The Register

9.0 What this chapter gives you#

1. You will be able to say what a register actually is — an authority’s written assertion that a stated fact is true — and explain why its value comes from who is willing to act on it, not from how the data is stored.
2. You will be able to name what a birth entry, a marriage entry and a death entry each guarantee, and, just as important, what each one does not guarantee.
3. You will be able to tell a population register from a functional register, give real examples of each, and predict the kind of error each one will make.
4. You will be able to read a twenty-character legal entity identifier, name the parts of it, say which standard defines it, and state what the system behind it checks and what it does not.
5. You will be able to use the terms authoritative source, source of truth and system of record correctly, and give a one-sentence reason why they are three different things and not three names for one.
6. You will be able to describe how a death travels from a bedside to a bank, name every place it stops, and quote real figures for how often it arrives against the wrong person.
7. You will be able to specify a correction procedure for a wrong register entry, say who must do what, and estimate the latency — the waiting time before a change takes effect — at every hop.
8. You will be able to quote the coverage numbers behind the United Nations legal identity target, say who the missing people are, and explain why registering a birth late is much harder than registering it on time.
9. You will be able to design a register that is append-only, that records corrections as new records, and that never silently overwrites a fact, and say what each of those three rules buys you on the day somebody disputes an entry.

There is a moment, in almost every argument about who somebody is, when one side says: it is written down. Not written down by them. Written down by an office. In a book, or in a database that is the descendant of a book, kept by a person whose job is to keep it, under a law that says the book exists. That is a register, and the sentence “it is written down” is doing an enormous amount of work.

A register is not a store of facts. It is an assertion. Somebody with authority has written that a thing is true, and has staked the standing of their office on it. The paper you are handed is not the fact; it is a copy of the assertion. This distinction sounds like hair-splitting until the day the assertion is wrong, and you discover that the world is not arguing with reality, it is arguing with the book.

The second thing to understand is harder, and it is the thesis of this chapter. A register is worth exactly what other people will do because of it, and nothing more. A register nobody relies on is a filing cabinet. A register that a bank, a school, an employer, a land office and a court will all act on without asking a second question is the most powerful object in a person’s life. The technology is identical in both cases. The difference is reliance, and reliance is a social and legal fact, not a technical one. This is why a beautifully engineered register can be worthless and a mouldering paper ledger can decide whether you own your house.

The third thing is the one that ruins people. Registers are wrong. Not rarely, not exceptionally — routinely, at a low but relentless rate, and in both directions. They say things that are not so, and they fail to say things that are. And because the whole point of a register is that others act on it without checking, a wrong entry propagates faster and further than a right one, and takes far longer to withdraw than it took to make. Everything in the second half of this chapter is about that asymmetry: how to design for it, how to correct it, and how to count the people who are missing from the book altogether.

The plain version#

A village with one book#

Imagine a village of about two thousand people. In the middle of the village there is a small office with a cupboard, and in the cupboard there is a book. One person, the keeper, is allowed to write in it. She is paid by the district, not by anybody in the village, and if she writes something false she loses her job and possibly her liberty.

Three kinds of thing go into the book. When a child is born, somebody comes within a few days and reports it, and the keeper writes a line: the day, the place, the child’s name, the parents’ names, and the name of the person who came and reported it. When two people marry, they come in front of the keeper with two witnesses, and she writes a line saying that on this day these two people became married to each other. When somebody dies, a doctor writes a note saying what they died of, a relative brings the note, and the keeper writes a line saying that this named person died on this day.

That is the whole system. It has no computers, no cards, no photographs. And yet within twenty years of the book being properly kept, the village runs on it.

The school will not admit a child without a copy of the birth line, because it needs to know the child’s age and it is not willing to take the family’s word for it. The land office will not transfer a field to a son without a copy of the death line for the father, because otherwise anybody could claim any dead man’s land. The pension office pays every month against the list of people who are in the book and are not in the death part of it. A man who wants to marry a second wife while the first is alive is stopped by the marriage line, because the keeper can look it up. The army recruiter, the passport clerk in the district town, the bank that opens a savings account — all of them, when they need to know something about a person, do not investigate. They ask for a copy of a line in the book.

Notice what has happened. The keeper does not have any power over these people. She cannot admit a child to school or transfer a field. What she has is something better: everybody else has decided in advance to do what her book says. That decision — made separately by the school, the land office, the pension office, the bank — is the entire value of the book. If tomorrow the school decided it would rather see a letter from a priest, and the land office decided it preferred two witnesses, the book would still be perfectly accurate and completely worthless.

This is the first thing to hold on to. The book is not powerful because it is correct. It is powerful because other people have agreed to act on it.

What each kind of line actually promises#

The three kinds of line in the book look similar. They promise very different things, and mixing them up is where most confusion starts.

The birth line promises that a birth happened, on a stated day, in a stated place, and that at the time it was reported the informant said these were the parents and this was the name. That is a lot, and it is the foundation of nearly everything else. It is also less than people think. It does not promise that the child is still alive. It does not promise that the man named as the father is the biological father — the keeper did not test anything, she wrote down what she was told by somebody the law allowed to tell her. It does not promise that the adult standing in front of you today, holding a copy, is that child. It is a statement about an event in the past, not about the person in front of you.

The marriage line promises something different in kind. A birth is an event that happened whether or not anybody wrote it down. A marriage, in most legal systems, is partly created by the writing down. Before the keeper writes, these two people are not married; after she writes, they are. The line does not describe a fact in the world so much as bring a legal status into being and record the moment it began. That is why marriage registers are guarded so carefully and why forging one is treated so seriously: you are not lying about the world, you are manufacturing a status.

The death line is the strangest and the most dangerous. It promises that a named person has stopped existing as a legal person. Every other line in the book gives somebody something. The death line takes everything away at once: the pension stops, the bank account freezes, the land passes to somebody else, the marriage ends, the vote disappears. It is the only line in the book that is a demolition order. And it is the line most likely to be acted on instantly and without question, because everybody downstream is frightened of paying a pension to a dead man and nobody is frightened of the opposite mistake.

Here is a way to feel the difference. If the birth line is wrong by a day, a child starts school a year early or late. If the marriage line is wrong, two people have a legal problem they can usually untangle. If the death line is wrong, a living person is turned off, and every day that passes makes it harder to turn them back on, because more offices have copied the wrong line into their own lists.

One book that lists everybody, and many books that list some people#

The village book tries to list everybody. Every birth, every death, everybody who lives here. That is one kind of book, and its ambition is completeness.

But the village has other books too, and they are a different kind. The tax office keeps a book of everybody who pays tax. The clinic keeps a book of everybody it has treated. The election office keeps a book of everybody entitled to vote. Each of these is a real list, kept by a real authority, and each is correct about its own subject. None of them is trying to list everybody. The tax book has no entry for a child. The clinic book has no entry for a healthy adult who has never been ill. The voting book has no entry for anyone under age.

