How Studios Lose Their Own Work
Link rot measured, content drift measured, and the formats most likely to strand you. An archive structure and naming convention built on standards rather than habit.
Date
Category
Writer
The first time it happened to me it was a folder of 2,400 layered files from a single year of work, sitting on a drive that mounted perfectly. Every file opened. Every file was missing its linked images and its fonts, which had lived somewhere else on a machine I no longer owned. Nothing was corrupt. Nothing was deleted. The work was simply no longer there in any useful sense, and I had been telling myself for six years that it was backed up.
Studios lose their own work constantly, and almost never in the way they plan against. The threat model everyone has is a failed disk or a fire. The actual threat model is slower, more boring, and much better documented than most designers realise, because librarians and lawyers have been measuring it for two decades.
The decay, measured
In 2014 Jonathan Zittrain, Kendra Albert and Lawrence Lessig published a study in the Harvard Law Review Forum of what had happened to hyperlinks in legal citations. They examined 555 hyperlinks in United States Supreme Court opinions, the earliest from 1996, and links in three Harvard journals over 1996 to 2012. Their summary: "more than 70% of the URLs within the above mentioned journals, and 50% of the URLs within U.S. Supreme Court opinions suffer reference rot." Precisely, 49.9 percent of the Supreme Court links no longer contained the cited material. In the Harvard Law Review only 26.8 percent of links still contained what was cited; in the Journal of Law and Technology 34.2 percent; in the Human Rights Journal 29.9 percent.
One number in that study deserves more attention than it gets. Of the Supreme Court links, 63.6 percent returned an HTTP 200 status code, and only 78.5 percent of those actually contained the cited material. A working link is not evidence of anything. Any automated link checker a studio runs against its own site will report those pages as healthy. This is the correction I would most like designers to absorb: the tooling that tells you your archive is fine is only checking whether a server answers.
Pew Research Center measured the wider web in a report published on 17 May 2024. Sampling roughly a million pages, about 90,000 from each year between 2013 and 2023, it found that 38 percent of webpages that existed in 2013 were no longer accessible a decade later, and that a quarter of all pages from the whole 2013 to 2023 range were inaccessible as of October 2023. Of 500,000 government pages checked, 21 percent contained at least one broken link. Of 500,000 news pages across 2,063 domains, 23 percent did. Of 50,000 English-language Wikipedia pages, 54 percent had at least one broken reference link. Of nearly five million tweets collected between 8 March and 27 April 2023 and watched until 15 June, nearly one in five were no longer publicly visible.
The third study is the one that frightens me most, because it describes a failure that leaves no trace. Jones, Van de Sompel and colleagues, writing in PLOS ONE in 2016, looked at 1,059,742 references to ordinary web resources in a corpus of 3.5 million science, technology and medicine articles published between 1997 and 2012. Using 19 web archives they found representative snapshots for about 30 percent of those references, and then compared each snapshot with the live page. Their finding: "for over 75% of references the content has drifted away from what it was when referenced." The precise figure was 76.35 percent, 184,065 of 241,091 references checked.
Library stacks. Institutions solved the durable-archive problem with boring, unglamorous discipline. Photograph by Chevsapher, CC0, via Wikimedia Commons.
Three failures, one word
The literature separates what studios lump together, and the distinctions are practically useful.
Link rot, in Zittrain's definition, is "The URL no longer serving up any content at all." This is the visible failure. You notice it.
Reference rot occurs "when a link still works but the information referenced by the citation is no longer present, or has changed." The page loads. What you cited is gone.
Content drift, in the PLOS ONE authors' words, is where "the resource identified by a URI changes over time. The resource's content evolves and can change to such an extent that it ceases to be representative of the content that was originally referenced." Nothing breaks. The meaning slides out from under the pointer.
For a studio, the equivalents are exact. Link rot is the client site that has gone. Reference rot is the client site that is still there, replatformed, with none of your design on it. Content drift is the case study page that is still live, still yours by attribution, and now shows a version of the work three redesigns removed from what you made.
Formats, and the thing that actually kills them
Adobe stopped supporting Flash Player on 31 December 2020 and began blocking Flash content from running on 12 January 2021, advising that "all users immediately uninstall Flash Player". Every piece of interactive work made in that format for roughly fifteen years became unplayable by ordinary means on a fixed date, announced years in advance, and most of it was lost anyway.
Flash is the dramatic case and therefore misleading. Formats rarely die by announcement. They die when the only software that reads them properly stops shipping, or when a version bump silently changes how a file renders. A layered file from 2009 that opens in current software and rasterises an effect differently has not failed; it has drifted, exactly like a web page.
National archives handle this by ranking formats rather than trusting them. The Library of Congress Recommended Formats Statement, currently at version 2025-2026, exists to identify "analog and digital formats suitable for the large-scale acquisition of and long-term access to library collections". For digital photographs it lists, in order of preference, TIFF, JPEG2000, PNG, then JPEG/JFIF. For textual works it prefers XML-based markup with an accessible schema and stylesheets, then EPUB3, then PDF/UA (ISO 14289-1) and PDF/A (ISO 19005), then ordinary PDF at the highest quality available with searchable text and embedded fonts.
