Between 2011 and 2014 Britain’s Department for Work and Pensions moved people off the old incapacity benefits and onto Employment and Support Allowance. The law required it to check each person for both parts of the new benefit. The conversion often checked only one, so people owed the income-related part, which carries the disability premiums, went without it.1 By June 2013 staff were finding individual cases and correcting them. In June 2014 the department told its decision-makers to stop making the error and to put old cases right if they came to light. It did not tell them to look for the old cases. In June 2015, on an appeal brought by a claimant identified as DJ, the Upper Tribunal held that a conversion had to consider both parts and that one which hadn’t should be revised as the department’s own error. The ruling prompted discussion and no clear action. The department set out to find everyone converted the same way only in 2017, and by early 2018 it estimated that about 70,000 people had been underpaid, by around £5,000 each on average.2 DJ’s appeal was heard, and it was right.
One claimant won while 70,000 others stayed underpaid.
It was filed at the wrong size.
Voice that climbs no rung is intake
Voice machinery measures expression. It doesn’t measure consequence, and it can be built to avoid measuring it. An arrangement can be extraordinary at collecting the experiences of the people it fails and stay perfectly autocratic. Contradiction often dies on the way in. Knowledge that survives intake still has to be allowed to do something.
Consequence comes in degrees, and the degrees form a ladder. From weakest to strongest, its rungs are an exception granted, a justification forced, a review opened, a pattern aggregated across cases, a resource moved, a rule reconsidered and a design changed. Voice that climbs none of them is intake, whatever it gets called.
The first two rungs are the size of a case. An exception is granted to a person, and a justification is owed to a person. From the third rung on, the object isn’t the case anymore. A review looks across cases, and a pattern is assembled from them. A resource, a rule, a design belong to the class the rule was written for. An individual case can establish a precedent or trigger a wider review. Where the remedy reaches only the claimant’s file, the other cases remain untouched.
The unit of standing
A rule is the size of a class. An appeal begins with a person. When the appeal can correct only that person’s file, winning it leaves the rule available to produce the same error again. The mismatch protects the rule without anyone having to refuse the appeal.
Standing only works at the size of the thing it contests. Standing to reopen your own file is real, and it tops out at the exception. The file is the biggest object that standing can reach, and the rest of the seventy thousand aren’t in it. To reach the rule, standing has to be able to hold them together and file at the rule’s size. A precedent that requires every affected file to be reviewed can cross that boundary. A remedy confined to one file cannot.
The rule can be defended without anyone refusing an appeal. Complaints are handled one at a time, and settlements are individual and confidential. Appeals are docketed per file, with no discovery across files, and the complainant is asked for her facts and never anyone else’s. Each of those is reasonable on its face. Together they keep every claim person-sized, and so below the rung where the rule lives. The waiting room is arranged so the people in it can’t become a party.
View source
%% title: The unit mismatch and categorical downscaling funnel
%% caption: An institutional contradiction exists at the scale of the class, but intake funnels it down to an atomized individual appeal—keeping the rule intact through per-case settlements.
flowchart TD
classDef system stroke-width:1.2px;
classDef gate stroke-width:1.6px;
classDef failure stroke-width:2.2px,stroke-dasharray:6 3;
classDef repair stroke-width:2.4px,font-weight:bold;
classDef binding stroke-width:3px,font-weight:bold;
C["Systemic Defect: Rule Size<br/>ESA conversion underpays about 70,000 claimants"]:::system --> F{"Intake Architecture"}:::gate
subgraph Atomized ["Downscaling Funnel (Individualized Appeal)"]
direction TB
A1["Per-file docket & confidentiality"]:::failure --> A2["One-off correction for 1 claimant"]:::repair
A2 --> A3["Rule insulated: the rest stay underpaid for years"]:::failure
end
subgraph ClassSized ["Feedback With Standing (Class Scale)"]
direction TB
S1["Collective grievance aggregator<br/>Claimants' union / Structural advocate"]:::gate --> S2["Filing at rule's size<br/>Exception log triggers mandatory review"]:::binding
S2 --> S3["Category on Trial<br/>Conversion process rewritten"]:::repair
end
F -- "atomize" --> A1
F -- "aggregate" --> S1
Who gets to hold the cases together
The design question is who is allowed to assemble the pattern, and whether the assembly carries standing once it is done. The institution can assemble the pattern itself: the exception log, the cluster threshold, the diagnostic tripwire. Those tools are necessary, but they’re owned by the party the pattern implicates, so the log tends to be the record nobody keeps. The department had such a tool: a framework officials wrote in 1979 for mistakes that affect a large volume of cases. The department later acknowledged that a tribunal ruling in October 2014 should have set it off.3
A third party can assemble it: a regulator’s complaint aggregation, a newsroom, a class action, a union grievance procedure. The destabilization right belongs here too, since it lets a court order restructuring when an institution’s failures are systemic instead of incidental.4 Or the people carrying the remainder can assemble it themselves, in a record kept outside the institution. There, a case that vanished once it was patched can still be counted next to the next one.