The trouble starts when somebody treats one of the second kind as if it were the first kind. If you want to know how many people live in the village, and you count the tax book, you will conclude that children do not exist. If you want to send a health worker to every household and you use the clinic list, you will visit only the households that have already been visited. Every one of those books is correct and every one of them will mislead you if you ask it a question it was not built to answer.

There is a second, quieter problem. The books disagree, and they have to be made to agree. The tax office would like to know when somebody dies, so it can stop chasing them. The clinic would like to know when a baby is born, so it can offer vaccinations. So messages pass between the books. And every message between two books is a place where a fact can be attached to the wrong person, because the books do not share a single unambiguous way of naming people. The village has four men called Rajan Pillai. Deciding whether two lines in two different books are the same person is a whole discipline of its own, and it is genuinely a matter of probability rather than certainty; chapter 8 is about that problem and does it properly.

A book with a page for something that is not a person#

Now imagine that the village gets a factory. The factory is not a person. It cannot be born and it will not die, but it can sign contracts, owe money, employ people and be sued. So the district opens a second book, in a different office, and in it the factory gets a line: the day it was created, its name, the address where letters can be sent to it, and the names of the people who are responsible for it.

Everything we said about the first book applies here. The line does not make the factory honest. It records that on a stated day somebody brought the right forms and paid the fee, and that from that day the factory exists as a thing the law recognizes. The line is worth exactly what the bank, the supplier and the court will do because of it.

There is one thing this second book does that the first one cannot. Two villages away there is another factory with exactly the same name. When the bank in the city lends money, it needs to be certain which of the two it is lending to. So the book gives each entry a number that no other entry anywhere in the world will ever have. Not a name, which repeats, but a number, which does not. That number turns out to be worth more than everything else on the page put together, because it is the only part that can travel between countries without ambiguity.

The rule that the keeper never rubs anything out#

One day a man comes to the office and says the book is wrong. His daughter’s name is Meera, and the keeper wrote Meena. He has lived with this for eleven years and now the school will not accept her form.

Here is where a well-run book differs from a badly run one. The bad answer is for the keeper to take a pen, cross out Meena, write Meera above it, and hand the man a fresh copy. It solves his problem this afternoon and destroys the book. Because from that moment nobody can tell, looking at the page, whether the book always said Meera, or whether somebody changed it last Tuesday, or who changed it, or why. The book’s whole value was that it was somebody’s honest record of what was reported at the time. A book that can be quietly edited is not a record; it is a current opinion.

The good answer, and the one the law usually requires, is this. The original line stays exactly as it was written in 1994. In the margin beside it the keeper writes a new, dated, signed note: on this day, on the strength of this evidence, this entry was corrected from Meena to Meera. Now the page carries two facts — what was recorded then, and what was corrected later — and anybody reading it can see both, and can see who did it and when. Nothing was destroyed. Something was added.

That single habit — never rub out, always add — is the difference between a register and a spreadsheet, and we will spend a good part of the technical half of this chapter on it, because most modern systems get it wrong in ways that paper got right two hundred years ago.

Meera Nair, carried all the way through#

Let us follow one person, with dates, and keep her for the rest of the chapter.

Meera Nair is born on 14 March 1994. Her father goes to the registration office on 19 March, five days later, well inside the twenty-one days the rules allow. The keeper writes the line and gives it a number. The family pays nothing, because it is on time.

The keeper writes the name as Meena. Nobody notices for eleven years, because the family never needs the paper until the school asks for it in 2005.

In 2005 the father applies to have the entry corrected. He brings a sworn statement and two documents. The correction is made as a dated marginal note. The original line is untouched. Elapsed time from the error being written to the error being corrected: about eleven years and two months. Elapsed time from the request to the correction: in this example, three months, because the office needed the sworn statement to be witnessed properly.

In 2019 Meera marries, and a marriage line is written in a different book, in a different town, by a different keeper.

In 2024 she moves abroad and starts a company, which gets a line in a company book and a number of its own.

In 2026 a bank freezes her account, because a list of dead people reached the bank with her name on it. She is not dead. Somebody else, with a similar name and a similar date of birth, is.

That last event is the one worth remembering. Nothing was hacked. No document was forged. A true fact about one person was attached to another person, and every office downstream did exactly what it was supposed to do with it.

Where the plain version stops being true#

There is no one book, and there never was#

The village analogy has one keeper and one cupboard. Real countries have neither. Civil registration is usually run locally — by a municipality, a panchayat, a county clerk, a parish — under national rules, and the national office holds an index or a copy rather than the original. England and Wales have had a General Register Office since the civil system began in 1837, but the entries are made by local registrars, and for most of that period the national copy was a quarterly transcript, not the record itself. India registers births and deaths through state governments under a national Act, with a Registrar General of India above them, and until very recently the national picture was assembled from state returns rather than held centrally.

So the phrase “the register says” almost always hides a question: which copy, held by whom, updated when. A correction made in a local register may not reach the national index for months. The certificate in your hand was true on the day it was printed and says nothing about afterwards.

The honest version: a register is a distributed system with a legal story told over it. The legal story says there is one authoritative record. The engineering reality is a primary record, a set of indexes, and a great many copies of varying age, and most disputes are between copies of different ages rather than between right and wrong.

“Authoritative” is a decision somebody made, not a property of data#

In the plain version, the book is authoritative because it is the book. That is circular, and in practice authority is granted, not intrinsic.

Somebody — a ministry, a regulator, a company’s own risk committee — decides that for a particular fact, a particular source will be believed. That decision can be wrong, can be political, and can change. The same underlying data can be authoritative for one purpose and inadmissible for another. A utility bill is accepted as evidence of address by a bank and refused by a passport office. A hospital’s record of a birth is excellent evidence that a birth occurred and is not the birth register.

This is why the technical half of this chapter spends time separating three phrases that are used interchangeably and should not be: authoritative source, source of truth, and system of record. They answer three different questions: who may I believe, where does this fact come from originally, and which system’s copy governs when copies disagree.

The register does not know the fact; it knows the report#

The village keeper did not witness the birth. She wrote down what an informant told her, and the informant was chosen by law — usually a parent, or the occupier of the house, or the person present at the death.

This means every register entry is at best a faithful record of an assertion made by a particular person on a particular day. If the informant was mistaken, the register is faithfully wrong. If the informant lied and the documents supported the lie, the register is faithfully fraudulent, and the fraud is now wearing the register’s authority.

The Lal Bihari case in India makes the point better than any argument. In 1975 a man in Uttar Pradesh was recorded as dead in the official records after, by his account, a relative bribed an official, so that his share of family land could be taken. He was alive the whole time. It took until 1994 — nineteen years — to have the death record annulled. He went on to found an association of people in the same position, which claims tens of thousands of members, and was awarded an Ig Nobel Peace Prize in 2003. The system worked exactly as designed at every step. The input was false.

A death does not travel by itself, and nothing ever says “still alive”#

The plain version says the pension office pays everybody who is in the book and not in the death part of it. Read that again and notice the shape of it. The pension office is not acting on a positive statement that you are alive. It is acting on the absence of a statement that you are dead.

