An archive faces a strange problem.
It wants people to use a record without slowly destroying the best version of that record.
The solution is one of the clearest ways to understand lossy systems: keep a rich master; make derivatives for particular jobs.
Quick Read
Professional preservation workflows distinguish between a high-fidelity preservation master and copies made for production, access, browsing or delivery. The master aims to retain enough information for long-term preservation and future reuse. Access copies can be smaller, faster and easier to distribute because they do not have to carry every property of the source at maximum fidelity.
This is not merely a file-format trick. It is an architecture for uncertainty. We do not know every future question a historian, scientist, student or restoration specialist may ask, so we preserve a richer source than today’s immediate use appears to require.
One-sentence answer: archives manage lossy representations safely by separating preservation from delivery, retaining the richest defensible source while allowing task-specific copies to become smaller.
The Newspaper Problem
Imagine a fragile newspaper from 1942.
A student only wants to read one headline. A historian wants the advertisements. A type designer wants letterforms. A conservator wants paper texture and damage patterns. A future researcher may ask a question nobody has imagined yet.
If we digitise only for the student’s present need, we may create a tiny black-and-white image that makes the headline readable and quietly destroys colour, texture, marginalia and tonal information.
The copy succeeded at today’s task and failed tomorrow’s unknown task.
A Master Is an Option on Future Questions
That is why preservation quality can look excessive.
High resolution. High bit depth. Careful colour management. Stable formats. Checksums. Metadata. Storage redundancy.
Why pay for all that?
Because information discarded today cannot be summoned merely because it becomes important tomorrow.
A rich master preserves optionality.
Access Copies Have a Different Job
The file suitable for preservation may be inconvenient for ordinary access.
It may be enormous. It may require specialist software. It may contain more resolution than a phone screen can display.
So the archive creates derivatives.
- a thumbnail for search results;
- a medium image for a catalogue page;
- a compressed PDF for classroom use;
- a streaming copy for audio or video;
- a production master for routine reuse.
Each derivative answers a receiver.
The mistake is not making a smaller copy. The mistake is forgetting that the smaller copy is a derivative and allowing it to replace the richer source.
Lossiness Is Relative to a Property
A transcript can preserve every spoken word and lose accent, timing and background sound.
A photograph can preserve visible appearance and lose depth, smell, temperature and movement.
A scan can preserve page appearance and lose the physical thickness of the paper.
So “lossless” must always be interpreted relative to what is being encoded. A mathematically lossless PNG is lossless with respect to its stored pixel values; it is not a lossless reconstruction of the physical world that produced them.
Capture Comes Before Compression
This distinction is easy to miss.
Suppose a manuscript is photographed badly: glare hides ink, focus misses the edge, and the camera clips highlights. Saving that photograph in a lossless file format does not restore what the camera failed to capture.
Loss can occur before the file exists.
Preservation therefore begins with acquisition quality, not merely compression settings.
Metadata Preserves What Pixels Cannot
A perfect image with no identity can become useless.
What is it? Who created it? When? From which physical object? Under what capture conditions? Has it been cropped? Colour corrected? Restored? Migrated?
Metadata preserves relationships and provenance that image data alone cannot express.
This gives us another form of loss: a representation can preserve content while losing context.
Checksums Protect Against a Different Kind of Loss
Compression asks whether we intentionally represent something with fewer bits.
Fixity asks whether the bits we decided to keep have changed unexpectedly.
A cryptographic checksum can help an archive detect accidental alteration during storage or transfer. It cannot tell us whether the original scan was adequate, but it can tell us whether the stored file still matches the file previously checked.
Different losses require different guards.
Migration: Preservation Is a Process, Not a Shelf
Digital files do not become immortal merely because they are digital.
Formats age. Storage media fail. Software disappears. Organisations change.
Long-term preservation therefore includes monitoring, replication, validation and sometimes migration to newer representations.
Every migration raises a question: what properties must survive?
The answer is not always “every bit in the old container”. It may include appearance, behaviour, structure, metadata and authenticity relationships. Preservation is preservation of significant properties, not worship of a filename extension.
Why Researchers Ask for the Original
“Can I see the original?” is not academic fussiness.
It is a request to move upstream in the information chain.
A quotation is derived from a paragraph. A paragraph is derived from a document. A scan is derived from an object. A catalogue description is an interpretation of that object.
Moving upstream can restore distinctions that downstream representations discarded.
The Lesson for Students: Keep the Full Problem Somewhere
Students make derivatives constantly.
Highlighting is a derivative. Notes are derivatives. Flashcards are derivatives. A summary is a derivative. A formula sheet is a derivative.
These are useful precisely because they are smaller than the source.
