Thirty-Seven Million Descriptions

On 15 June 1215 King John sealed Magna Carta. Four exemplifications of that charter survive. Four exemplifications of the 1297 confirmation also survive, and between the two sit reissues of 1216, 1217 and 1225.

The usual name makes this sound like one document with a long afterlife. The record does not. The 1215 text was annulled within weeks. The 1297 text went into England’s statute law, and three of its clauses are still in force: the liberty of the Church, the customs of London and other towns, and due process against arbitrary imprisonment.

One name. Five instruments. Eight surviving exemplifications across two versions.

Every one of those is a true count. The error starts when a total arrives without saying what was made countable.

One name, five instruments #

Magna Carta is not a hard case. It is a clean demonstration of something record systems usually hide.

The 1215, 1225 and 1297 texts differ materially in content, and the 1215 charter has no direct legal descendants of its own. Quote it as though it were the operative text and you supply words a court would not recognise as current law. The version number is not a bibliographical courtesy added afterwards. It fixes the object of the claim.

Physical survival is a separate unit again. The British Library, Lincoln Cathedral and Salisbury Cathedral hold the four surviving 1215 exemplifications between them. A catalogue can describe each parchment, its seal, its condition, its digital image and its transcription, which makes the catalogue fuller and leaves the number of surviving originals at four.

The Dead Sea Scrolls show the same split from the other end. The corpus runs to roughly 981 separate manuscripts, mostly leather, ranging from near-complete scrolls to fragments a few centimetres across, and it is catalogued piece by piece. Counting descriptions is genuinely useful there, because the things described are so unlike each other. It is still not a count of manuscripts, or of texts, or of images.

Every archive total picks a unit #

The Vatican Apostolic Archive reports about 85 kilometres of shelving. The US National Archives reports more than 13 billion pages. The Library of Congress reports roughly 173 million items.

Those are not rival estimates of a common substance. Shelf length measures occupied storage. A page count measures surfaces. An item count measures whatever an institution has decided its members are. One bound volume is simultaneously a single item, several hundred pages and a few centimetres of shelf.

The UK National Archives makes the last unit explicit, and the explicitness is the point. Its Discovery catalogue held more than 37 million item descriptions as of October 2024, of which more than 9 million resolve to material available for direct download. The first figure says how many things the catalogue can name. The second says how many of those names lead anywhere.

Neither says the archive acquired 37 million objects.

This is why a catalogue migration can produce a gratifying coverage chart. A volume-level record is split into file-level records. A file gains separate rows for a scan, a transcription and a correction. The previous description is kept as a historical version. The new total is real as a count of records, and it has quietly stopped answering the old question.

The discovery trail is separate #

There is a nearby case that gives the same rising graph an entirely different meaning, which is that more material has actually turned up.

Known surviving Dunlap broadsides of the Declaration of Independence went from 14 in 1949 to 26 today, the most recent identified in the UK’s National Archives in 2009. Those increases are not the result of splitting catalogue rows. Each one is a physical object entering the known-survival record for the first time.

The distinction is not that one number is older or more respectable. It is that the discovery trail exists independently of the description trail. A new broadside is another object in the world. A new catalogue field is another assertion about an object that may have been sitting in a known collection for decades.

Financial data needs the same separation. A single filing is one economic event, several document versions, many line items and an expanding population of vendor records. A history that counts every row is a report about the vendor’s representation. A history that collapses rows without keeping versions destroys the information needed to reconstruct what a backtest actually consumed.

What disappears is the denominator #

Archives hold one advantage a data vendor may not: the accession and custody trail stays inspectable when the catalogue changes underneath it.

The 1297 Magna Carta on display at the National Archives is privately owned by David Rubenstein, who bought it at Sotheby’s in December 2007 and lent it. Ownership, custody, version and description are four records there, and none of them answers for the others.

A financial master table can simply be overwritten. The newer catalogue is better described, more available and larger on every dashboard measure, and the prior denominator has left with the old rows. Nothing downstream can then determine whether the coverage gain was a new issuance, a recovered object, a restatement, or a new way of describing what was already in hand.

Thirty-seven million Discovery descriptions tell a reader how much the catalogue can name.

Four 1215 originals tell a reader what survived. Substitute the first for the second and the chart goes up while the denominator quietly leaves the building.