That asymmetry runs through every identity system in the world, and it causes both of the big failure modes. If the death message does not arrive, a dead person keeps drawing money, and this is the failure that auditors and newspapers care about. If the death message arrives against the wrong record, a living person is switched off, and this is the failure that ruins somebody’s year and that almost nobody counts.

Because only the first failure is politically expensive, systems are tuned to accept death reports eagerly and to make un-death slow. That is a design choice, and it is usually made without anybody deciding it.

Being absent from the book is not evidence that you do not exist#

The village book is complete because the village is small and everybody uses it. Globally, that assumption fails, and it fails for hundreds of millions of people.

If your birth was never registered, you have no line to copy. Every office you meet asks for the copy. The usual route to getting a line late is to produce other documents, and the other documents are exactly the ones you could not obtain without the line. This is not an edge case handled by an exceptions process; for many people it is a wall.

And there is a worse version, where the book existed, you were in it, and the record was destroyed or was never made because nobody thought it would be needed. When the United Kingdom’s Windrush scandal broke in 2018, people who had lived in the country legally for decades could not prove it, partly because the state had not issued them documents at the time and had not kept usable records of its own. The National Audit Office reported on 5 December 2018 that the Home Office had identified 164 people who had arrived before 1973 and had been detained or removed since 2002, and that by 30 September 2018 its taskforce had issued documentation to 2,658 people. The lesson is not about immigration policy. It is that the absence of a record was treated as evidence of absence of entitlement, when it was in fact evidence of nothing except that no record existed.

The correction is not the fix#

The plain version ends with a tidy marginal note and everybody happy. The technical reality is that correcting the register is only the first of many steps, and often the easiest.

By the time a wrong entry is found, it has been copied. Copies live in banks, employers, insurers, credit bureaux, hospital systems, electoral rolls, tax files and private data brokers. Some of them refresh from the register regularly, some refresh only when the customer forces them to, and some never refresh at all. Correcting the source does not retroactively correct the copies, and there is usually no list of who holds a copy.

The honest version: a correction is a new fact that has to be propagated, and it propagates through the same slow, unreliable channels that carried the error, except that the error travelled with a following wind and the correction is going uphill. Plan the correction path when you design the register, not when the first complaint arrives.

The technical version#

Civil registration, defined precisely#

The United Nations definition, used by the statistical system and by the UN Legal Identity Agenda, is worth quoting because every word in it is load-bearing. Civil registration is the continuous, permanent, compulsory and universal recording of the occurrence and characteristics of vital events pertaining to the population, as provided through decree or regulation in accordance with the legal requirement in each country.

Continuous means it never stops and never resets. Permanent means the record outlives the event, the registrar and usually the government. Compulsory means the reporting duty sits on named people, with a penalty attached, rather than on the citizen’s convenience. Universal means it covers everybody in the territory, not citizens only. Those four adjectives are the specification of a register, and each of them is routinely broken in practice.

The UN Legal Identity Agenda, coordinated by a task force established from September 2018 after a January 2018 decision by the Executive Committee of the Secretary-General, brings thirteen UN agencies together under the joint chairmanship of UNDP, UNICEF and the UN Department of Economic and Social Affairs. Its operational definition of legal identity is equally precise: the basic characteristics of an individual’s identity, such as name, sex, place and date of birth, conferred through registration and the issuance of a certificate by an authorized civil registration authority following the occurrence of birth. It adds two clauses that people forget. Where there was no birth registration, legal identity may be conferred by a legally recognized identification authority, which should be linked back to civil registration. And legal identity is retired by the issuance of a death certificate upon registration of death. Identity, in the UN’s framing, has a defined beginning and a defined end, and both are register events.

Now, precisely, what each vital event register asserts.

A birth entry is evidence of the fact and time of birth and of the particulars as reported by a qualified informant. In most common-law systems the certified copy is admissible as evidence of those particulars, which is a much stronger position than an ordinary document enjoys. A marriage entry is usually constitutive: the legal status begins with the registration or with the ceremony that the registration records, so the entry is not merely evidence of a marriage, it is part of the marriage. A death entry is the instruction that ends legal personality, and in most systems it is the trigger for succession, for the ending of the marriage, for the cancellation of entitlements and for the release of assets.

Time limits, late registration, and what a late entry is worth#

Registers impose deadlines because the value of a report falls with time. Fresh reports come from people who were there. Late reports come from people reconstructing.

In England and Wales the framework is the Births and Deaths Registration Act 1953, sitting on a civil system that began on 1 July 1837 under the Births and Deaths Registration Act 1836, which created the General Register Office. Before that, the record of English life was the parish register, which Thomas Cromwell ordered every parish to keep in 1538 — one of the earliest population-wide registration mandates in Europe, and a useful reminder that the institution is nearly five hundred years old and the database part is very recent.

India’s framework is the Registration of Births and Deaths Act, 1969. Its section 13 sets out a graded late-registration ladder that is worth knowing because it is a model of how registers price staleness. A birth or death not registered within the prescribed period may be registered within thirty days of occurrence on payment of a late fee. Between thirty days and one year, it requires the written permission of the prescribed authority, the prescribed fee, and an affidavit sworn before a notary public. After one year, it may be registered only on an order made by a magistrate of the first class. The evidence burden rises with every step, because the register is buying less and less certainty for its assertion.

That ladder is also the wall described earlier. An adult who was never registered at birth must obtain a magistrate’s order, and to obtain it must produce evidence of the birth, and the evidence a court will accept is usually documentary, and the documents are usually obtainable only by people who already have a birth certificate.

The Registration of Births and Deaths (Amendment) Act, 2023 — introduced in the Lok Sabha on 26 July 2023, passed by the Lok Sabha on 1 August 2023 and by the Rajya Sabha on 7 August 2023 — changed the character of the Indian system in three ways that matter here. It provides for a national database of registered births and deaths, with state-level databases feeding it. It makes the birth certificate the single document to prove date and place of birth for a listed set of purposes, including admission to an educational institution, inclusion in an electoral roll and appointment to government employment. And it provides for Aadhaar numbers to be collected from parents and informants, which creates the join key that makes register-to-register reconciliation possible at national scale. Whether concentrating so much reliance on one document is wise is a live argument, and both sides have a point: concentration reduces fraud from weaker documents and increases the damage when the single document is wrong.

Population registers and functional registers#

A population register attempts to hold a current record for every resident, continuously updated from vital events and from moves. A functional register holds a record for every person who has a particular relationship with a particular function of the state: paying tax, receiving healthcare, voting, driving, receiving a benefit.

Denmark is the cleanest example of the population-register model. Its Central Person Register was established on 2 April 1968, with the enabling bill passed unanimously by the Folketing on 29 May 1968 and signed in June of that year; every resident received a personal number, and the register became the spine that other systems hang from. Sweden had gone earlier, introducing a nine-digit personnummer in 1947. The Netherlands runs a personal records database, the Basisregistratie Personen, as one component of a formal scheme of base registrations; the Dutch scheme comprises twelve base registers as described in the Netherlands’ own submissions to the European interoperability programme, and its governing idea is that a given fact has exactly one designated home, that public bodies are expected to use the data from that home rather than keep private copies, and that when a body suspects an error it is expected to report it back to the register that owns the fact rather than fix it locally.