But if the learner keeps only the flashcard and discards the worked example, conditions, counterexample and explanation, later understanding may become brittle.
A good learning system therefore has layers: source, explanation, compressed notes, retrieval prompts and application.
The Lesson for News: A Clip Is Not the Event
A ten-second clip may faithfully show ten seconds and still misrepresent a two-hour event.
The pixels can be authentic while the selection is lossy.
This is why provenance must include not only whether media was altered but how it was selected, cropped and contextualised.
Authenticity and completeness are different dimensions.
The Lesson for AI: Do Not Confuse a Derivative With the World
A model may operate on captions, transcripts, embeddings, resized images, summaries or database fields rather than raw events.
Each representation makes some questions easier and others impossible.
If a system cannot return to richer evidence when the compressed representation is insufficient, it may become confidently trapped inside its derivative.
The archival principle scales beautifully: keep a route back to the best available source.
What Archives Teach Civilisation
Civilisation is full of access copies.
Textbooks compress scholarship. Maps compress territory. Statistics compress populations. Laws compress social expectations. Museum labels compress objects. Search results compress libraries.
We need these reductions. Nobody can carry the full world.
But healthy knowledge systems remember that reductions are reductions.
They preserve provenance. They maintain archives. They permit challenge. They keep higher-resolution evidence available when a consequential decision demands it.
A Practical Five-Layer Model
- Source: the physical or born-digital original.
- Preservation master: the richest managed representation intended for long-term retention.
- Production derivative: a practical working copy.
- Access derivative: a receiver-optimised copy for delivery.
- Interpretation: description, summary, annotation or argument built from the record.
Not every institution uses exactly these labels, and real workflows vary. The value of the model is conceptual: do not collapse different jobs into one file.
Frequently Asked Questions
Why not distribute the preservation master?
Sometimes an institution can. Often the master is unnecessarily large, operationally awkward or restricted. A derivative can provide faster, safer and cheaper access while the master remains protected.
Does lossless compression make a preservation file perfect?
No. It preserves encoded data exactly through compression and decompression, but capture quality, metadata, colour management, cropping and source limitations still matter.
Can a derivative become the only surviving copy?
Yes, and that is why preservation planning matters. Once richer information is gone, a smaller derivative cannot reliably recreate distinctions it never retained.
Preservation Is Really a Decision About Future Ignorance
When an archive digitises an object, it knows the questions being asked today.
It does not know the questions that will be asked in fifty years.
That uncertainty changes the economics of preservation. A richer capture can look inefficient if judged only against current demand, but become extraordinarily cheap if it prevents future re-digitisation, protects a fragile original or preserves a property no one realised would matter.
The master file is therefore not only a copy. It is a hedge against ignorance.
Resolution Is Not the Same as Information
More pixels do not automatically mean more useful information.
If the optics are poor, the object is out of focus or the capture is noisy, a huge file can contain very little additional recoverable detail.
Good preservation therefore considers the whole imaging chain: source condition, lens, sensor, illumination, focus, sampling, colour calibration, tonal response, file encoding and quality control.
Storage size is an outcome. Fidelity is the objective.
Bit Depth Protects Subtle Tonal Differences
A faded manuscript can contain distinctions that are invisible in a coarse tonal representation.
Higher bit depth allows more numerical levels between dark and light. That can matter when later processing needs to separate faint ink from stained paper or recover subtle gradients.
Again the point is not that more is always better without limit. The point is that preservation decisions should be justified by the properties of the source and the plausible future uses of the capture.
Colour Management Is a Promise About Interpretation
A pixel value such as 180, 120, 90 does not carry universal colour by itself.
Colour depends on how those numbers are interpreted within a colour space and on the characteristics of capture and display devices.
Colour management attaches meaning to numerical colour values so that future systems have a better chance of reproducing the intended appearance.
Without that interpretive context, exact numbers can survive while visual meaning drifts.
Cropping Is One of the Most Underestimated Losses
A crop can make an image cleaner and easier to use.
It can also remove binding edges, scale references, handwritten notes, neighbouring objects, page damage or context showing how the object sat within a larger collection.
For access, a tight crop may be excellent.
For preservation, the uncropped capture may contain evidence that nobody wants to lose permanently.
Compression Artefacts Are Not the Only Risk
People often reduce preservation quality to one question: JPEG or TIFF?
That is too small.
- a wrongly identified object can be perfectly encoded;
- a badly focused scan can be mathematically lossless;
- a missing page can have flawless checksums;
- a colour cast can survive every migration exactly;
- a cropped signature can never be recovered from the crop alone.
Integrity has layers. File fidelity is only one.
