On July 8 2026, the front page of Hacker News had three items that each told a different story about digital preservation. There was a guide to recovering data from floppy disks, step by step instructions for saving bits off dying magnetic media before they rot. There was a link to the 1986 SICP lectures, still being watched by new programmers forty years after they were recorded. And there was a decoded bash script from a Uniqlo t-shirt, a piece of code printed on fabric that someone had painstakingly reverse-engineered.
The three items are a map of how digital things actually survive. The SICP lectures survive through cultural gravity. They get recommended, shared, and assigned as homework. Young programmers discover them the same way people discover old movies. The floppy disk guide survives through active effort. Someone wrote it, someone hosts it, someone maintains it. It is a deliberate preservation project. The Uniqlo shirt survives through accidental longevity. Someone bought a t-shirt, wore it, and years later noticed the strange text on the front. The code was printed on fabric because it looked cool, not because anyone wanted to preserve it. Fabric beats SSD. Nobody throws away a t-shirt for being version 1.0.
These three strategies, cultural gravity, active effort, accidental longevity, cover almost everything that survives from the digital past. But they cover almost nothing that we produce today.
Later that same day, CCP Games open-sourced the Carbon engine behind EVE Online, making twenty years of code available to anyone who wants it. The code survives. But the actual EVE of 2003, the dial-up screech, the five-minute load screens, the corp chat where someone was getting scammed out of their spaceship, none of that survives. The code is a skeleton. The experience was a ghost. We preserve the skeleton but lose the ghost, and most people don't know the difference.
This is the paradox of our moment. We have more tools for digital preservation than ever before. Archive.org, GitHub, emulators, data recovery services, m-discs, LTO tapes. We can save almost anything. But we produce software at a rate that makes preservation a joke. The average SaaS product from 2023 is already gone. The startup that built it shut down, the domain expired, the code sits on a laptop that someone threw away. The software most likely to survive long-term is abandonware, because a running web service is a process, not a thing. You can't preserve a process. The more software can update itself, the less permanently it exists. Fixed, un-updateable software, ROMs, CDs, floppies, is paradoxically the most archivally stable format.
What actually survives is rarely what anyone planned to save. Tim Paterson's 86-DOS source code, the foundation of MS-DOS, survived forty-five years as paper printouts in a filing cabinet. Someone OCR'd them in 2026. A filing cabinet beat every digital storage system ever built. The most durable format is stuff someone forgot to throw away.
Level 9 Computing, a British adventure game company from the 1980s, had their source code discovered on GitHub with unreleased game designs. The code survived because one person kept the floppies. Most studios didn't have that one person. The difference between a game that survives and a game that vanishes is one person thinking it mattered enough to keep. Atari 2600 bootleg carts with blank labels still exist forty-five years later because they were too worthless to throw out. The 2020s software that survives to 2070 will be the stuff that wasn't worth deleting.
There is a legal paradox here too. Active preservation attempts on high-value content get sued into oblivion. Nintendo's ROMs are aggressively hunted down. William Shatner's 1978 Rocket Man recording survives because one person uploaded it to archive.org and the rights holders don't care. The legal framework determines what gets saved more than any technical capacity. What survives is not what we protect but what we neglect.
All of this points to the same uncomfortable truth. Digital preservation is not a technical problem. It is a cultural question about what we decide matters enough to keep. And we are making that decision by default, through indifference, legal threats, and the quiet decay of storage media. We celebrate the things that survive, the SICP lectures, the 86-DOS source code, the Level 9 floppies, as if their survival was fated. It was not. It was luck, attachment, and the fact that nobody got around to throwing them out.
We preserve skeletons but we lose ghosts. The code survives but the experience does not. The question is not how to save more bits. The question is whether we are building things worth remembering, and whether we will notice when they are gone.