That last rule is the important one and it is almost never implemented elsewhere. It converts every user of the register into a sensor for errors in it. Without it, every downstream system quietly patches its own copy, the copies diverge, and the register slowly becomes the least accurate description of the population in the country, while remaining the most authoritative.

The functional registers make a different error. Because they only see people through one lens, they systematically miss and systematically double-count. A tax file does not know about people below the threshold. An electoral roll does not know about people who moved and did not tell it, so it accumulates ghosts. A health register knows about births in hospitals and not about births at home.

The temptation, once several functional registers exist, is to treat the biggest one as a population register. That is how countries end up governing from the electoral roll or from the ration-card database. Every such use imports that register’s blind spot as national policy.

Company registers and the legal-entity identifier#

A company register does for legal persons what civil registration does for natural persons, with one large difference: the entity is created by the register entry, not merely recorded by it. There is no company before incorporation. The register is constitutive in every case, which is why company-register fraud is so effective — you are not forging a record of something, you are creating the thing.

Some real scale, as of the most recent published figures. The United Kingdom’s Companies House reported a total register size of 5,479,045 companies at 31 March 2026, up 0.94 per cent on the previous year, with 815,277 incorporations and 787,120 dissolutions during the financial year ending 2026, and an average company age on the register of 9.2 years. Of those, 5,114,182 were registered in England and Wales, 275,310 in Scotland and 89,553 in Northern Ireland.

For most of its history Companies House was explicitly not a verifier. It registered what it was given. That changed under the Economic Crime and Corporate Transparency Act 2023: from 18 November 2025, identity verification became a compulsory part of incorporation and of new appointments for directors and people with significant control, with a twelve-month transition through the annual confirmation statement for existing directors and people with significant control. This is a large and instructive change, because it is a register acquiring, late in life, a check it never had, and doing so against a stock of several million existing entries that were never checked.

For identifying legal entities across borders, the relevant instrument is the legal entity identifier, or LEI. Its history is a direct consequence of the 2008 financial crisis, when supervisors discovered that they could not answer the question “who is on the other side of this contract” across jurisdictions, because every firm had its own naming. The G20 and the Financial Stability Board drove the work forward; the first LEIs were issued in December 2012; the Global Legal Entity Identifier Foundation was established in 2014 under Swiss law and operates from Frankfurt, with later offices in the Americas, Japan and Singapore. The code itself is standardized as ISO 17442-1:2020, Financial services — Legal entity identifier (LEI) — Part 1: Assignment, published in August 2020 and confirmed as recently as January 2026. A second part of ISO 17442 covers the use of the LEI inside digital certificates. [UNVERIFIED: the exact publication year and edition of ISO 17442 part 2]

The structure is twenty alphanumeric characters:

LEI layout (ISO 17442)

  positions 1-4    LOU prefix, the issuing organization
  positions 5-6    reserved, currently always "00"
  positions 7-18   entity-specific part, assigned locally
  positions 19-20  check digits, ISO/IEC 7064 MOD 97-10

  example, GLEIF's own published sample:
  506700GE1G29325QX363
  |__| |____________| |_|
  LOU   entity part   check

Three things a practitioner should know about that layout. First, the prefix identifies the issuer, not the country of the entity: an LOU may issue to entities anywhere, so you cannot read nationality out of an LEI. Second, positions 5 and 6 being fixed at “00” is a reservation for future use, not a checksum, and code that treats them as meaningful will break if the reservation is ever spent. Third, the check digits use the MOD 97-10 scheme from ISO/IEC 7064, the same family used by the IBAN, which catches all single-character errors and most transpositions. A validator is a few lines:

def lei_valid(code):
    # 20 chars, uppercase letters and digits only
    if len(code) != 20 or not code.isalnum():
        return False
    code = code.upper()
    digits = ""
    for ch in code:
        if ch.isdigit():
            digits += ch
        else:
            digits += str(ord(ch) - 55)   # A=10 ... Z=35
    return int(digits) % 97 == 1

The Global LEI Index held 3,403,779 LEI records when queried on 18 August 2026. Each record carries what the system calls Level 1 data, “who is who”: the legal name, the legal form, the registered address, the identifier from the local business register, and the record’s own status and next renewal date. Level 2 data, “who owns whom”, records direct and ultimate parent relationships where they exist and are reportable.

The LEI illustrates the chapter’s thesis with unusual clarity. Nothing about a twenty-character string makes it valuable. What makes it valuable is that regulators required it. In the European Union, from 3 January 2018, an investment firm could not execute a reportable trade for a client that had no LEI — the rule the industry summarized as “no LEI, no trade”. Reliance created the register’s value in a single day, by regulation, and the code had been technically identical for five years before that.

There is one honest caveat that practitioners should carry. An LEI confirms that an entity was registered and that its reference data was checked against a local business register at renewal time. It does not confirm that the entity is solvent, honest, or currently trading, and a lapsed record does not mean the entity has ceased to exist. The status field matters as much as the code.

Authoritative source, source of truth, system of record#

These three phrases are used as synonyms in most procurement documents. They are not synonyms, and a system designed by somebody who thinks they are will have an unowned data quality problem within a year.

Authoritative source. NIST Special Publication 800-63-4 defines it as an entity that has access to, or verified copies of, accurate information from an issuing source, such that a credential service provider has high confidence that the source can confirm the validity of the identity attributes or evidence supplied by an applicant during identity proofing. The same document defines an issuing source as an authority responsible for the generation of data, digital evidence such as assertions, or physical documents that can be used as identity evidence. NIST adds a sentence people skip: an issuing source may also be an authoritative source, and authoritative sources are often determined by a policy decision of the agency before they can be used at all. Authority here is conferred by policy, not discovered in the data.

Source of truth is an architecture term, not a legal one. It names the place a fact originates and from which all other copies are supposed to derive. It answers “where did this come from” and implies a direction of flow.

System of record is an operational term. It names the system whose copy governs when two systems disagree, usually because that system is the one the organization will defend in an audit or a court. It answers “whose copy wins”.

Work through Meera Nair. The issuing source for her date of birth is the local registrar in Kollam who wrote the 1994 entry. The authoritative source, for a British bank in 2026, may be neither that registrar nor her Indian certificate but her passport, because the bank’s policy names passports and not foreign civil registers. The source of truth inside the bank is the customer master record populated at account opening. The system of record for the purpose of a regulatory report is whichever system the bank’s compliance function will produce in an inspection, which is quite often a third system that pulls from the second.

Four different answers, all correct, all about the same date. Most identity data disasters are a failure to notice that these are four questions.

Register-to-register reconciliation#

Reconciliation is the process of making two registers agree about a population they both describe. It is a batch problem, a matching problem and a governance problem, in that order of visibility and reverse order of difficulty.

