Undercurrent research29 Mar 202612 min read

Undercurrent · Deep Dive · Technology & Governance · Part III

The 1987
Mainframe

Britain's most critical public systems run on code that belongs in a museum. Every attempt to modernise them has either failed catastrophically or been quietly abandoned. And the people who understand the code are dying.

TechnologyGovernmentPensionsPolicy

1988

Year built

PSCS core system

£90bn

Annual pensions

administered through it

£1bn+

Underpayments

caused by its failures

134,000

Pensioners

underpaid (errors to 1985)

The Clean Interface Is a Lie

Imagine you apply for your State Pension online. The GOV.UK form is clean, modern, reassuring. You get a confirmation email. You assume the data is flowing through some secure government cloud. What actually happens: a civil servant prints your application on paper and types it into a computer built in 1988.

That computer is not accredited to modern government security standards. It processes your data overnight, in batches. If the calculation is complex, a caseworker does it manually on a spreadsheet. This is not a historical curiosity. This is how £90 billion a year in State Pensions is administered right now.

Context — The Interface

There Is a Particular Kind of Institutional Deception

That works not through active dishonesty but through the careful management of appearances. The GOV.UK design system is one of the most admired in the world — clean, accessible, well-written. When you interact with a government service online, it looks like a modern digital organisation. The confirmation email arrives in seconds. The progress tracker updates. Everything feels contemporary.

The interface is real. What it connects to is not what you imagine.

Underneath the clean front-end of Britain's most important public services sits a patchwork of legacy infrastructure that would be unrecognisable to the average citizen. The core system administering your State Pension was built in 1988. The database that police officers query to check criminal records runs on a mainframe from 1974. HMRC's VAT processing system was designed in the 1970s, before VAT existed in its current form. The NHS spent £10 billion trying to build a unified patient record system, failed completely, and today your medical records are still routinely faxed between hospitals.

This is not a story about government incompetence, though there is plenty of that. It is a story about a structural trap that is almost impossible to escape — and about what happens when you build critical national infrastructure on a foundation you can never replace.

Context — The Code

What Is COBOL, and Why Should You Care?

COBOL stands for Common Business-Oriented Language. It was invented in 1959 by a team led by Grace Hopper, a US Navy rear admiral and computer scientist, specifically to process business transactions on the mainframe computers of the era. It is, in computing terms, ancient. The average COBOL developer today is over 55 years old. Many of the people who originally wrote the systems running Britain's public infrastructure have retired. Some have died.

And yet COBOL is everywhere. There are an estimated 220 billion lines of COBOL code still in active production globally. It processes approximately 95% of all ATM transactions worldwide. It handles the majority of credit card transactions, airline reservations, and insurance claims. The reason it has survived is the same reason it is so dangerous: it works. It is extraordinarily reliable for processing large volumes of structured data in sequential batches. It has simply never been replaced, because replacing it is harder than maintaining it — until, eventually, it isn't.

The UK government's relationship with COBOL is particularly acute. Unlike the United States, which has at least acknowledged its COBOL dependency and begun investing in modernisation programmes, the UK has spent the past two decades layering modern-looking interfaces on top of legacy systems rather than replacing the systems themselves. The result is a kind of digital Potemkin village: the facade is contemporary, the infrastructure behind it is not.

Context — The Pension System

The PSCS: A Case Study in Institutional Paralysis

The Pensions Strategy Computer System (PSCS) was introduced in 1987 and fully rolled out in 1991. It runs on a COBOL codebase and was originally hosted on ICL VME mainframes — hardware that has since been replaced, but whose replacement tells its own story. The DWP completed a programme called VME-R in 2021, which migrated the PSCS from old ICL hardware to modern virtualised infrastructure. The DWP's own blog described this as "the largest programme of its kind in Europe." It involved migrating 10 billion rows of data and converting 26 million lines of application code.

Note the word "converting." The COBOL application logic was not replaced. It was translated and moved onto newer hardware. The underlying system — the rules, the calculations, the data structures, the batch-processing architecture — remains a product of the late 1980s.

The DWP's own blog noted, with striking candour: "We had to completely redesign the batch processing service, as this service was so old and complex and no-one was around who knew it anymore."

Read that again. The people who understood how the system worked were gone.

The DWP was redesigning a critical national infrastructure system based on reverse-engineering code that nobody fully understood. This is not an unusual situation in government IT. It is the norm.

The PSCS — in combination with the manual processes built around it — systematically underpaid 134,000 pensioners, mostly women, by a total of over £1 billion. Some of those pensioners have since died. The DWP has been running correction exercises — for the ninth time since 2018 — at a cost of £24.3 million in staff time alone.