Born-Digital Records Create a Different Preservation Problem
A paper document has a physical object that can be rescanned.
A born-digital interactive work may depend on software, fonts, databases, linked media, network services and operating-system behaviour.
Preserving the file alone may preserve only the shell.
Digital preservation can therefore require emulation, migration, documentation and capture of relationships among components.
The richer the behaviour, the harder it becomes to define what a future “faithful copy” should reproduce.
Audio and Video Reveal the Same Architecture
A preservation audio master may retain more sample precision and use a stable uncompressed or lossless format.
A podcast listener does not need that master. A compressed access file is lighter, faster and entirely appropriate for ordinary listening.
Video makes the distinction even more dramatic because preservation masters can be extremely large.
The architecture remains the same: preserve richly where future re-use matters; deliver efficiently where access matters.
The Economics of Rich Masters
Preservation has costs.
Large files require storage, replication, monitoring, migration and administration. Institutions cannot preserve everything at infinite fidelity.
So preservation is itself a resource-allocation problem.
The mature question is not “maximum quality at any price”. It is “what level of capture and management protects the significant properties of this material at a sustainable cost?”
That is rate–distortion thinking in institutional clothing, except the distortion measure includes historical, evidential and cultural value rather than only perceptual error.
Redundancy Is Not Waste in Preservation
Compression tries to remove redundancy from a representation.
Preservation deliberately adds redundancy at the storage-system level.
Multiple copies in separate locations protect against hardware failure, disaster and administrative error.
The apparent contradiction disappears once we notice the layers.
Redundant pixels may be unnecessary inside one encoded file. Redundant copies of that file may be essential for survival.
Provenance Is What Lets a Copy Stay Honest
A derivative should know what it came from.
If a catalogue thumbnail has a persistent identifier pointing to the full record, the small image remains useful without pretending to be self-sufficient.
If a transcription points to the scan, and the scan points to the physical object, readers can move upstream when interpretation becomes uncertain.
Provenance turns a stack of copies into a navigable evidence system.
Preservation and Access Should Not Be Enemies
There is a temptation to think that preservation means locking things away while access means opening them.
Digital derivatives allow both goals to strengthen each other.
People can inspect, cite, teach from and search a record without repeatedly handling a fragile original.
Meanwhile the master remains available for new derivatives when technology or research questions change.
Good compression can therefore increase preservation by reducing unnecessary handling and distribution of the richest copy.
A Student Version of the Archive Model
The same architecture can organise study materials.
- Source layer: textbook, teacher explanation, primary reading, full worked solution.
- Master notes: structured notes preserving mechanisms, definitions, examples and exceptions.
- Revision derivative: compressed chapter summary.
- Retrieval derivative: flashcards or question prompts.
- Exam derivative: ultra-compact checklist used immediately before performance.
The error is not making the exam checklist.
The error is deleting everything else and expecting the checklist to regenerate understanding.
A Public-Knowledge Version
Encyclopaedias, museum labels, search snippets and classroom textbooks are all access layers.
They are useful because they are compressed.
Healthy knowledge ecosystems connect those compressed layers back to richer evidence: references, collections, datasets, archives, scholarship and primary sources.
The link upstream is part of the knowledge.
What Can Be Lost Even When the Master Survives?
Sometimes the file survives and the institution around it does not.
Cataloguing knowledge can disappear. Staff expertise can retire. Folder structures can lose meaning. Rights information can become unclear. Relationships among files can break.
Preservation is therefore organisational memory as well as storage engineering.
A file without context can become an archaeological object inside its own archive.
The Deep Rule: Preserve Reversibility Where Reversibility Matters
A derivative can always be made later if the master remains.
A master cannot always be reconstructed later if only the derivative remains.
That asymmetry should guide decisions.
When a transformation is cheap to repeat but impossible to undo, preserve the richer upstream state whenever the stakes justify it.
That is not only archival wisdom.
It is a general law for any civilisation that wants to compress without becoming forgetful.
Sources and Further Reading
- U.S. National Archives and Records Administration, guidance and definitions concerning preservation copies and digital preservation.
- Federal Agencies Digital Guidelines Initiative, technical guidance for digitising cultural heritage materials.
- Library of Congress, Sustainability of Digital Formats resources.
- Digital Preservation Coalition, Digital Preservation Handbook.
- International standards and institutional guidance on metadata, fixity and long-term digital preservation.
Final Thought: Small Copies Are Safe When the Richer World Still Exists
There is nothing wrong with a thumbnail.
There is something wrong with mistaking the thumbnail for everything that could ever be known about the picture.
The wisdom of an archive is not that it refuses compression.
It compresses constantly.
Its wisdom is that it knows which copy is allowed to forget.