The mechanics are simple to state. Register A produces a file of change events. Register B ingests it, tries to find its own record for each event, and applies the change. Everything hard is inside “tries to find its own record”.

If both registers carry the same unique identifier, the join is exact and the work is trivial. This is why national identity numbers are so attractive to administrators and so contested by everybody else: the identifier is not primarily an identity technology, it is a join key. India’s 2023 amendment collecting Aadhaar numbers at registration is best understood in exactly those terms.

If they do not share an identifier, the join must be made on names, dates and places, and it becomes probabilistic. Chapter 8 covers that mathematics properly; here we only need its consequence. A probabilistic join has four outcomes, not two, and a register that handles only two of them will do damage.

Reconciliation outcomes and correct handling

  matched, agreeing   -> apply, log, no human
  matched, conflict   -> hold, queue for review
  no candidate found  -> hold as unresolved, do not
                         create a new record silently
  two or more equally
  good candidates     -> hold, never pick the first

Failure mode in the field: systems implement rows 1
and 3 only. Row 2 becomes a silent overwrite. Row 4
becomes "pick the highest score", which is a coin
toss with a decimal point on it.

The governance layer is where reconciliation actually fails. Two registers under two ministries have two definitions of a person, two correction procedures, two update cadences and two legal owners, and there is usually no written agreement about which of them wins when they disagree, how long the loser has to fix its record, or who tells the individual. The technical join is a week’s work. The memorandum of understanding takes two years.

The death-record propagation problem#

Death is the reconciliation case that matters most, because the consequence of acting on it is total and because the subject cannot complain.

Here is the shape of the propagation, in the general case:

  bedside / hospital
        |  medical certificate of cause of death
        v
  informant (relative, occupier, official)
        |  reports within statutory period
        v
  LOCAL CIVIL REGISTRAR  -- writes the death entry
        |                        |
        |                        +--> certified copies
        |                             (paper, to family)
        v
  national index / vital statistics
        |
        +--> pensions and benefits
        +--> tax authority
        +--> population register / ID authority
        +--> electoral roll
        +--> passport and licence authorities
        +--> banks, insurers, credit bureaux
                (usually via a shared death file
                 or via the family presenting a
                 certified copy)

Every arrow is a place a death can be lost, delayed, or attached to the wrong person. Two national systems give us real numbers for both failures.

In the United States, the Social Security Administration maintains the Numident, its master file of Social Security number assignments and associated identity records, and publishes death information through the Death Master File, which other federal agencies and external entities use for matching and fraud prevention. The agency’s Office of the Inspector General reported, in an audit released on 2 July 2026 (report 032311, “Beneficiaries Incorrectly Recorded as Deceased”), that SSA posted approximately 5.6 million death records to the Death Master File in calendar year 2025, and that it subsequently determined that 12,504 of them — 0.22 per cent — were erroneous. The same audit reviewed a random sample of beneficiaries recorded as deceased between January 2020 and December 2024 whose death records were later removed, and found that in 45 per cent of the corrected cases technicians had not documented why the death was recorded in the first place or why it was removed, as agency policy required.

Sit with those two numbers. A 0.22 per cent error rate is, by the standards of large administrative systems, respectable. It is also 12,504 living people in one year, in one country, in one register, each of whom had to discover the problem and argue their way out of it. And in nearly half of the corrections, the file does not record why the error happened, which means the error cannot be prevented and cannot be explained to the person it happened to.

In India, the Unique Identification Authority of India has been reconciling Aadhaar against civil registration death records. A Press Information Bureau release of 16 July 2025 reported that UIDAI had obtained approximately 1.55 crore — 15.5 million — death records from the Registrar General of India, and had deactivated approximately 1.17 crore Aadhaar numbers after validation, with civil-registration-based reporting available in 24 states and union territories, and a “Report Death of a Family Member” facility on the myAadhaar portal launched on 9 June 2025 requiring the deceased person’s Aadhaar number and death registration number.

Notice the gap between 1.55 crore records received and 1.17 crore deactivations. That gap is not incompetence; it is what validation looks like. Some records did not match, some matched ambiguously, some matched a record that was already inactive. A system that closed that gap to zero by matching harder would be deactivating living people.

The United Kingdom’s Tell Us Once service is the same problem attacked from the other end. Instead of pushing death records between databases, it collects the notification once at the point of registration, with the informant’s consent, and fans it out to central and local government bodies — pensions and benefits, tax, passport, driving licence, local council services — so that the family does not have to write to each in turn. It is voluntary, it is offered by most local authorities on behalf of the Department for Work and Pensions, and it does not reach private sector holders, who still generally require a certified copy.

Three design lessons come out of all this.

First, deaths propagate as positive assertions and life does not. No register ever sends a message saying “this person is still alive”. The absence of a death record is doing the work of a positive claim, and absence is not evidence.

Second, the propagation is a fan-out with no acknowledgement path. The registrar does not know which of the twelve downstream systems applied the update, or which applied it to the wrong record. Add acknowledgements and you can measure your own propagation; without them you are broadcasting into a void and finding out from complaints.

Third, the reverse path is missing almost everywhere. Systems are built to distribute deaths and not to distribute un-deaths. When a death is retracted, the retraction usually travels by telephone and letter, person by person, at human speed.

When the register is wrong: correction and its latency#

Good registers have always known that they will be wrong, and the law usually specifies the remedy in more detail than the software does.

In England and Wales, section 29 of the Births and Deaths Registration Act 1953 governs it. Subsection (1) sets the rule that no alteration shall be made in any register of live births, still-births or deaths except as authorized by that or another Act. Subsection (2) allows clerical errors to be corrected in the prescribed manner and subject to prescribed conditions by persons authorized by the Registrar General. Subsection (3) is the important one: an error of fact or substance is corrected by an entry in the margin, without any alteration of the original entry, on production of a statutory declaration by two qualified informants or, failing that, by two credible persons. Subsection (4) gives the coroner a route to correct errors in information from an investigation, again by marginal entry without altering the original.

India’s Registration of Births and Deaths Act, 1969 reaches the same design in section 15. Where an entry is erroneous in form or substance, or was fraudulently or improperly made, the registrar may correct the error or cancel the entry by suitable entry in the margin, without any alteration of the original entry, and shall sign the marginal entry and add the date.

Two legal systems, different centuries, same rule: correct by addition, never by deletion, and sign and date the addition. The paper registers solved the audit-trail problem before there were databases to get it wrong in.

The procedure is not the whole story. The latency is.

The first row is the largest and the least discussed. Most register errors are discovered years after they are made, at the moment somebody needs the record — a school admission, a passport application, a pension claim. In our worked example, Meera Nair’s name error was written in 1994 and discovered in 2005, because nothing before 2005 required the certificate.

The last row is the one that decides whether the correction actually helps. A corrected register with uncorrected copies downstream produces the worst possible situation for the individual: they now hold a document that disagrees with the records the world is actually using, and they are the only person motivated to fix each of them, one at a time, by telephone.

