This morning's feed delivered a curious genre collision: in one corner, a debate about "tombstones" for deleted data in search indexes; in the other, the story of a Martian lander found eleven years later in someone else's image archive. Both plots revolve around the same idea we prefer to ignore: information doesn't survive on its own. It survives only if someone deliberately preserves it. And right now, in humanity's most "documented" era, we risk leaving behind fewer traces than Bronze Age civilizations.
The starting point was a thesis from an engineering discussion: when data in a system gets deleted, it's more honest to leave a "tombstone"—a marker saying "something was here and it's been deleted"—instead of silent disappearance. Beautiful argument: silent deletion is indistinguishable from a fact that never existed, and sooner or later this ends in confident hallucination.
Right next to it in the morning digests lay the story of the Beagle 2 probe: a British lander touched down on Mars in 2003, went silent—and was only found in 2015 when an engineer manually reviewed archive images from an orbital spacecraft. What saved it wasn't telemetry but the discipline of someone else's archive: HiRISE images were stored in order, and the "needle" could be found.
The combination of these two stories gave the theme. It'll be amusing if our century—the age of terabyte archives, cloud backups, and "everything's recorded, right?"—goes down in history as the digital equivalent of the early Middle Ages: an era from which future historians will inherit fragments and gaps.
Let's start with the most resonant and still-fresh case. In 1986, the BBC marked the 900th anniversary of the Domesday Book—England's famous land survey of 1086—with an ambitious multimedia project. The country's residents told their own stories: a million participants, 250,000+ place names, 25,000 maps, 50,000 photographs, 3,000 data sets, and 60 minutes of video, recorded on 12-inch laser discs. Budget: £2.5 million, serious money by 1986 standards. Technologically, it was one of the world's first interactive hypermedia products: fifteen years before the mass web.
The problem emerged quickly. The discs could only be read on BBC Micro computers with special boards and Philips LV-ROM drives—a bespoke setup nobody manufactured. By 2002, when the story hit the press hard, the system was already unreadable: almost no hardware remained, the software was dead, and some discs had physically degraded. Robin McKie delivered the verdict in The Guardian's headline: "Digital Domesday Book lasts 15 years not 1000." Computer scientist Paul Wheatley formulated the irony that's been classic ever since: "It's lucky Shakespeare didn't write on an old PC."
The tastiest part here isn't the loss itself but what happened next. A team from the UK National Archives, partnering with the CAMiLEON research project (Michigan + Leeds), launched a rescue operation: emulating the ancient system on modern hardware, extracting data from the last working drive, debating whether to preserve "clean" data or include the "corrupted picture" of the original experience. In December 2002, they demonstrated the emulator publicly—the project was saved literally on the last gasps of the original hardware. The rescue operation's authors honestly admitted: they got lucky, and repeating this adventure as routine practice is impossible. The rescue happened not because it was planned that way, but because enthusiasts showed up in time.
This story's good because there's not a single "savage with a stone axe" in it: educated people who spent millions arranged the loss, and enthusiasts—also in another century—saved it. Perfect entry to the problem: digital survival isn't a property of the medium, but a property of the institution around the medium.
If Domesday is paradox, what follows are chronically sad cases. Genre classic—the tapes from the first Moon landing. Original recordings of slow-scan television (SSTV) from Apollo 11, stored in NASA archives, were erased and recorded over—in an era of magnetic tape scarcity, routine satellite data was written over old material. NASA specialist Dick Nafziger later explained: by the standards of the time, nobody did anything wrong; tape was an expensive consumable. Humanity made a giant leap, documented on film—and erased the original to save on supplies. Today we watch the first steps on the Moon through restored broadcast-quality copies, while the original, comparable in clarity, is lost forever.
The second case, wrapped in beautiful legend—the 1960 US Census. Urban legend says: census data was recorded on tapes for UNIVAC computers, the machines became obsolete, "only two reading devices remain in the world—one at the Smithsonian, the other in Japan," and the data perished. The National Archives specifically dismantled this myth—and the picture turned out subtler. Valuable data planned for preservation was saved: by 1979, the Census Bureau had rewritten 640 of 642 tapes to a modern format. But two tapes physically disappeared (about 0.5% of valuable records), and another less than 0.2% of records couldn't be read due to wear. The culprits weren't "obsolete format" but failed inventory and lazy accounting. The UNIVAC myth covers up banal organizational negligence—which, honestly, makes the story even more instructive.
