Public failures, scored
Each case was established by a court, an inquiry, or a regulator. Each is scored on six safeguards, each named for what it keeps in check: capability, authority, evidence, dependency, standing, and correction. Each score names the draft clause this project argues would have caught the failure. The findings are the court's, the inquiry's, or the regulator's; the scores are this project's reading of them. Each case names who carried the errors while the system ran. A seventh mark records whether the institution changed the process that produced the errors, or corrected the errors and left the process as it was. The scores are not a verdict on anyone. They show where the safeguards broke.
What the record established and where the safeguards broke
Standing, the right of the people a system decides about to challenge a decision and be answered, held in none of the five. Correction held once, at Apple Card, where the issuer changed its own policies without a court, though it took a regulator to prompt it.
held in place and working · drifted in place but out of step with what it should track · failed absent, or its absence is what the inquiry found
After the failure — did the institution change? Changed the rules = it changed the rule, category, or workflow that produced the errors · Partly changed = some change, late or imposed from outside · Process unchanged = the errors were corrected; the process that produced them was not. Read the theory
-
Automated welfare-debt raising: annual tax-office income averaged across fortnights to assert overpayments, with the burden of disproof placed on the recipient.
- Capability
- drifted
- Evidence
- failed
- Dependency
- drifted
- Standing
- failed
- Correction
- failed
- After the failure
- Process unchanged
- Scale
- About 470,000 debts raised unlawfully. The class-action settlement the Federal Court approved in 2021 covered roughly 381,000 people, with repayments, wiped debts, and interest valued at about A$1.8 billion.
- Time to halt
- Three years and four months
- Errors carried by
- Recipients, who had to produce payslips, often for years past, to disprove debts raised by averaging. The scheme had been booked as a budget saving of more than a billion dollars before it raised a debt.
- Halted by
- The Federal Court, on a consent order the Commonwealth agreed to hours before a hearing it would have lost.
-
Risk-scored fraud detection on childcare benefit claims, with an all-or-nothing rule that reclaimed the entire benefit for any irregularity and an intent label that barred repayment arrangements.
- Capability
- drifted
- Evidence
- failed
- Dependency
- drifted
- Standing
- failed
- Correction
- failed
- After the failure
- Partly changed
- Scale
- More than 30,000 parents wrongly accused of fraud and made to repay benefits, often tens of thousands of euros; children placed in care; the cabinet resigned.
- Time to halt
- About seven years
- Errors carried by
- Parents, more than 30,000 of them, made to repay benefits, often tens of thousands of euros, without being told why they had been flagged. The people working the review queue saw the pattern first and had no standing to raise it.
- Halted by
- The Council of State reversing its own case law in October 2019, then a parliamentary inquiry.
-
A branch accounting system whose reported shortfalls were treated as proof of theft or false accounting, by an operator that was also the investigator and the prosecutor.
- Capability
- drifted
- Evidence
- failed
- Dependency
- failed
- Standing
- failed
- Correction
- failed
- After the failure
- Process unchanged
- Scale
- More than 900 prosecutions; hundreds imprisoned, bankrupted, or both; the inquiry's first volume links at least thirteen suicides to the scandal and identifies roughly 10,000 eligible for redress.
- Time to halt
- About twenty years
- Errors carried by
- Subpostmasters, who were contractually liable for the shortfalls Horizon reported, and were then investigated and prosecuted by the operator on that evidence. Each was told the system was robust and that no one else had complained.
- Halted by
- A group of 555 subpostmasters in civil litigation, then the Court of Appeal, then an Act of Parliament quashing convictions in bulk.
-
A standardization model that replaced canceled A-level and GCSE exams by fitting each school's historical grade distribution to its current cohort, overriding teachers' assessed grades for all but the smallest classes.
- Capability
- held
- Evidence
- failed
- Dependency
- held
- Standing
- failed
- Correction
- drifted
- After the failure
- Process unchanged
- Scale
- About 39% of A-level grades issued below the teacher-assessed grade; the effect fell hardest on large cohorts in state schools and lightest on small classes, which were exempt.
- Time to halt
- Four days
- Errors carried by
- Students in large cohorts at state schools, where the model overrode teachers' grades most. They could appeal only through their school, and not on the ground that the model had ranked them wrongly. Small classes were exempt.
- Halted by
- The Secretary of State, after Scotland had already reversed its equivalent and universities had begun allocating places on the model's grades.
-
Automated credit-limit decisions on a consumer card issued by Goldman Sachs, with no route by which an applicant could learn why their limit differed from a spouse's, and a reconsideration path that was a barrier: a six-month wait on a limit increase, removed only after the November 2019 complaints.
- Capability
- held
- Evidence
- drifted
- Dependency
- held
- Standing
- drifted
- Correction
- held
- After the failure
- Changed the rules
- Scale
- No unlawful discrimination found. The regulator's finding was that applicants, and the bank's own staff, could not explain individual outcomes, and that the reconsideration path that existed was a barrier.
- Time to halt
- About seventeen months to a policy change; the model was not withdrawn.
- Errors carried by
- Spouses whose finances were shared but whose credit histories were not, offered lower limits with reconsideration gated behind a six-month wait. The bank's customer service staff took the complaints with no explanation to give.
- Halted by
- Nobody. The issuer changed its policies after a regulator's investigation found the process, not the model, deficient.
Time to halt, drawn to one scale
Each bar runs from first harm to the halt. England's 2020 exam grades: four days, too short to see at this scale, so its bar is drawn at the minimum width. Where the record dates an end only to the year or month, the bar measures from the middle of it.
The record, drawn
How often each safeguard held, drifted, or failed
Capability, whether a system stayed within what it was declared and approved to do, failed outright in 0 of 5 cases. Standing, the right of the people a system decides about to challenge a decision and be answered, failed in 4. Evidence failed in 4, and correction in 3.
How to read a score
A score is a claim about one safeguard at the time of the harm, made from the primary record. It is not a finding of fault, and it is not a measurement of the system today.
- held
- The safeguard was written down and still matched what it was meant to check. Capability was declared; authority named its limits; the evidence was the kind that could justify the action taken.
- drifted
- The safeguard existed but stopped matching what it was meant to check. Authority that outlived the evidence it was issued on. A capability that grew about fiftyfold while the review capacity beside it stayed still.
- failed
- The safeguard was absent, or its absence is what the inquiry found. A right to challenge that won individual cases but never changed the system that decided them. Evidence that the party relying on it never held.
Every finding is linked to the clauses it falls under and to the law it applies. If a case is scored wrongly, the primary sources are listed on its page; tell us which one we misread.
The seventh mark is not a safeguard. It records what happened after the failure: whether the institution changed the rule, category, or workflow that produced the errors, or corrected the errors and left that process as it was. A case can score badly on the safeguards and still end in change, or score well and still end with the process unchanged. The theory essay on exception learning explains the difference.
For three composite scenarios, written as remediation checklists and drawn from no single institution, see the composite scenarios.