The US National Archives publishes something similar as a hard rule for transfer: preferred TIFF versions 4, 5 and 6 for born-digital photographs, with JFIF, DNG, PNG and JP2 merely acceptable; preferred ASCII, Unicode, ODT and PDF/A-1 and PDF/A-2 for textual records, with ordinary PDF, DOCX and DOC acceptable. Its Digital Preservation Framework goes further, with preservation plans for sixteen record categories that identify "Significant Properties, the properties that should, if possible, be retained in any format migration", and preservation action plans for over 700 file formats.
That phrase, significant properties, is the one worth stealing. The question is never "will this file open". It is "which properties of this file matter, and does the format I am migrating to preserve them".
Format | Standing with national archives | Main archival risk | What I actually do |
|---|---|---|---|
TIFF | Preferred by both LoC and NARA for still images | Size; variant flavours and compression schemes | Flattened TIFF as the archival copy of every finished raster piece |
PNG | Preferred by LoC (third), acceptable to NARA | No CMYK, limited colour management | Screen deliverables only, never the master |
JPEG | Preferred by LoC (fourth), acceptable to NARA | Lossy, and generation loss compounds silently | Distribution copies only; never re-saved over itself |
PDF/A (ISO 19005) | Preferred by LoC and NARA for text | Requires embedded fonts and no external dependencies | Every document and every approved layout, exported at sign-off |
Ordinary PDF | Acceptable, not preferred | May reference external fonts; features vary by version | Fine for sending, not counted as an archive copy |
Native layered files (layout, raster, vector) | Not addressed as preservation formats | Vendor-controlled, version-dependent, dependent on linked assets and fonts | Kept, but always alongside a flat archival export and a packaged assets folder |
Browser-based design tool documents | Not addressed; not a file in any local sense | Exists only in a vendor account; export fidelity is partial | Exported to a static format at every milestone, on a schedule, not on a whim |
Flash SWF | Obsolete by vendor action | No supported runtime | Kept as an artefact with a screen recording beside it, because the file alone is now unreadable |
The migration
On 18 March 2019 MySpace confirmed the loss of more than twelve years of uploaded media. The statement it gave users was three sentences long: "As a result of a server migration project, any photos, videos, and audio files you uploaded more than three years ago may no longer be available on or from MySpace. We apologize for the inconvenience." Reporting at the time put the scale at over 50 million tracks from 14 million artists.
No malice, no hack, no disaster. A migration. This is the most common way institutional archives die, and it is the one a studio is most exposed to, because a studio's archive is now mostly other people's servers: a design tool, a file sync service, a project management app holding the only record of why a decision was made, a font subscription that governs whether a file will render, and a client's content management system holding the only live copy of the work.
Every one of those is a migration risk you do not control and will not be consulted about. The lesson is not to avoid cloud services. It is that anything which exists only inside a vendor's account is not archived, it is merely stored, and the two words have been used interchangeably for so long that most studios cannot tell you which one describes their own work.
What I got wrong
Three things, specifically.
I treated backup as archiving. A synced folder propagates deletions and corruption at the speed of the network. It protects against a dead disk and nothing else. An archive needs at least one copy that is not connected to anything and does not change.
I trusted that a file opening meant a file was intact. It does not. A layered document with missing links opens. A document with substituted fonts opens. Nothing warns you at the moment of loss, only years later when you need the file.
And I kept everything, which is the same as keeping nothing. An undifferentiated archive of every version of every file is unsearchable, and an unsearchable archive is functionally lost. Selection is part of preservation, not a compromise of it.
A standard that survives
The two pieces of published infrastructure worth adopting are both free and both boring.
The first is BagIt, specified in RFC 8493 (Kunze, Littman, Madden, Scancella and Adams, October 2018). A bag is "a directory containing the payload files and other accompanying metadata files known as 'tag' files". The payload lives in a directory called data. A manifest file lists every payload file with its cryptographic checksum, so any silent corruption is detectable. The specification notes that BagIt "differs from serialized archival formats such as MIME, TAR, or ZIP" in that the files stay directly accessible through the operating system, which matters a great deal in twenty years.
The second is Dublin Core, published as DCMI Metadata Terms. Fifteen elements: Title, Creator, Contributor, Publisher, Date, Description, Subject, Type, Format, Identifier, Source, Language, Relation, Coverage, Rights. It is deliberately minimal, which is why it survives. The Library of Congress asks for essentially the same set as essential metadata on still images: "Title, Creator, Creation Date, Place of publication, Publisher/producer/distributor, Contact information", with embedded IPTC data as desirable.
Here is the convention we run. It is not clever, which is the point.
