What Actually Destroys A Digital Record

ANothing is easier to copy than a digital file, and nothing has turned out to be harder to keep. A sheet of paper fades slowly and stays partly legible while it does so; a file is either present and perfect or absent altogether, and the passage between those two conditions can take an afternoon. The point is usually made with the Merridew Survey, a national record of land use assembled in 1986 and issued on videodiscs that no ordinary machine could read fifteen years later, while the parish registers it was built to replace sat unharmed on their shelves. Episodes of that kind are repeated as warnings about technology, and they have shaped how money for preservation is requested and spent. Whether the technology is what actually destroys digital records, or merely the part of the process that is visible once they are found to be gone, has become the central dispute of the field.

BSoren Nagy, an archival scientist at the Thurloe Centre for Recorded Memory, holds that the visible part is the wrong part. Between 2011 and 2019 her team followed 340 collections kept by universities, broadcasters and local authorities, and examined every item that staff reported as no longer readable, 8,600 of them in all. In only nine per cent of those cases was the obstacle a format that no available software could open. The remainder had gone when a server was decommissioned without a receiving system being named, when a department was reorganised, or when the one person who knew where a set of tapes lived retired. Nagy's summary is that the file which survives is the file somebody is paid to move. She does not deny that formats age. She argues that ageing formats are seldom granted the time to become a problem, because the collection around them is dispersed first.

CThe claim yields a prediction, and Nagy has tested it. If attention rather than decay does the damage, collections whose funding has never been interrupted ought to hold their oldest material however obsolete its format may be. Among the twelve archives she could identify with an unbroken budget line since the 1990s, ninety-six per cent of the material deposited in that decade was still being opened and read, although almost all of it sat in formats that no consumer program now recognises. Retention at that level was confined to bodies in which a named post carried the duty; where the work was shared out among existing staff, the figure fell to about half. Nagy offers the result as a limit on the technological account rather than a refutation of it, and observes that twelve is a small number of archives on which to rest anything at all.

DGerard Ruiz, who runs the emulation laboratory at the Cape Verity Institute of Computing History, regards the reasoning as circular. A study which counts only what institutions themselves report, he says, measures what institutions notice, and a body that has quietly lost a format has no means of reporting it. His own group rebuilt in software nine machine environments from the 1980s and 1990s and used them on material that three archives had written off; sixty-two per cent of it came back intact and a further quarter in a degraded but usable state. Emulation, on this account, converts obsolescence from a permanent loss into a postponed expense. Ruiz concedes two points. All nine environments belonged to one family of processors, so nothing follows about machines outside it, and the expense of the work was estimated indirectly, from hours logged by his engineers, because no agreed method of costing recovery exists.

EThe quarrel has consequences well outside the reading room. Deposit rules introduced in 2021 in several countries oblige any body holding public records to file a preservation plan, and inspectors have begun rejecting plans that name no responsible officer. Commercial suppliers of archival storage quote Nagy's findings in their sales material, on the reasoning that a service bought under contract cannot be forgotten when a department is reorganised. Nagy points out, awkwardly for her own case, that her argument does license that inference, and that she has no evidence at all about what becomes of such contracts when a supplier fails. Ruiz replies that those same inspectors treat every successful recovery as proof that nothing is ever truly lost, which makes laboratories like his harder rather than easier to fund. On one point the two agree: the habit of publicising individual rescues, instead of counting what ordinarily disappears, has served the subject badly.

FSomething like a settlement is now taking shape, in which the two accounts are given separate jobs instead of being ranked against each other. Across the first decade of a record's life the dangers are physical and technical, failing media and formats abandoned by their vendors, and these are the problems that emulation and routine copying handle well. Over longer periods the threat is that the context which made a file intelligible disappears: the codes, the abbreviations, the knowledge of why the record was made at all, none of which any machine can rebuild from the bits themselves. What neither side can settle is the fate of material that never entered an institution, since the only account of it is held by the very systems that lose it. Both reject the assumption behind the warning stories, that a technology which failed is the same thing as a memory that was never kept.