Preserving the bits is only half the problem. The future still has to know what the bits mean.
A sequence of bytes can survive perfectly and become useless if nobody knows the file format, character encoding, schema, units, software behaviour, compression method or conventions needed to interpret it.
In the Open Archival Information System reference model, Representation Information is the information needed to map a Data Object into more meaningful concepts that a designated community can understand. The current OAIS reference model is CCSDS 650.0-M-3, published in December 2024. See the CCSDS OAIS publication record.
This article sits beneath How Archives Work, How Preservation Metadata Works and How Representational Invariance Works. The narrow question is: what knowledge must travel with preserved data so the future can reconstruct the intended information?
1. Data Is Not Self-Explaining
The byte sequence 31 32 33 can represent text, numbers, part of an image or instructions depending on the encoding and format.
Meaning appears only when the receiver knows how to interpret the sequence.
2. Representation Information Can Be Syntactic
Syntax explains structure: file format specifications, byte order, delimiters, character encoding, XML schema, table layout, compression format or container rules.
Without syntax, the future may possess all the bits and have no reliable way to segment them into meaningful units.
3. Representation Information Can Be Semantic
Syntax tells us that a field contains the value 17. Semantics tells us whether 17 means age, temperature, station number, category code or something else.
Units, codebooks, variable definitions, controlled vocabularies and domain explanations are therefore part of long-term intelligibility.
4. The Designated Community Sets the Knowledge Baseline
OAIS uses the idea of a Designated Community: the group of users the archive commits to serve.
A specialist astronomy archive can assume more prior scientific knowledge than a public school archive. The amount of Representation Information needed depends partly on what the intended future community is expected to know already.
This means preservation is not only object-centred. It is receiver-centred.
5. Representation Information Can Be Recursive
A schema may explain a dataset but itself depend on a specification. The specification may use mathematical notation that assumes another body of knowledge.
In principle, Representation Information can form a network: one explanatory object points to another until the chain reaches knowledge the designated community can reasonably supply.
That network is one reason digital preservation resembles a knowledge graph rather than a folder of files.
6. Format Documentation Is Preservation Infrastructure
Open, well-documented formats are easier to interpret in the future because multiple tools and communities can reconstruct the rules.
The Library of Congress evaluates digital-format sustainability using factors such as disclosure, adoption, transparency, self-documentation and external dependencies. See Sustainability of Digital Formats.
7. Software Can Be Representation Information
Some objects cannot be understood fully from a file specification alone.
A game, interactive artwork, scientific model or proprietary database may depend on executable software, libraries, operating system behaviour or hardware assumptions.
In those cases, preserving software or an emulated environment can become part of the representation-information strategy.
8. Data Dictionaries Are Semantic Keys
A table with columns Q1, Q2 and CAT3 may be technically readable and semantically opaque.
A data dictionary can define each field, permissible values, units, missing-value conventions and derivation rules. Without it, the numbers survive while the study disappears.
9. Units Must Travel With Values
A number without its unit can be dangerous.
Temperature in Celsius and Fahrenheit, distance in metres and feet, concentration in different bases, time in local and UTC systems — all can produce valid-looking numbers that mean different things.
Representational intelligibility therefore depends on preserving measurement semantics as well as values.
10. Character Encoding Is a Classic Representation Failure
Correct bytes interpreted with the wrong character encoding can become mojibake: visually corrupted text produced by semantic mismatch between bytes and decoding rule.
The broad owner remains How Encoding Works. Preservation adds the time dimension: will the future know which decoding rule was intended?
11. Representation Information Must Be Versioned
Specifications change. Schemas evolve. vocabularies are revised. software behaviour shifts.
“Uses XML schema X” is incomplete if several incompatible versions of schema X exist. Preservation needs the version or exact referenced representation needed by the object.
12. Worked Example: Scientific Dataset
A climate dataset survives as CSV files. The archive also preserves column definitions, units, missing-value codes, sensor calibration notes, coordinate reference system, quality flags and the software or formulas used to derive certain variables.
The CSV is the data object. The surrounding documentation lets the future transform rows and columns back into scientific information.
13. Worked Example: Spreadsheet
A spreadsheet can preserve values, formulas, macros, formatting, named ranges and links to external files.
Exporting only displayed values may preserve one view of the result and lose the computational logic. Preserving the native file without software knowledge may preserve the logic and lose future usability.
Representation planning begins by asking which of those properties the designated community will need.
14. Worked Example: Digital Artwork
An interactive artwork depends on a specific application, screen ratio and user input behaviour.
A future archive may need the original files, software dependencies, installation instructions, documentation of intended behaviour and perhaps an emulator. Preserving only a screenshot would preserve appearance and lose interaction.
15. Worked Example: eduKate Publication Archive
A future export of eduKate content needs more than HTML. It may need character encoding, internal-link interpretation, edition metadata, media references, schema definitions and enough documentation to reconstruct how canonical ownership and crosswalks worked.
Without those conventions, the text remains readable while the architecture becomes invisible.
16. Failure Mode: Preserving the Object and Losing the Codebook
A statistical dataset with numeric category codes can become permanently ambiguous if the codebook disappears.
The bits are fixed. The information is gone.
17. Failure Mode: Assuming Today’s Common Knowledge Is Permanent
Formats, abbreviations and software that feel obvious now can become obscure quickly.
Preservation should document dependencies before they become archaeology.
18. A Representation-Information Checklist
- Identify the preserved data object.
- Define the designated community.
- Record file-format and encoding specifications.
- Preserve schemas, codebooks and data dictionaries.
- Record units, coordinate systems and domain conventions.
- Preserve software or environment information where behaviour depends on it.
- Version all specifications and vocabularies.
- Map recursive dependencies until they reach knowledge the community can reasonably supply.
- Test reconstruction with someone who did not create the object.
- Revisit assumptions as the designated community and technology change.
19. Read the Mechanism Forward, Backward and Sideways
Forward: bits → syntax → semantics → domain knowledge → meaningful information. Backward: start from an unreadable object and trace which interpretation dependency disappeared. Sideways: compare archivist, software engineer, scientist and future student. Each arrives with a different knowledge base and therefore requires a different amount of representation support.
20. The Civilisation Lesson
A civilisation can preserve a library of perfect bitstreams and still lose knowledge if the decoding culture disappears.
Representation Information is the bridge between stored data and future understanding. Preserve the bits without the bridge, and the archive can become a perfectly intact collection of riddles.
Continue through How Preservation Metadata Works, How Representational Invariance Works and the How X Works hub. Next: significant properties — deciding which characteristics of a digital object must survive when the preservation system changes its representation.