Feedback with standing is the absorbers’ own record with a third party’s authority. The absorbers control the aggregator, and the institution has to respond to its pattern at the size of the rule. Structural advocates are one form. A claimants’ union with the right to file against the conversion process rather than one award is another. In both, the pattern arrives as a filing the institution has to meet. A suggestion box only supplies evidence that officials may weigh or set aside.
The aggregator gets corrupted in familiar ways. It gets captured, funded by the institution, staffed from it, reporting into it, so that assembling the pattern becomes one more way of hearing it. Or it gets starved, left to the stamina of whoever is angriest this year, so the class-sized right exists on paper and nobody exercises it.
What makes a finding bind
Assembling the pattern gets the underpaid claimants as far as a filing. Whether the filing binds depends on what the department can do next, and the answer has to be very little.
The claimants, or the organization that speaks for them, start the review without asking the department’s leave, because a review the department must approve can only report what its officials can live with. Once one award is shown wrong, the process is examined against every conversion it touched, and the department has to hand over the files. If the process keeps underpaying people while it is examined and the shortfall could cost someone their home, an independent authority can order the missing amounts paid to the people at risk. That way they don’t carry the cost of the delay. A finding that holds then sets off remedies fixed in advance: arrears, a rewritten process and, if the process keeps failing, the department’s authority to convert claims at all. The department can’t declare the remedy finished. Someone else checks that the department paid the arrears and changed the process. Skipping that check has a consequence of its own, because an institution that certifies its own repair has brought back self-certification at the last step. When the National Audit Office examined this repair in 2018, it found that the search for underpaid claimants missed some people in the work-related activity group, and the department committed to reviewing 6,000 more cases.5
What the claimants hold is the power to start a review rather than a promised outcome. That power has force because a finding that holds can’t be filed and forgotten. It also has to be cheap to start and to finish. Count the claimants’ hours and money, the months before anyone is repaid, and whether the next claimant gets relief without filing anything. If the people who won carried most of the cost of making the department lose, the mechanism is running on their endurance.
Who can force the institution to learn
Participation asks whether people are heard. Corrigibility asks whether what they know can change the system that made speaking necessary. Democratic experimentalism comes close: pooled local knowledge revises practice, as if learning were what institutions lack. It isn’t. Institutions routinely hold accurate information about their own failures and carry on undisturbed, so a theory of organizational learning can’t be the standard. The standard is who can force the institution to learn, and at what size. Feedback with standing is conflictual on purpose. It hands some party the standing to make silence expensive, which is exactly what feedback without standing is arranged to avoid.
Adapting and correcting belong with the same people
An institution can’t avoid producing mismatch, because abstraction is how it acts at all. Someone is always supplying the intelligence, flexibility, care, time or endurance that closes the gap. Once you have counted that burden, the arrangement is democratizing if it learns from the people who absorb its remainder, and only using people up if it consumes their capacity to compensate for it.
Institutional power is democratized when the burden of adapting and the power to correct sit with the same people, and that alignment includes size. The burden is carried one person at a time. The correction has to be filed at the size of the rule. Some arrangements grant the absorbers standing over their own files and call that alignment. They have aligned the two at the exception rung and stopped.
Alignment at the size of the rule distributes exposure to error, authority to interpret it, capacity to contest it, and liability for its costs. Fallibility needs no distributing, because everyone has it. The people who carry an institution’s errors need enough standing to make those errors consequential for the institution. That means enough to grant an exception, force a justification, move a resource and revise a rule. It also means enough of them, held together, to reach the rung where the rule lives.
Neither ownership nor procedure settles it. Participation without rungs does not settle it either, and a rung the absorbers cannot reach together, or cannot afford to keep reaching, settles nothing. The same test applies whether the institution is a market, a state, a profession or a model.
The test secures control over the institution. It doesn’t secure freedom from needing the institution’s permission, and the two come apart. An insurer could have independent appeals, automatic remedies and binding review of its rules and still be the only route to care, on condition that the person bought it. That institution is corrigible and still holds a lever over everyone who can’t leave. Whether the lever should exist is a question no procedure answers.
Revision
29 October 2026: the appeal argument now distinguishes a remedy confined to one file from an individual case that establishes precedent or triggers review of every affected case. It previously treated the person-sized appeal as a limit inherent in appeals.
Notes
National Audit Office (NAO), Investigation into errors in Employment and Support Allowance, HC 837, March 21, 2018, nao.org.uk. Legislation from 2010 required the check on both parts. The department told the auditors its process had aimed to avoid delaying transfers, reduce administration costs and prevent breaks in payment.
NAO, Investigation, summary paragraphs 1–12, paragraph 3.5 and Figures 3 and 8. Arrears could be paid only from 21 October 2014, the date of an earlier Upper Tribunal ruling, which the department estimated left £100 million to £150 million of underpayments it could not pay at all.
NAO, Investigation, paragraph 3.8 and Figure 9. The framework is Legal Entitlements and Administrative Practices (1979), which the department calls a LEAP exercise when it runs one.
Charles Sabel and William Simon call it a destabilization right: when an institution’s failures are systemic rather than incidental, the remedy is a duty to restructure rather than damages for the person who happened to sue.
NAO, Investigation, summary paragraph 9 and paragraph 3.6.