Third case—the most massive and fresh. In 2019, MySpace officially admitted: during server migration, all music files uploaded by users from 2003 to 2015 were damaged and not transferred. Twelve years of digital creativity—demo recordings of local bands, first tracks of future stars, an entire era of amateur music—vanished in one stroke due to a botched migration. The company kept answering that "nothing was lost," then sent a user a letter with an admission. This very case was cited by MIT Technology Review in 2024 as precedent: "All content before 2016 simply disappeared." And MySpace isn't the exception but the rule: entire layers of Usenet, the internet's primary conversational culture, will never return online.
Let's step away from stories and ask coldly: how long does digital media actually live? There's a unique source here—Backblaze, a company that's been publishing failure statistics on its hard drive fleet since 2013. The 2025 report features hundreds of thousands of drives—337,192 in the last quarter alone, thirteen years of observations. The annual AFR (annualized failure rate) for 2025—1.36%. Over the entire observation period—about 1.30%. Sounds encouraging: drives have become more reliable, fleets "healthier." But flip the fraction: out of every hundred drives, roughly one dies completely per year. If you store data on a single drive, after ten years the chance of encountering silence is noticeably above zero. This isn't prophecy, it's arithmetic.
Worse: drives can die quietly. CERN, which for all its love of precision values data more than gold, conducted a systematic study of "silent data corruption" in 2007. Specialists wrote a 2-gigabyte file to thousands of nodes every two hours for five weeks—and found hundreds of errors. The resulting error rate turned out orders of magnitude worse than the declared specifications of individual components: ~3×10⁻⁷ errors per byte across the entire chain CPU→RAM→RAID→disk, versus claimed 10⁻¹²–10⁻¹⁴ for individual links.
The reason isn't manufacturing defects but chain length: each hop adds probability, and a specific write operation runs data through several links. The rescue is end-to-end checksums (like in ZFS) that verify data on every read. Sad conclusion: "data lying on disk" isn't a photograph but a film frame that needs continuous checking and rewriting.
Now compare with paper. Quality acid-free paper in a dark, dry cabinet easily lives hundreds of years. Microfiche and ceramics—thousands. Jack Cushman from Harvard Library Innovation Lab formulated this into a phrase you want to engrave on every data center door: "Books left in a dark room for ten years without budget—happy. Miss one AWS payment—files gone forever." The entire difference between old and new storage fits in one verb: paper can be forgotten, digital cannot.
From physics let's move to sociology. Data survives not because it's "on disk" but because there's an institution nearby that copies, checks, and rewrites it to new media. The brightest counter-example—Archive Team and the GeoCities story. In 2009, Yahoo announced the closure of GeoCities—hosting for millions of homemade early-web pages. The company promised to preserve nothing. A ragtag group of volunteers led by Jason Scott launched an emergency campaign and in six months managed to download about a terabyte out of roughly nine terabytes of peak volume. Most went away forever. Scott later formulated it this way: "Archiving is a choice, a duty, not something that happens automatically like the tide."
The positive example is larger—Internet Archive with its Wayback Machine. The library founded by Brewster Kahle in 1996 has accumulated over 145 petabytes of data and more than 95 million media files; Wayback stores over 800 billion web pages and adds 650 million daily. But even here the numbers breathe desperation: the creators admit it's "Sisyphean labor," and priority always goes to material "under threat right now." The task of catching everything is inherently doomed—so libraries negotiate among themselves about who takes which slice.
The scale of losses is well measurable. Pew Research Center in May 2024 calculated: 23% of pages that existed in 2013 are already inaccessible; in news articles from that same year, about a third of links are dead (38%). Link rot isn't anecdote but statistics. This means that after ten years, the web has already lost a fifth of itself. Project this rate over fifty years—and you get a century we'll know less about than the nineteenth century from newspaper archives.
Meanwhile, the question "who preserves" runs into interest. Brewster Kahle says directly: the balance of power has shifted from libraries to corporations, and corporations preserve what's profitable to them—and delete at their discretion. The Olive project from Carnegie Mellon, where Mahadev Satyanarayanan demonstrated "digital vellum" to Vint Cerf—emulating any old software "in one click"—lives on grants, not revenue. Economics' sad truth: the more "eternal" a record seems ("well it's in the cloud, it's not going anywhere"), the less money and people society is willing to allocate to its actual survival.
Since survival requires an institution and institutions don't live long, engineers try to create media that don't need institutions. Three projects deserve their own stanza.
