the full bill for reconciling an institution’s abstractions with a life they don’t describe. Research on administrative burden has shown that these costs decide whether a nominally available program is available at all, and that where they land is a political fact.2 The finding travels well past the welfare office. A hospital runs on staff patching coordination failures the org chart can’t see. A platform runs on users fixing records the schema mangled. A household runs on one person noticing and compensating for everyone else. Wherever an abstraction meets a life it didn’t model, the difference becomes somebody’s job.

Who actually pays when the rule fails

Economists stopped confusing the person who hands over a tax with the person who carries it. A levy on the seller can land on the buyer, and a levy on the employer can land on the worker. The statutory payer is a formality, and you find the real one by watching where the price moves. Applied to institutional error, the same discipline asks who pays when the rule fails, and almost no institution asks that of itself.

The payer might be the treasury, the agency, the vendor, management, the frontline worker, the claimant, the caregiver, the family, or nobody, because the right just went unrealized. Those are different answers and the paperwork doesn’t distinguish between them. The denial letter says resolved and the record says processed. Either way, the account stops at the system’s boundary. The real account keeps going, in unpaid leave and missed wages and a kitchen table at midnight and the specific exhaustion of proving, for the fourth time, a fact the institution could have known.

The political economy of rules lives in the gap between where a failure formally sits and where its costs actually land.3 An institution can externalize its representational failures the same way a factory externalizes smoke: keep the abstraction, price the cleanup at whatever the absorber will tolerate, book the savings. The burden is one of the system’s outputs. The category’s speed and cheapness are real, and they were bought with the claimant’s month, the daughter’s afternoon, the user’s unpaid repair. The agency’s ledger records the rule as working, and the unpaid labor that keeps the record accurate appears nowhere in it.

The remainder statement

Federal agencies have estimated the burden of their own forms since 1980. The Paperwork Reduction Act makes every information collection carry a number: how many respondents, how many hours each, what the hours are worth. Added together for fiscal 2023, the numbers came to 10.5 billion hours, nearly two-thirds of them at the Treasury, in an accounting OMB published under the title Tackling the Time Tax. The estimate is sometimes wrong, and GAO has found its errors running both ways. Even so, the burden estimate is the most honest document most rules ever produce, an admission on the record that the rule costs the people subject to it something. The estimate counts only the hours of the people who fit.4

A rule owes the same admission about the people who don’t. That admission is a remainder statement, five lines filed with the rule before the rule runs. What the category drops, named as circumstances rather than as an error rate. How much to expect, as an estimate, published so it can be wrong in public. Who carries the dropped cases, meaning the role assigned to reconcile each one, with the reconciling counted as work. At whose expense, meaning whether the institution pays for the reconciling or the person does. And a ceiling, meaning the volume of remainder at which the estimate is declared blown and the rule, not the cases, goes under review.5

Diagram

Five lines the agency files before the first claim

View source

%% title: Five lines the agency files before the first claim
%% caption: Before the eligibility window runs, the agency writes down what it will drop, roughly how many claims that is, who reconciles them and on whose budget. Past the ceiling, the rule goes under review instead of the woman’s fifth file.
flowchart TD
  classDef system stroke-width:1.2px;
  classDef gate stroke-width:1.6px;
  classDef binding stroke-width:3px,font-weight:bold;
  classDef failure stroke-width:2.2px,stroke-dasharray:6 3;

  subgraph Statement ["The Remainder Statement (Filed with Rule)"]
    direction TB
    L1["1. What it drops<br/>Seasonal lump income, records the portal refuses"]:::system
    L2["2. How many<br/>Estimated 3.2% of claims, published"]:::system
    L3["3. Who reconciles<br/>A paid agency case worker, not the claimant"]:::system
    L4["4. Who pays<br/>The agency budget, not her unpaid hours"]:::binding
    L5["5. Ceiling<br/>Past 5%, the rule is reviewed, not her fifth file"]:::failure
    L1 --> L2 --> L3 --> L4 --> L5
  end

The statement leaves the rule exactly as accurate as before and changes what its failures are. Without the statement, each of the woman’s four files is an anomaly, and anomalies are hers. With the statement, the fifth file is a line item over the estimate, and a blown estimate belongs to whoever wrote the rule. Incidence, which the paperwork was built to lose, gets written into the paperwork before the first case shows up.

A rule unable to produce the statement still has a remainder. Left undeclared, it lands on whoever cannot refuse. Once declared, the statement can be checked by the people who carry the remainder, and they can falsify it in the one direction that matters. They can compare what was declared against what they’re actually carrying. For the first time, their ability to make the mismatch count has a document to attach itself to.

Who is made responsible, and with what power

The usual questions are “is the rule correct on average” and “is the category fair across the population it was drawn for.” Both treat the nine thousand as the institution and the woman as rounding error. Treat her as a party and the question flips. For this rule, who got made responsible for what the rule cannot represent, and what power do they have to make the mismatch count?

Held against the question, arrangements with identical intentions split three ways. Some declare the remainder and distribute it on purpose: paid coordination, discretionary budgets, staffed appeals, records fixed on the institution’s dime. Some assign it by default, to whoever can’t say no. And some do something subtler. They keep every rule intact by letting a reliable few absorb all the variance, and then they mistake the absorption for competence.6