A file can survive every storage failure and still become almost useless.
Imagine finding a folder named OLD containing scan001.tif, scan002.tif, and document-final.pdf. All three files open perfectly. Their checksums match. Nothing is corrupt. But nobody remembers who created them, when they were made, what project they belonged to, or whether document-final.pdf was actually the final version.
The bytes survived. The identity did not.
Metadata is the part that tells you what survived
Metadata can be extremely simple: a title, creator, date, location, description, source, or identifier. It can also be technical or archival information recording file format, rights, preservation events, relationships to other objects, and changes made over time.
The National Archives gives surprisingly practical advice for ordinary family collections: when digitizing papers and photographs, add basic metadata answering Who, What, Where, and When. Those four questions become much harder to answer after the person who knew the collection is gone.
Institutional archives formalize the same problem. NARA’s guidance for digital photographic records requires descriptive information such as a unique identifier, caption, photographer, and the “who, what, when, where, why” needed to understand an image later. The Library of Congress-backed PREMIS preservation metadata model formalizes five kinds of entities around preserved material: Intellectual Entities, Objects, Events, Rights, and Agents.
The reason for all that structure is simple: the file alone often cannot tell its own story.
Context can sometimes be reconstructed
Lost metadata is not always permanently lost. Researchers can infer clues from surrounding material.
A photograph may contain EXIF camera data. A PDF may identify an author internally. A file’s directory may reveal the project it came from. Email correspondence can establish when an attachment was circulated. Sibling files may contain matching names or version numbers. Archived webpages may show where a download originally appeared.
But reconstructed metadata has to be treated differently from original metadata. A filesystem modification date may reflect when a file was copied, not when it was created. A filename can be misleading. An author’s name inside a document may identify the subject rather than the creator. Guessing is not preservation simply because the guess is plausible.
An unidentified artifact is a partial survival
This distinction matters when people say that a digital collection has been “saved.” Saving the files is necessary, but it is not the whole job.
A thousand unlabeled images may preserve enormous historical value while forcing future researchers to spend months reconstructing basic facts. A source-code archive without repository history may preserve code but lose authorship and chronology. A folder of government PDFs without titles, dates, or source URLs may be impossible to cite confidently.
Digital extinction can therefore happen in layers. First the original website disappears. Then the files survive somewhere else. Then, years later, the knowledge connecting those files to their source disappears too.
The artifact remains on disk, perfectly healthy and increasingly mysterious.