Information can change form without becoming different information — but only if the right things survive the transformation.
A temperature can be written in Celsius or Fahrenheit. A train route can be shown on a geographic map or a schematic network diagram. A mathematical function can be represented by an equation, graph, table or verbal rule. A story can move from speech to text to film. Each representation changes surface form. The deeper question is what must remain unchanged for the receiver to say, “this is still the same underlying thing.”
Representational invariance is the preservation of selected structure, relationships or decision-relevant properties across a change of representation.
This article sits beneath How Information Works, How Encoding Works and the existing Cognitive Art route What Is Transformation?. Encoding explains how symbols survive machine representation. Transformation explains how thought changes form. Representational invariance asks the narrower cross-domain question: which properties are not allowed to change when the form changes?
1. Every Representation Selects
No representation contains everything about its source.
A road map preserves location and connectivity but not the smell of the street. A photograph preserves one optical projection but not the full three-dimensional scene. A financial statement preserves selected accounting relationships but not every event inside the organisation.
Invariance therefore cannot mean “everything remains the same.” It means the properties needed for the representation’s job remain stable enough to reconstruct the intended relationship.
2. The Receiver Defines the Invariant
A representation can be faithful for one task and unfaithful for another.
A schematic MRT map may preserve station order and interchange structure while distorting physical distance. For route planning, that can be excellent. For estimating walking distance between lines, the same representation may be inadequate.
The invariant is therefore receiver-relative: what must survive depends on what the receiver needs to do next.
3. Unit Conversion Preserves Quantity
Convert 1 metre into 100 centimetres. The symbols and number change. The represented length does not.
The conversion works because a transformation rule preserves the physical quantity while changing the numerical representation.
This is a clean example of invariance: form changes; referent stays fixed.
4. Coordinate Changes Preserve Geometry Selectively
The same point can be represented in different coordinate systems. The coordinates change; the underlying location does not.
More generally, mathematical transformations preserve selected invariants: distance under rigid motion, angle under some transformations, topology under continuous deformation, probability under valid reparameterisation.
The power of mathematics often comes from identifying what remains invariant while everything else changes.
5. Translation Tries to Preserve Meaning, Not Words
A literal translation can preserve words and destroy function.
If a polite request in one language is translated into a grammatically correct but socially blunt sentence in another, lexical correspondence survived while pragmatic meaning did not.
Good translation therefore chooses a deeper invariant: communicative force, reference, tone, relation or intended effect.
See How Translation Layers Work.
6. Compression Preserves Some Information by Losing Other Information
A JPEG image can throw away pixel-level information while preserving enough visual structure that the photograph still looks acceptable.
A summary loses sentences while preserving the argument’s main structure. A transit map loses geographic fidelity while preserving route logic.
Compression works only when the discarded information is less important to the receiver than the preserved invariant.
The broad owner remains How Compression Works.
7. Schema Translation Preserves Fields Imperfectly
When one metadata system is translated into another, fields may not align exactly.
One schema may distinguish creator, editor and contributor. Another may have only one generic “agent” field. Translation is possible, but the finer distinctions can collapse.
The existing Metadata Crosswalks branch owns that applied problem. Representational invariance gives the general principle: define what must survive before translating.
8. Invariance Can Be Tested Through Round Trips
One useful diagnostic is a round-trip transformation.
Convert representation A to B, then reconstruct A. Which properties survive? Which disappear? Which drift?
Machine encoding often uses this test. Translation and archival migration can use analogous checks. A perfectly reversible round trip is not always required, but the losses become visible.
9. Invariance Can Fail Because the Receiver Lacks Context
A representation may preserve the intended structure while the receiver cannot reconstruct it because the required convention is missing.
A mathematical symbol without the notation rules, a map without a legend, a dataset without units, or a photograph without date and location can become ambiguous.
Representation and metadata therefore work together: the content carries state; the metadata helps the receiver interpret the state correctly.
10. Invariance Can Fail Through Hidden Scale Change
An average can preserve a population total poorly. A map at national scale can hide street-level barriers. A monthly statistic can hide daily volatility.
When representation changes resolution, relationships visible at one scale can disappear at another.
The representation is not necessarily wrong. The invariant changed because aggregation altered what distinctions remain recoverable.
11. Worked Example: Algebra, Graph and Table
The function y = 2x + 1 can be shown as an equation, a graph or a table of values.
The equation makes symbolic transformation easy. The graph makes rate and geometry visible. The table makes individual input-output pairs explicit.
The representations differ, but the relationship between x and y should remain invariant. If the graph shows a different slope, the translation failed.
12. Worked Example: MRT Network Map
A schematic network map bends lines and changes distances. That is deliberate distortion.
What must remain invariant is station sequence, line membership, interchange structure and direction logic. The map succeeds because the distortion sacrifices geography to preserve navigation structure.
The MRT owner remains How MRT Works | It’s Mathematics.
13. Worked Example: Photograph and Caption
A photograph preserves visible evidence from one viewpoint. A caption adds names, date, event and interpretation.
Move the image into an archive without the caption and visual structure survives while contextual identity may disappear. Keep the caption with the wrong image and metadata survives while reference fails.
The Photography owner remains How Photography Works.
14. Worked Example: Education
A learner can solve a fraction problem using a diagram, equation or verbal explanation.
If the learner understands the underlying ratio relation, performance should transfer across representations. If the learner memorised only one surface procedure, the invariant does not survive the change.
Representation transfer is therefore a powerful test of real understanding.
15. Worked Example: Legal and Institutional Translation
A law can exist as legislative text, operational regulation, administrative procedure and frontline decision rule.
Each layer transforms representation. The institution succeeds only if the legally relevant invariants survive the handoffs. If the operational procedure changes the right or obligation, representation drift has become governance failure.
16. A Representational-Invariance Checklist
- Define the source representation and target representation.
- State which properties must remain invariant.
- State which properties may change or be discarded.
- Identify the receiver and the task after transformation.
- Check units, scale, context and conventions.
- Test round-trip reconstruction where useful.
- Compare outputs across several representative cases.
- Make irreversible losses explicit.
- Reject transformations that preserve surface appearance while breaking decision-relevant structure.
17. Read the Mechanism Forward, Backward and Sideways
Forward: source representation → chosen invariant → transformation → target representation → receiver reconstruction. Backward: start from a wrong downstream decision and ask which invariant disappeared during representation change. Sideways: compare mathematician, translator, engineer, archivist and teacher. Each makes different compromises about what can change and what must survive.
18. The Civilisation Lesson
Civilisation constantly moves knowledge between forms: speech to text, text to database, map to GPS, law to procedure, observation to statistic, memory to archive.
Every transformation creates a possibility of drift. Mature systems therefore do not ask merely whether the new representation looks reasonable. They specify the invariants that must survive before the transformation begins.
Representational invariance is the discipline of deciding what reality must remain recoverable after the form has changed.
Continue through How Notation Systems Work, How Encoding Works and the master How X Works hub. Next: multimodal representation — how different media preserve different parts of one reality.