For a wrongly recorded death, the recovery path is worse still, because the individual has to prove a negative to institutions whose systems have no field for “alive again”. Practical advice, learned from people who have done it: get the register corrected first, obtain multiple certified copies of the corrected entry on the same day, and then work the downstream list in order of financial damage, keeping a dated log of every call, because the second and third institutions will each ask you to prove that the first one has already accepted it.

Coverage: who is missing from the books#

The United Nations target is SDG 16.9: by 2030, provide legal identity for all, including birth registration. Its indicator, 16.9.1, is the proportion of children under five years of age whose births have been registered with a civil authority, by age, with UNICEF as custodian agency. There is a second, less-quoted indicator that measures the system rather than the child: indicator 17.19.2 counts the proportion of countries that have conducted at least one population and housing census in the last ten years and have achieved 100 per cent birth registration and 80 per cent death registration.

That eighty per cent for deaths, against one hundred for births, is an admission written into the target itself. Deaths are harder to register than births, everywhere.

The current picture, with dates attached to every number.

UNICEF’s 2024 update, The Right Start in Life, puts about 150 million children under five — around two in ten — as unregistered, with global coverage close to 77 per cent. The regional spread is enormous: Australia and New Zealand and Europe and Northern America at 100 per cent, Latin America and the Caribbean at 95 per cent, Eastern and South-Eastern Asia at 94 per cent, sub-Saharan Africa at 51 per cent, and Oceania excluding Australia and New Zealand at 26 per cent.

The World Bank’s Identification for Development programme measures the adult end of the same problem. Its 2021 Global Dataset put just under 850 million people without official proof of identity. The refreshed estimate published on 30 October 2025, drawing on the ID4D-Findex 2025 data, puts the figure at more than 800 million, down from 850 million in 2021 and just over 1 billion in 2017 — about one person in ten. Over half of them are children whose births were never registered. Adult foundational ID coverage in sub-Saharan Africa stood at 81 per cent, up from 72 per cent in 2017, with eight countries below 70 per cent. The same release reported that 2.8 billion people lacked access to a digital identity usable for online transactions in 2025, down from 3.3 billion in 2021, and that about 41 per cent of adults without an ID said they could not buy a SIM card.

Deaths are worse counted than births, and by a wide margin. The World Health Organization’s civil registration and vital statistics material states that an estimated two-thirds of deaths worldwide are never registered and are therefore absent from the vital statistics system. [UNVERIFIED: the date and underlying study for the WHO two-thirds unregistered deaths estimate]

Put those numbers next to the propagation problem and you get the real state of the world’s registers: hundreds of millions of people who cannot prove they exist, and a majority of deaths that no register ever learns about. Both failures land on the same households.

Register design: append-only, corrections as records#

Everything above converges on a small set of design rules. They are not exotic. They are what the 1953 and 1969 Acts already require, expressed as data engineering.

Rule one: append-only. A register table takes inserts. It does not take updates or deletes on recorded facts. If a fact changes, or was wrong, you insert a new row that says so and refers to the old one.

Rule two: corrections are records. A correction is itself a registrable event with its own date, its own authority, its own evidence reference and its own author. “The name was corrected” is a fact about the register, and it belongs in the register.

Rule three: two timelines, never one. Every fact needs the date the fact was true from, and the date the register learned it. The database literature calls these valid time and transaction time, and a store that keeps both is bitemporal. Without both, you cannot answer the two questions that every dispute asks: what was true on the day, and what did we believe on the day. Those are different questions and they have different answers, and an organization that cannot separate them cannot defend its own decisions.

Rule four: the current view is derived, not stored. What the world reads is a projection built from the event history, not a mutable row. If the projection and the history disagree, the history wins and the projection is rebuilt.

Rule five: no silent success. Every write records who authorized it, under what power, on what evidence. A row with no author is not a register entry; it is a rumour with a primary key.

Here is what the Meera Nair record looks like under those rules. The entry is the 1994 birth registration; the correction is 2005; the erroneous death report and its retraction are 2026.

{"event": "birth_registered",
 "entry_id": "KL-KLM-1994-0031477",
 "valid_from": "1994-03-14",
 "recorded_on": "1994-03-19",
 "name_given": "Meena",
 "name_family": "Nair",
 "informant": "father",
 "authority": "registrar, Kollam municipality"}

{"event": "entry_corrected",
 "entry_id": "KL-KLM-1994-0031477",
 "corrects_field": "name_given",
 "from": "Meena", "to": "Meera",
 "valid_from": "1994-03-14",
 "recorded_on": "2005-06-22",
 "evidence": "affidavit + school record",
 "authority": "registrar, s.15 RBD Act 1969"}

{"event": "death_reported",
 "entry_id": "KL-KLM-1994-0031477",
 "valid_from": "2026-02-03",
 "recorded_on": "2026-02-09",
 "source": "batch file, external register",
 "match_score": 0.91,
 "match_basis": "name + dob, no identifier"}

{"event": "death_retracted",
 "entry_id": "KL-KLM-1994-0031477",
 "retracts_event": "death_reported",
 "recorded_on": "2026-03-30",
 "reason": "matched wrong subject",
 "authority": "registrar, on production of
               subject in person"}

Read that history and notice what it lets you say. The name was recorded as Meena and corrected to Meera, and the correction is backdated to 1994 for validity while being recorded in 2005 — so a certificate issued in 1999 saying “Meena” was correct at the time it was issued and is not evidence of anybody’s dishonesty. The death was reported on a probabilistic match with no shared identifier and a score of 0.91, which is the whole explanation of the failure and which a system that overwrote the record would have destroyed. And the retraction names its own authority.

Now the counter-example. In a system with one mutable row per person, the same story leaves this behind:

person_id  name    dob         status
31477      Meera   1994-03-14  ALIVE

That is all of it. No 1994, no Meena, no correction, no 0.91, no reason, no author. Everything a person would need to defend themselves, and everything an engineer would need to fix the matching rule, has been erased by an UPDATE statement. This is the single most common serious defect in identity systems built in the last twenty years, and it is committed by teams who would never dream of altering a paper register.

One caution against over-reading the rule. Append-only is about the integrity of the record, not about hoarding personal data forever. Retention limits, erasure rights and archival transfer are real legal obligations, and they are satisfied by deleting or transferring whole records under a recorded, authorized process — which is itself an event — not by silently editing fields. There is a genuine tension here between an append-only design and a right to erasure, and honest practitioners resolve it case by case rather than pretending it does not exist. It is also worth saying plainly that append-only does not require a blockchain; a permissioned append-only table with signed entries and an authority column has been sufficient for civil registration since 1538 and remains sufficient now. Distributed ledgers for civil registration are, as of August 2026, mostly pilots and marketing rather than production practice.

The worked reconciliation, end to end#

Let us run Meera Nair’s 2026 incident all the way through, with the numbers, because the abstract description hides where the damage actually comes from.

A national register of deaths sends a weekly change file to a clearing service used by banks. In the file for the week of 9 February 2026 there is a record for a woman with the family name Nair, given name Meera, date of birth 14 March 1994, place of birth recorded as a district in Kerala. The subject of that record is a different woman with the same name and the same birth date. There is no shared identifier in the file, because the two registers were built in different decades by different ministries.