— Between 1985 and 2021

Beyond Google

The PSCS is not accredited to modern government security standards. The DWP decided in 2007 that it could not meet those standards and chose not to pursue accreditation rather than fix the system. This means that the database holding the pension records of millions of British citizens operates outside the security framework that governs other government systems. The DWP told the Public Accounts Committee that replacing the PSCS would be "incredibly complex and very risky." That is true. It is also true that not replacing it has already cost over £1 billion in underpayments and will cost more.

The Mechanism

Five Reasons These Systems Can Never Be Replaced

1. The Knowledge Problem

The original developers are gone. The code is largely undocumented — written at a time when documentation was considered an optional extra. The Police National Computer, built in 1974, runs applications in COBOL, C, 390 assembler, and Easy Case assembler, with code so undocumented that simple tasks involve 40 business rules where five were expected. Changing one thing breaks five others. Nobody knows which five.

2. The Data Problem

Decades of records — pension histories, tax records, criminal records — are encoded in data formats that don't map cleanly to modern databases. The DWP migrated 10 billion rows of data in its VME-R programme and described it as the most complex data migration in European government history. Even then, the data was moved, not transformed. The underlying structures remain legacy.

3. The Political Problem

No politician wants to own the system failure that stops pension payments to 12 million people. The safest career move — always, in every government, of every party — is to extend the legacy system for another five years and let the next minister deal with it. The PSCS has been "due for replacement" since at least 2007. It is still running.

4. The Procurement Problem

Government IT procurement is designed for buying things, not building things. The standard model is to write a specification, put it out to tender, and award a contract to a large systems integrator — Fujitsu, Capita, IBM, EDS. The contractor builds to the specification. The specification is wrong, because nobody fully understood the legacy system. The contractor charges for changes. Costs double. The programme is cancelled or descoped. This has happened with the NHS NPfIT (£10 billion), the HMRC ASPIRE contract (£10.4 billion), Universal Credit IT (£1 billion+), and the NLEDS Police National Computer replacement (£1.1 billion over budget, five years late). The procurement model itself guarantees failure.

5. The Wrapper Trap

Each generation of politicians commissions a modern-looking front-end to make the legacy system appear contemporary. The DWP's Customer Account Management (CAM) system provides a modern interface over the PSCS. HMRC's online tax portal connects to a 1970s batch-processing backend. The NHS Spine connects to hospital systems that haven't been updated since the 1990s. Each wrapper makes the underlying system harder to replace, because now you have to replace both layers simultaneously — and the wrapper has introduced new dependencies that nobody fully mapped.

Context — The Pattern

The PSCS Is Not an Outlier. It Is the Rule.

The State Pension system is the clearest example because the consequences — £1 billion in underpayments, 134,000 people affected — are so concrete and so well-documented. But the same pattern repeats across every major public system in Britain.

Police National Computer (1974)

Built in 1974, runs on Fujitsu mainframe with applications in COBOL, C, 390 assembler, and Easy Case assembler. In January 2021, a routine maintenance script accidentally deleted 413,000 records — including fingerprints and DNA evidence — because the code was so old and poorly understood that a single error cascaded through the system. The NLEDS programme, designed to replace the PNC, has been running since 2016. It is now five years late and £1.1 billion over budget.

HMRC VAT System (1970s)

Collects over £500 billion in tax annually on systems whose core VAT architecture dates to the 1970s. The ASPIRE contract — a £10.4 billion outsourcing deal with Capgemini that ran from 2004 to 2017 — provided stability but locked HMRC into technological stagnation. When the NPS (National Insurance and PAYE Service) was introduced in 2009, its flawed implementation created a backlog of 18 million PAYE cases affecting 15 million people, delaying repayment of overpaid tax and putting £1.4 billion in underpaid tax at risk.

NHS (NPfIT 2003-2011)

The National Programme for IT was launched in 2003 with the ambition of creating a unified digital health record for every patient in England. It was abandoned in 2011 after spending £10 billion. Nothing was built. The money was gone. Today, NHS trusts operate on a patchwork of incompatible systems. When the WannaCry ransomware attack hit in May 2017, it crippled over 60 NHS trusts not because of a sophisticated exploit but because they were running Windows XP — an operating system Microsoft had stopped supporting in 2014. The cost was £92 million and 19,000 cancelled appointments.

Britain's catalogue of government IT disasters, 2004–2021. Combined cost: over £25 billion. This is not a run of bad luck. It is a structural failure of how government procures, builds, and maintains technology.