Step seven is the one everybody skips and the only one that detects silent failure. A checksum you never verify is a comment, not a control.
What is worth keeping
Not everything. The working files from a project are interesting for about eighteen months and then they are weight. What holds value over a decade is narrower than a designer's instinct suggests: the final artwork in an archival format, the source material you or someone you paid actually produced, the reasoning in plain text, the licences, and a dated capture of the thing as it went into the world.
That last item is the one studios almost never have, and the research above says why it matters. If 38 percent of pages from a decade ago are gone, and three quarters of the surviving ones have drifted from what they were, then a studio that points at live URLs as its portfolio is pointing at something with a measured half-life. The work you did is not on the client's server. It is on the client's current server, which is a different thing, and it will stop being true without anyone telling you.
Sources
Jonathan Zittrain, Kendra Albert and Lawrence Lessig, "Perma: Scoping and Addressing the Problem of Link and Reference Rot in Legal Citations", Harvard Law Review Forum, volume 127, 2014. 555 Supreme Court hyperlinks, 49.9 percent no longer containing cited material, 63.6 percent returning HTTP 200 of which 78.5 percent contained the material; Harvard Law Review 26.8 percent, Journal of Law and Technology 34.2 percent, Human Rights Journal 29.9 percent; definitions of link rot and reference rot. https://harvardlawreview.org/wp-content/uploads/2014/03/forvol127_zittrain.pdf
Pew Research Center, "When Online Content Disappears", 17 May 2024. 38 percent of 2013 pages inaccessible; a quarter of all 2013 to 2023 pages inaccessible as of October 2023; 21 percent of government pages and 23 percent of news pages with at least one broken link; 54 percent of Wikipedia pages with a broken reference; nearly one in five tweets no longer publicly visible; sample sizes and collection dates. https://www.pewresearch.org/data-labs/2024/05/17/when-online-content-disappears/
Shawn M. Jones, Herbert Van de Sompel, Harihar Shankar, Martin Klein, Richard Tobin and Claire Grover, "Scholarly Context Adrift: Three out of Four URI References Lead to Changed Content", PLOS ONE, 2016, doi 10.1371/journal.pone.0167475. 1,059,742 URI references from 3.5 million STM articles 1997 to 2012; 19 web archives; representative snapshots for about 30 percent; 76.35 percent content drift; the definition of content drift. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0167475
Adobe, "Adobe Flash Player End of Life". End of support 31 December 2020; blocking of Flash content from 12 January 2021; the recommendation to uninstall. https://www.adobe.com/products/flashplayer/end-of-life.html
Library of Congress, Recommended Formats Statement, version 2025-2026, table of contents and still image and textual works sections. Purpose statement, preferred and acceptable format hierarchies, essential and desirable metadata. https://www.loc.gov/preservation/resources/rfs/TOC.html
Library of Congress, Recommended Formats Statement, still image works. TIFF, JPEG2000, PNG, JPEG/JFIF in order of preference; essential metadata list; IPTC as desirable. https://www.loc.gov/preservation/resources/rfs/stillimg.html
Library of Congress, Recommended Formats Statement, textual works. XML-based markup, EPUB3, PDF/UA (ISO 14289-1), PDF/A (ISO 19005), then ordinary PDF; acceptable formats including XHTML, DOCX, ODF and RTF. https://www.loc.gov/preservation/resources/rfs/text.html
US National Archives and Records Administration, "Appendix A: Tables of File Formats", records management transfer guidance. Preferred and acceptable formats for born-digital photographs, audio and textual records. https://www.archives.gov/records-mgmt/policy/transfer-guidance-tables.html
US National Archives and Records Administration, "Digital Preservation Framework for Risk Assessment and Preservation Planning". The risk matrix, preservation plans for 16 record categories, the definition of Significant Properties, and preservation action plans for over 700 formats. https://www.archives.gov/preservation/digital-preservation/risk
NPR, "Myspace Says It Lost Years Of User-Uploaded Music", 18 March 2019. The verbatim MySpace statement about the server migration project, and the reported scale of more than 50 million tracks from 14 million artists. https://www.npr.org/2019/03/18/704458168/myspace-says-it-lost-years-of-user-uploaded-music
J. Kunze, J. Littman, E. Madden, J. Scancella and C. Adams, "The BagIt File Packaging Format (V1.0)", RFC 8493, October 2018. Definition of a bag, the data payload directory, manifests with checksums, bag-info.txt, and the distinction from serialised formats such as MIME, TAR and ZIP. https://www.rfc-editor.org/rfc/rfc8493.html
Dublin Core Metadata Initiative, DCMI Metadata Terms, recommendation of 2020-01-20. The fifteen elements and their definitions. https://www.dublincore.org/specifications/dublin-core/dcmi-terms/
Related reading
Latest Articles.
© Moc Mallen Studio
Thoughts, ideas, and perspectives on design, simplicity, and creative process.