The clearing service matches on name plus date of birth plus a coarse place field. Our Meera scores 0.91 on that comparison. The service’s threshold for automatic application is 0.90. It applies.

Three consequences follow within seventy-two hours, none of which involve any further check. The bank marks the account deceased and stops outgoing payments. The credit bureau adds a deceased flag, which is reported to every lender that queries it. The employer’s payroll, subscribing to the same feed for pension administration, suspends the salary.

Meera discovers the problem on 12 February when a card is declined. She telephones the bank. The bank tells her, correctly by its own rules, that it cannot reverse a deceased marking on a customer’s say-so and that she must have the source corrected.

Eighty-six days from a matching score of 0.91 to the last flag clearing, of which the register’s own correction was forty-nine days and the downstream propagation of the correction was another thirty-seven. The error propagated in three days. The correction took twenty-eight times as long.

Where should the design have stopped this? Not at the correction procedure, which worked. Four earlier places, in order of cost.

At the source, by publishing a stable identifier in the change file, so that the match is exact and the score never arises. At the clearing service, by treating a match on name and date of birth alone, with no identifier, as insufficient for an automatic destructive action — the rule should be that the strength of evidence required scales with the harm of the action, and deceased is the most harmful flag in the file. At the receiving institutions, by holding the flag in a pending state for a short window and attempting one contact before acting. And in the retraction path, by making the clearing service capable of publishing a retraction with the same reach as the original assertion, which is the part almost nobody builds.

What real registers actually fail on#

If you audit register systems for a while, the same defects recur, and they are rarely the ones the risk register lists.

The first is silent update. A field changes and nothing records that it changed, who changed it, or what it was. Everything else on this list is survivable; this one destroys the ability to investigate any of the others.

The second is one-way propagation. The system can distribute a fact but cannot distribute its withdrawal, so retractions travel by telephone.

The third is the missing “unknown” state. Systems model true and false and have no way to say “we do not know”, so absence of evidence becomes a negative assertion. This is the coding error underneath both “not in the register, therefore not entitled” and “no death record, therefore alive”.

The fourth is threshold worship. A single similarity threshold governs every action, so a match good enough to send a leaflet is also good enough to stop a pension. The strength of evidence must scale with the harm of the action.

The fifth is copy drift. Downstream systems patch their local copies instead of reporting errors back, so the register is slowly outvoted by its own children. The Dutch report-back rule, under which a body that suspects an error tells the owning register instead of patching its own copy, is the clearest published answer, and it is still uncommon.

The sixth is coverage blindness. The system measures its accuracy over the people it contains and has no measurement at all for the people it does not contain, which is exactly the population that policy most affects.

The seventh is the certificate as a live object. A certified copy is a snapshot with a print date, and it circulates for decades. Any process that treats a paper certificate as current is reading a value that could be thirty years stale. Whether a person may hold several different identities across contexts on purpose, and what it costs to link them, is the subject of chapter 10.

9.98 Common wrong ideas#

Wrong: A register stores facts. Right: A register stores an authority’s dated assertion about a fact, made on the basis of a report from a named informant; the fact and the assertion are separate things, and the whole discipline of correction exists because they can diverge.

Wrong: If it is in the register, it is true. Right: If it is in the register, an authorized person recorded it as true on a stated date using stated evidence; registers are faithfully wrong whenever their inputs are wrong, as the nineteen years it took Lal Bihari to annul his 1975 death record demonstrate.

Wrong: Authoritative source, source of truth and system of record mean the same thing. Right: They answer three different questions — whom policy permits you to believe, where a fact originates and flows from, and whose copy governs a dispute — and a single fact routinely has different answers to all three.

Wrong: A birth certificate proves who you are. Right: It is evidence that a birth with those particulars was registered; connecting that historical event to the adult standing in front of you is a separate step that the certificate cannot perform, which is why it is called a breeder document and treated as a starting point rather than an identification.

Wrong: No death record means the person is alive. Right: It means no death has been reported to that register, which is a different claim entirely; the World Health Organization estimates that most deaths worldwide are never registered at all, so absence of a death record is weak evidence in exactly the places where it is most relied on.

Wrong: Correcting the register fixes the problem. Right: Correcting the register is the first hop of many, and in our worked example took forty-nine days out of an eighty-six day recovery, with the remaining thirty-seven days spent pushing the correction into copies that had accepted the error in three days.

Wrong: An append-only design means you can never delete personal data. Right: It means you never silently edit a recorded fact; deletion, retention limits and erasure rights are satisfied by removing or transferring entire records under an authorized, recorded process, which is itself an event in the log.

Wrong: A company register verifies the companies on it. Right: Historically most did not verify anything beyond the completeness of the forms; the United Kingdom only made identity verification compulsory for new directors and people with significant control on 18 November 2025, against an existing register of 5,479,045 companies at 31 March 2026 whose earlier entries were never checked.

Wrong: A legal entity identifier proves an entity is legitimate. Right: It proves the entity was registered and its reference data was validated against a local business register at issue or renewal; it says nothing about solvency, honesty or current trading, and the record status field carries information the code itself does not.

Wrong: Two registers can be reconciled by matching names and dates of birth. Right: Without a shared identifier the join is probabilistic, with four possible outcomes rather than two, and treating an ambiguous or conflicting match as a clean one is the single most common cause of a living person being switched off.

9.99 Chapter summary in 20 lines#

1. A register is an authority’s dated, signed assertion that a stated fact is true, made on the strength of a report from a named informant.
2. The value of a register comes entirely from who is willing to act on it, which makes reliance a legal and social achievement rather than a technical one.
3. The United Nations defines civil registration as the continuous, permanent, compulsory and universal recording of vital events, and each of those four words is a demanding requirement.
4. A birth entry asserts that an event occurred with stated particulars; it does not assert that the holder of the certificate is the person the entry describes.
5. A marriage entry usually creates the legal status rather than merely recording it, which is why forging one manufactures a status instead of misreporting a fact.
6. A death entry is the only common register entry that removes rights rather than conferring them, and it is acted on faster and questioned less than any other.
7. Population registers try to describe everybody, while functional registers such as tax, health and electoral rolls describe only the people who interact with one function, and each fails differently.
8. Denmark’s Central Person Register was established on 2 April 1968 and Sweden introduced a nine-digit personal number in 1947, both making the population register the spine of the state.
9. Company registers are constitutive, because the entity does not exist before the entry, which is why registration fraud creates rather than merely misdescribes.
10. The legal entity identifier is a twenty-character code defined by ISO 17442-1:2020, with a four-character issuer prefix, two reserved characters, twelve entity characters and two ISO/IEC 7064 MOD 97-10 check digits.
11. There were 3,403,779 LEI records in the Global LEI Index when queried on 18 August 2026, and the system’s value came from regulators requiring it rather than from the code itself.
12. Authoritative source, source of truth and system of record answer three different questions, and NIST Special Publication 800-63-4 makes clear that authority is conferred by policy decision.
13. Register-to-register reconciliation is exact when a shared identifier exists and probabilistic otherwise, with four outcomes to handle rather than two.
14. The Social Security Administration posted about 5.6 million death records in calendar year 2025 and later found 12,504 of them, or 0.22 per cent, to be erroneous.
15. India’s UIDAI obtained about 1.55 crore death records from the Registrar General of India and deactivated about 1.17 crore Aadhaar numbers after validation, as reported in July 2025.
16. Both English law, in section 29 of the Births and Deaths Registration Act 1953, and Indian law, in section 15 of the Registration of Births and Deaths Act 1969, require corrections to be made by dated marginal entry without altering the original.
17. Correction latency is dominated by the two ends of the chain: discovering the error, and pushing the correction into copies that accepted the error far faster than they will accept its withdrawal.
18. About 150 million children under five were unregistered as of the 2024 UNICEF update, and more than 800 million people lacked official proof of identity on the World Bank’s 2025 estimate.
19. Late registration is deliberately harder than timely registration, which protects the register’s accuracy and simultaneously walls out the people who most need to be let in.
20. A register should be append-only, should record corrections as new dated and authorized records, should keep both the date a fact was true and the date it was learned, and should never silently overwrite anything.

