Digital preservation churn

Why does a digital archive need constant work to stay readable when a clay tablet needs none?

A clay tablet from Mesopotamia sits in a drawer and does nothing. Nobody budgets for it annually. Dig it up in another thousand years and the marks are still marks; the only requirement is somebody who reads the script. A digital archive of the same content, left alone for thirty years, is quite likely to be unusable — and a well-run one costs money every single year forever.

That looks like a straightforward defeat for the newer medium, which is odd, because digital copying is perfect in a way no scribe ever managed. Both things cannot be simply true. So what is the digital archive actually spending its money on?

Begin by asking what a tablet is. The mark and the carrier are the same object: a wedge pressed into clay is simultaneously the storage and the encoding. There is exactly one interpretive layer between the object and a reader — knowledge of the script — and it is a layer held in human heads and in other tablets, which is why the script could be recovered after being lost.

Now count the layers under a digital file. Magnetised domains or trapped charge, read by a drive that understands a physical encoding; blocks assembled by a filesystem; bytes interpreted as a format; the format understood by an application, itself running on an operating system and instruction set. Every one of those is a convention rather than a physical fact, maintained by somebody with their own reasons.

Here is the step that matters: those layers fail independently, and only one of them is what people usually mean by decay. Carrier decay is real — magnetic binder degrades, optical dye fades, flash cells leak charge. But suppose you defeat it entirely. You still have a file that opens in nothing, because the format was proprietary and its vendor moved on, or the codec needed a licence server that no longer answers, or the document is a database export whose schema lived in a manual nobody kept.

So the archive is fighting two clocks: one physical and roughly predictable, the other social, which can fire at any moment while your media is in perfect condition.

And now the compensating asymmetry, which is why the situation is not hopeless. Digital copies are exact. A tablet copied by hand accumulates error; a file copied with verification does not. That single property converts preservation from conservation — slowing a decay you cannot stop — into repetition: copy it forward before the carrier or the format expires, verify by checksum, repeat. The churn is not a flaw sitting on top of an otherwise stable medium. It is the medium's actual preservation strategy, and it is available precisely because copying is lossless.

The falsification test. If carrier decay were the whole story, then files on intact, readable media would open fine, and the failures would correlate with the physical age and condition of the medium. The refuting observation for that account — and it is observed constantly in practice — is the file that reads off the disc bit-perfectly, passes its checksum, and opens in nothing: an early word-processor document, or a multimedia work whose runtime no longer exists. Bits intact, meaning gone. Conversely, an account resting on format obsolescence alone would be refuted by an archive whose formats are all open and standardised failing anyway — which also happens, when the tapes were never refreshed. Both clocks are real; neither alone explains the failures.

Four things are worth separating, and the last is where most of the confusion lives.

First, this is a different failure from the one the companion piece on silent data corruption treats. There the bits themselves change and the copies disagree; here the bits are perfect and the trouble is that nothing can interpret them. Fixity checking catches the first and is entirely blind to the second.

Second, the two standard responses are not rivals. Migration converts the content into a current format, accepting that some properties may not survive the conversion. Emulation keeps the original bits and reconstructs the environment that read them, preserving behaviour — which matters enormously for anything interactive — at the cost of maintaining the emulator instead. Serious archives do both, choosing by what the object is.

Third, "digital is forever" is simply false, and its cousin — "the cloud is a backup" — is worse, because it converts a technical dependency into a contractual one. A service can end. The bits' survival then depends on somebody's business decisions.

Fourth, and this is the point I would most want to land: format churn is not a property of digital technology. Nothing about bits requires the format to move. It moves because the entities defining formats have an interest in it moving — upgrade cycles, differentiation, lock-in. The affordance of the technology is perfect copying, which favours permanence; the incentive of whoever owns the format favours change. Confusing those two produces the fatalistic version of this story, in which churn is an inherent cost of being digital. It is not. It is why open, documented, widely implemented formats are the single most effective preservation decision available — they decouple the file's readability from any one party's commercial interest. The technology did not create the problem; the ownership of the encoding did.

I will not quote a media lifetime figure here. Published estimates for magnetic tape and optical disc longevity vary by an order of magnitude depending on formulation, storage conditions and who funded the study, and any single number would be more misleading than useful.

• How distributed schemes such as LOCKSS make survival a property of a network rather than of any single custodian.

Every term the collection defines is gathered in the glossary.