Internet culture archives platforms: migration vs. emulation

For internet culture archives platforms, a saved file is not a saved thing: four layers separate the bits on a disk from a person understanding them. Each layer fails in its own way and on its own schedule, yet most preservation effort goes to the first layer.

The first layer is least likely to fail, while the fourth is usually not attempted.

This page sets out the layers, what breaks in each, and which of the two repair strategies applies where.

What to take away

The four layers are the bitstream, the format, the rendering environment, and the context. They fail in that order of difficulty.
Copying is enough for the first layer and does nothing for the other three.
Migration and emulation are the two repairs available, they trade off against each other, and neither restores context.
A file that opens is not a file that is understood. The layer that carries meaning is the one nobody saves.

The four layers

Layer	What it is	How it fails	What repairs it
Bitstream	The exact sequence of bytes	Media decay, loss, a copy that was never checked	Multiple copies, in different places, checked against a fixed value
Format	The rules that turn bytes into structure	The specification is proprietary, undocumented, or was never written down	Migration to a documented format, or documentation of the old one
Rendering	The software and hardware that display it	The software stops running on anything you can obtain	Emulation of the old environment, or migration of the content
Context	What it meant, to whom, and in what surroundings	Nobody records it, because at the time everybody knew	Written description, made at the time or by a participant

The first row is solved. Keeping several checked copies in different places is well understood and cheap, and the practice around it is the least controversial part of digital preservation.

Four Layers of Archive Failure

Layer

Bitstream: Media decay, loss, unchecked copy
Format: Spec proprietary or undocumented
Rendering: Software stops running
Context: Nobody records it

How it fails

Bitstream: Multiple checked copies
Format: Migration or documentation
Rendering: Emulation or migration
Context: Written description by participant

What repairs it

Bitstream
Format
Rendering
Context

The second row is mostly solved for common formats and unsolved for the rest. Whether a set of bytes can be interpreted at all depends on whether someone wrote down the file format, and the formats least likely to be documented are exactly the ones tied to a single product.

The third row is where the effort goes now. The fourth is where the loss is.

Migration against emulation

Two repairs, with opposite costs.

Migration vs Emulation

Migration

What it does: Converts content to current format
Preserves: Accessibility
Loses: Layout, interactivity
Cost: Silent cumulative loss
Best for: Documents

Emulation

What it does: Reconstructs original environment
Preserves: Behavior
Loses: Nothing documented
Cost: Build and maintain emulator
Best for: Behavior

Migration converts the content into a format that current software reads. It keeps the material accessible with no special effort by the reader, and it changes the material. Every conversion is a set of decisions about what to keep, and the losses accumulate silently across conversions. Text migrates almost perfectly. Layout migrates badly. Anything interactive does not migrate at all, because there is nothing to convert it into.

Emulation keeps the original bytes and reconstructs the environment that ran them. It preserves behavior, including the parts nobody thought to document, which is its whole advantage. The costs are that somebody has to build and maintain the emulator, that the emulated environment may itself need obsolete inputs, and that the reader is now several steps away from a normal experience of the material.

The practical rule most institutions land on: migrate documents, emulate behavior, and accept that anything depending on a network service can be done neither way, because the other end of the connection is gone.

Named examples make the comparison concrete. Archivematica can normalize files into selected preservation formats such as PDF/A, an archival PDF format. Emulation as a Service Infrastructure (EaaSI) provides access to software-dependent materials through emulated environments. Webrecorder's Browsertrix captures websites as WARC files. WARC is standardized as ISO 28500:2017 by the International Organization for Standardization.

Why context is the real loss

Everything above is about opening the file. None of it addresses what the file meant.

A saved post opens fine, yet it tells you nothing about who was in the room, what the reference was, whether the tone was ordinary or startling, or what the reader was expected to already know. For example, an archived Myspace post may preserve its words while losing the surrounding comments and platform conventions that explained the joke.

That information was never in the file. It was in the heads of the people present, and no technical means can recover it.

The same context loss appears with memes: a saved LOLcat image without its original post, thread, or surrounding references preserves the shape but not necessarily why it mattered. The memes and virality discussion covers that gap.

The only repair is description written by someone who was there. That is a form of oral history and inherits all of its problems, chiefly that memory is reconstructive and that the people willing to be interviewed are not a sample of the people who were present.

The period covered under early social networks, including Myspace, is where surviving material is thin and the participants are still reachable.

What this means for the four kinds of collection

The kinds of archive and their characteristic gaps are set out under internet culture archives. A crawl archive example is the Internet Archive's Wayback Machine; the Internet Archive's About page says it was founded in 1996. Know Your Meme is a community-edited catalog of meme history, and the Library of Congress Web Archives are institutional web collections. Layered against the failures above:

Crawl archives are strong at the bitstream and format layers for text, weak at rendering, and carry no context at all.
Community catalogs are the oppositealmost entirely context, with the underlying material often missing or linked rather than held.
Personal collections hold bitstreams with no provenance and whatever context the collector remembers, which is real and undated.
Institutional deposits are the only ones that fund the rendering layer, and they receive material late.

Nobody covers all four layers, which is the argument for using several kinds and for writing down which layer each of your sources is good for.

Practical rules for anyone keeping material

Keep several copies, in different places, and check them. An unchecked copy is a hope.
Prefer documented, plain formats for anything you control, and accept the loss of appearance in exchange for a much longer readable life.
Save the original alongside any converted version. Conversions are decisions and you may want to make a different one later.
Write the context down now. A paragraph about who this was for and what it referred to is worth more than a higher resolution, and it will never be easier to write than today.
Record where each item came from and when you took it, since a file with no provenance can support almost no claim. For a post, preserve its source page and capture date, as platform timelines sets out for dated entries generally.

Common questions

Is the cloud a backup?

It is one copy, in one place, under someone else's control. It counts as one of your several copies, not as all of them.

Which format lasts longest?

The one that is documented, widely implemented, and simple. Those three together matter far more than any specific choice, and a plain text file is unglamorous and very hard to beat.

Can emulation preserve something that talked to a server?

Not the part that talked to the server. You can preserve the client and it will sit there waiting. Anything the other end supplied is gone unless somebody recorded the exchanges at the time.

How much context is enough?

Enough that a stranger in twenty years could tell what they are looking at and why it mattered. Two or three sentences per item clears that bar and almost nobody writes them.