Chapter sources: the United Nations Legal Identity Agenda pages of the UN Statistics Division for the UN operational definition of legal identity, the definition of civil registration as continuous, permanent, compulsory and universal recording of vital events, the establishment of the UN Legal Identity Agenda Task Force from September 2018 following the Secretary-General’s Executive Committee decision of January 2018 with thirteen UN agencies under the chairmanship of UNDP, UNICEF and UN DESA, and the wording of SDG target 16.9 and indicator 17.19.2 on 100 per cent birth registration and 80 per cent death registration; the UN Statistics Division metadata for SDG indicator 16.9.1, proportion of children under 5 years of age whose births have been registered with a civil authority, by age, with UNICEF as custodian agency; UNICEF’s global birth registration database and the 2024 update The Right Start in Life: Global levels and trends in birth registration, for the figures of about 150 million unregistered children under five, about 77 per cent global coverage, 51 per cent in sub-Saharan Africa, 26 per cent in Oceania excluding Australia and New Zealand, 95 per cent in Latin America and the Caribbean and 94 per cent in Eastern and South-Eastern Asia; the World Bank Identification for Development Global Dataset 2021, Volume 1, for the estimate of just under 850 million people without official proof of identity in 2021, and the World Bank blog post of 30 October 2025, Global progress in identification: 3 findings from the latest data, drawing on the ID4D-Findex 2025 database, for more than 800 million people without official proof of identity, just over 1 billion in 2017, over half of them children, 81 per cent adult coverage in sub-Saharan Africa against 72 per cent in 2017, 2.8 billion without online digital ID in 2025 against 3.3 billion in 2021, and 41 per cent of adults without an ID reporting they cannot buy a SIM card; the World Health Organization’s civil registration and vital statistics pages for the estimate that two-thirds of deaths are never registered; ISO 17442-1:2020, Financial services — Legal entity identifier (LEI) — Part 1: Assignment, first edition, published August 2020 and confirmed January 2026, together with GLEIF’s published material on the LEI code structure, the four-character LOU prefix, the reserved characters 5 and 6, the entity-specific characters 7 to 18 and the ISO/IEC 7064 MOD 97-10 check digits at positions 19 and 20, the sample code 506700GE1G29325QX363, the issuance of the first LEIs in December 2012, the establishment of GLEIF in 2014 under Swiss law with its office in Frankfurt, and Level 1 and Level 2 reference data; the GLEIF public API, queried on 18 August 2026, for the figure of 3,403,779 LEI records; NIST Special Publication 800-63-4 and 800-63A-4 for the definitions of authoritative source and issuing source and the note that authoritative sources are determined by a policy decision; the Births and Deaths Registration Act 1953 of England and Wales, section 29, subsections (1) to (4), for correction by marginal entry on a statutory declaration by two qualified informants without alteration of the original entry, and the Births and Deaths Registration Act 1836 which brought civil registration into effect on 1 July 1837 and created the General Register Office, following Thomas Cromwell’s 1538 injunction requiring parish registers; the Registration of Births and Deaths Act, 1969 of India, sections 8, 13 and 15, for the prescribed reporting period, the graded late-registration ladder of thirty days with a late fee, up to one year with written permission and a notarized affidavit, and beyond one year only on the order of a magistrate of the first class, and for correction or cancellation by suitable signed and dated entry in the margin without alteration of the original entry; PRS Legislative Research on the Registration of Births and Deaths (Amendment) Bill, 2023, introduced in the Lok Sabha on 26 July 2023, passed by the Lok Sabha on 1 August 2023 and the Rajya Sabha on 7 August 2023, providing for a national database, the birth certificate as the single document proving date and place of birth for listed purposes, and the collection of Aadhaar numbers; the Social Security Administration Office of the Inspector General audit Beneficiaries Incorrectly Recorded as Deceased, report 032311, announced 2 July 2026, for approximately 5.6 million death records posted to the Death Master File in calendar year 2025, 12,504 or 0.22 per cent subsequently determined erroneous, a sample drawn from January 2020 to December 2024, and the finding that documentation was missing in 45 per cent of corrected cases; the Press Information Bureau release of 16 July 2025 on UIDAI measures, for approximately 1.55 crore death records obtained from the Registrar General of India, approximately 1.17 crore Aadhaar numbers deactivated after validation, coverage of 24 states and union territories, and the Report Death of a Family Member service launched on 9 June 2025; the GOV.UK guidance on the Tell Us Once service operated by local authorities on behalf of the Department for Work and Pensions; Companies House statistical release Companies register activities April 2025 to March 2026, for a total register of 5,479,045 companies at 31 March 2026, 815,277 incorporations, 787,120 dissolutions, an average company age of 9.2 years and the country breakdown of 5,114,182, 275,310 and 89,553; the Companies House outline transition plan for the Economic Crime and Corporate Transparency Act 2023 and the Companies House announcement that identity verification became compulsory for new directors and people with significant control from 18 November 2025 with a twelve-month transition for existing appointments; the National Audit Office report Handling of the Windrush situation, published 5 December 2018, for the 164 people who arrived before 1973 and were detained or removed since 2002 and the 2,658 people documented by the taskforce to 30 September 2018, together with the Windrush Lessons Learned Review by Wendy Williams published in March 2020; the account of the Danish Central Person Register established on 2 April 1968 with the enabling bill passed by the Folketing on 29 May 1968 and signed in June 1968, and of Sweden’s nine-digit personnummer introduced in 1947; the Netherlands system of base registries as described in the Dutch submissions to the European interoperability portal, comprising twelve base registers with mandatory use of authentic data and a duty to report suspected errors back to the owning register; and the published record of the Lal Bihari case in Uttar Pradesh, dead on the record from 1975 until the annulment in 1994, founder of an association of people recorded as dead, and recipient of the Ig Nobel Peace Prize in 2003.