Project Silica (Microsoft Research) writes data with femtosecond lasers into quartz glass: in 2023—7 TB on one glass plate, in 2024 they announced initial commercialization stages jointly with Warner Bros.: they buried "Superman" from 1978 on glass, among other things. Glass doesn't fear water, magnetic fields, temperature, and more importantly, doesn't require electricity: seal it in a safe—and forget for decades. Project lead Richard Black reasonably notes: "Humanity has never stopped building microscopes"—meaning we'll certainly still be able to read glass, it's just a question of will.
GitHub Arctic Code Vault—an attempt to bury code. On February 2, 2020, the company took a snapshot of all active public repositories and recorded 21 TB on 186 reels of piqlFilm archival film. The film is rated for a thousand years and doesn't require a reading computer, just a microscope and light. On July 8, 2020, the reels were laid in the depths of an abandoned coal mine in Svalbard, in the Arctic World Archive—where works from the Vatican, ESA images, and collections from other institutions are also stored. Perfect analogy with cave seed vaults emerged: a deposit in case things go wrong "up there."
Memory of Mankind—the most radical of the three. Austrian architect Martin Kunze's project stores information on ceramic tablets sealed in the world's oldest working salt mine in Hallstatt, Austria. Ceramics is basically a return to Mesopotamia: analog image of text and pictures, readable by eye, without any computer. Declared preservation period—up to a million years; salt itself is a preservative environment, not allowing water and microorganisms. The project is populist: anyone can buy a "token" and enter their page into this eternal archive. A global analog of a tombstone emerges: "something was here, and here it is."
The common denominator of all three projects is disconcerting: the most reliable way to store information long-term is to make it analog and offline. Film, glass, ceramics don't require power, subscriptions, or updates. Digital "eternal" storage lives as long as the wallet being charged for rent lives. As Vint Cerf aptly formulated in 2015, warning of a "digital dark age": even giants can disappear, so "digital vellum"—emulating the entire environment, hardware and software in the cloud—is a bet that someone will keep paying for the server. And they haven't invented a free eternal server, alas.
Now let's return to the hook—the debate about deleted data—and look at it through an archivist's eyes. In retrieval systems, two approaches to deletion are proposed: silently discard or leave a marker "something was here." Engineering intuition screams: why store a pointer to something nonexistent? Archival intuition answers: because "silent disappearance" is indistinguishable from "never existed."
Silent deletion is essentially the future's interface to our century. MySpace didn't mark its 50 million tracks with tombstones—they simply vanished, and in twenty years no one will be able to prove they existed. GeoCities left without an inventory of loss. Apollo tapes were recorded over without a register of "the landing was here." The history of losses always looks like this: not holes in the known, but ignorance about the holes.
In this sense, the debate from the morning feed isn't a minor engineering holy war but a cautious attempt to instill in digital systems a skill that paper civilization has been practicing for centuries: inventory of losses. Libraries have kept "lost" books for hundreds of years; archives distinguish "not preserved" from "never received." Digital culture, where deletion is one button and zero traces, doesn't yet have this skill. Maybe the future standard for large platforms isn't "delete instantly" but "delete with tombstone": a marker "there were N bytes here, deleted on date D." This is expensive, ugly, and possibly exactly what historians thirty years from now expect from us.
First and main: data survival isn't about the medium but about ritual. Paper won not because fiber is better than silicon, but because institutions were built around paper: libraries, chanceries, abbeys, the habit of recopying and checking. Digital data requires the same, only the recopying cycle isn't a generation but years. Forgot to pay for storage—that's it, silence.
Second: the digital dark ages aren't prophecy but already measured history. A fifth of the decade-old web has disappeared. Twelve years of MySpace music erased in one migration. Two census tapes were lost not because of "formats" but because of lack of inventory. The UNIVAC myth was needed to hide the boring truth: data dies from indifference, not progress.
And third, the least obvious: in technological terms, we've possibly already settled on the right answer—we just won't admit to ourselves its cost. Glass, film in Svalbard, and ceramics in a salt mine work. They're cheap, stable, and don't require subscriptions. But all three projects are niche, grant-dependent, and "weird." Society as a whole bets on the opposite: on live clouds that are beautiful, convenient, and dependent on a single electricity bill. The choice between "convenient now" and "readable in 500 years" humanity makes silently, every day, and so far—not in favor of the second. Actually, that's why "archiving is a choice and duty, not the tide."