So What — International Comparison

Estonia Built an Entire Digital Government for Less

In 1991, Estonia regained independence from the Soviet Union. It had no functioning public administration, no legacy IT systems, and very little money. By 2005, it had introduced e-voting. By 2007, filing a tax return took three minutes. Today, 99% of Estonian residents have a digital ID card, 100% of government services are available online, and the country's X-Road data exchange layer allows every government agency to share data securely with every other — with the citizen's consent, with a full audit trail, and with zero paper. Estonia ranks sixth in the UN E-Government Survey 2024. Denmark ranks first.

The lesson that politicians draw from Estonia is usually the wrong one. They say: "Estonia is a small country of 1.3 million people. You can't compare it to Britain." This is a deflection. The relevant comparison is not population size. It is institutional culture.

Estonia built its digital government on a single foundational principle: data should be entered once, and once only. Every agency queries a shared data layer rather than maintaining its own database. The citizen fills in one form. The system does the rest. Britain, by contrast, has built each government department's IT system independently, over decades, with incompatible data formats, incompatible security standards, and no shared identity layer — because the UK has no national digital identity system.

India enrolled 1.4 billion people in the Aadhaar biometric identity system in under a decade, starting in 2009. The programme cost approximately $1 billion — less than the NHS NPfIT spent before it was abandoned. India is not a wealthy country with sophisticated public administration. It is a country of 22 official languages, extreme geographic diversity, and a history of bureaucratic dysfunction. It succeeded because it had political will, a clear technical architecture, and — crucially — no legacy systems to protect.

The UK's problem is not technical. The technology to build a modern digital government has existed for twenty years. The problem is institutional: a procurement culture that favours large contracts with large suppliers; a political culture that punishes failure and therefore incentivises inaction; and a civil service that has systematically outsourced its technical capability to contractors, leaving it unable to specify, build, or evaluate the systems it needs.

So What — Personal Impact

This Is Not an Abstract Infrastructure Problem

It is tempting to read this as a story about government dysfunction — interesting, perhaps outrageous, but ultimately remote from daily life. It is not. The fragility of these systems has direct, personal consequences for every person in Britain.

If you are approaching retirement age: Your State Pension calculation may already contain errors introduced by the PSCS's manual processes. The DWP has identified over £800 million in underpayments as of March 2025, and the correction exercise is ongoing. The Public Accounts Committee found there is "a risk that the errors that led to underpayments in the first place will be repeated in the correction exercise." You would not know if your pension was wrong. The system has no mechanism to tell you.

If you are admitted to hospital: The doctor treating you may not have access to your GP records, your allergy history, or the results of tests done at a different trust. This is not a hypothetical. It is the daily operational reality of NHS clinical staff, who spend an estimated 13% of their working time chasing information that should be instantly available. The NHS Long Term Plan acknowledged this in 2019. The problem has not been solved.

If you use a high street bank: The mobile app on your phone connects to a core banking system that may be running on 1970s COBOL mainframes. The 2012 RBS outage — caused by a botched update to a 40-year-old CA-7 batch scheduler — locked millions of customers out of their accounts for weeks. The 2018 TSB migration disaster cost £330 million and locked out nearly two million customers. These were not freak events. They were the predictable consequence of running critical financial infrastructure on code that nobody fully understands.

The Structural Trap in Plain English

• Systems are too old to be safe, but too embedded to be replaced.

• Every year they run, the knowledge of how they work becomes scarcer.

• Every year they run, more data accumulates in formats that don't translate cleanly to modern systems.

• Every year they run, the cost and risk of replacement increases.

• The political incentive is always to extend, never to replace.

This is not a problem that will solve itself.

One Takeaway

Britain has spent over £25 billion on failed government IT projects since 2004 — more than Estonia spent building an entire digital government from scratch. The reason these systems haven't been replaced is not technical. It is political: no minister wants to own the failure that stops pension payments to 12 million people. So they extend the legacy system for another five years, add a modern-looking front-end, and call it digital transformation. The code underneath gets older, the people who understand it get fewer, and the risk of catastrophic failure grows — quietly, invisibly, year by year.

The Dinner Party Line

"The system that pays your grandmother's pension requires her online application to be printed out and retyped into a computer from 1988 that isn't accredited to modern security standards. Britain has spent more on failed IT projects to replace it than Estonia spent building an entire digital government from scratch."

You’ve looked beneath the surface.

Now follow the connection.

Industrial Geography·14 min read