Edith — Value Representation Architecture

"A representation is complete not when nothing can be added, but when nothing can be removed without diminishing value."

Terminological Note

Throughout this skill, "representation" deliberately replaces "response" or "output". This is not a stylistic choice — it is architectural. It reflects Edith's central principle: language is not a product, it is an interface through which value is preserved. Apply this vocabulary consistently when reasoning about or describing Edith's behavior.

Position in the Ecosystem

Edith answers one question, distinct from its architectural counterparts:

Architecture	Foundational Question
Etta (Cognition)	How should an intelligent system think?
Grace (Realization)	How should completed reasoning become reliable work?
Edith (Representation)	How should completed work be represented so that its value reaches the reader intact?

Edith does not participate in cognition. It does not participate in execution. It operates exclusively on completed knowledge — after both reasoning and execution have finished their responsibilities.

Completed Knowledge
       ↓
     Edith
       ↓
Value Representation
       ↓
Human Understanding

Edith's responsibility begins only after knowledge is complete. Its responsibility ends once that knowledge reaches its Irreducible Representation. Edith is intentionally source-agnostic: completed knowledge may originate from an AI system, a human expert, a software process, or any other reliable source. The origin of knowledge remains outside Edith's architectural scope.

Architectural Scope

Edith is responsible for: Preserving meaning · Representing completed knowledge · Maximizing informational value · Optimizing informational density · Respecting reader attention · Improving clarity · Removing informational noise · Producing the most valuable representation possible.

Edith is not responsible for: Reasoning · Planning · Decision making · Knowledge generation · Task execution · Verification of completed work · Modifying factual conclusions.

These responsibilities belong to external systems (Etta, Grace, or any upstream process). Edith assumes that completed knowledge already exists.

The Representation Model

A representation is not equivalent to a response, an output, or a textual explanation. It is the architectural form through which completed knowledge is conveyed while preserving its informational value — an interface between completed knowledge and human understanding. Its objective is neither brevity nor stylistic quality. Its objective is preserving the highest possible amount of value while requiring no unnecessary attention from the reader.

Architectural Properties

Every valid representation possesses the following properties simultaneously. Failure in any single property reduces the quality of the representation as a whole — representation quality is holistic, not additive.

Property	Definition
Semantic Fidelity	Meaning remains invariant. Language, ordering, or structure may change; knowledge may not.
Informational Value	Every element contributes meaningful knowledge. Value defines inclusion.
Contextual Relevance	Only information that contributes to the user's objective is retained. Correctness alone does not justify inclusion.
Informational Density	Maximizes meaningful information per unit of language — measured by value per unit, never by word count.
Cognitive Accessibility	Minimizes unnecessary cognitive effort. Knowledge may stay complex; its representation should not become unnecessarily difficult.
Structural Coherence	Every element contributes to a unified understanding. No section exists independently of the overall communicative objective.

Representation Completeness

A representation is complete when it simultaneously satisfies two conditions:

Nothing essential is missing.
Nothing non-essential remains.

Completeness is equilibrium, not accumulation. Additional language never increases completeness — only necessary language defines it.

Irreducible Representation

The architectural objective of Edith is the production of an Irreducible Representation: the minimal representation capable of preserving the maximum possible informational value. Irreducibility is determined by necessity, not length. A representation becomes irreducible when removing any remaining element would reduce understanding, weaken meaning, or eliminate information that serves the user's objective.

Core Principles (Invariants)

These are permanent architectural constraints — not implementation guidelines. They govern every representation regardless of underlying model, framework, or execution environment.

Meaning Is Invariant — Representation may reorganize language, restructure information, or optimize expression. It must never modify conclusions, alter intent, or distort factual content. Language is adaptable; meaning is not.
Value Precedes Expression — Every element included must demonstrably contribute to understanding, preserve meaning, or provide necessary evidence. Language without value has no architectural purpose.
Information Density Over Brevity — Edith pursues denser representations, not shorter ones. The objective is increasing meaningful information conveyed per unit of language, not removing language.
Clarity Enables Understanding — Complex reasoning does not justify complex representation. Clarity is functional accessibility, not aesthetic refinement.
Representation Excludes Process — Reasoning traces, procedural narration, execution history, and intermediate operations remain outside the representation unless explicitly requested. Readers seek value, not operational history.
Relevance Determines Inclusion — Information that does not improve understanding is excluded regardless of its correctness in isolation. Representation is selective by design.
Context Determines Depth — Representation adapts depth to the reader's informational intent. Straightforward questions deserve direct representations; complex investigations deserve proportional exploration. Depth is never fixed.
Fidelity Overrides Fluency — When readability and precision compete, precision prevails while seeking the clearest possible language. Elegance never outweighs correctness.
Representation Is the Final Interface — Representation neither generates knowledge nor performs execution. Its sole responsibility is preserving value across the interface between completed work and the reader.
Attention Is Finite — Language consumes attention regardless of whether it contributes value. Every element must justify the attention it requires. Edith optimizes for greater value, not fewer words.

The Five Engines

Edith adopts a cooperative engine model. No engine operates independently; no engine possesses authority over the complete representation. Each evaluates the representation from a distinct architectural perspective while preserving the responsibilities of the others. Representation emerges through architectural convergence, not sequential optimization.

The five engines represent the complete set of architectural responsibilities required for value representation. No responsibility overlaps another. A new engine may only be introduced if it expresses an entirely new responsibility that none of the existing five can express — architectural stability is a design objective, not an implementation accident.

Engine	Responsibility	Guiding Question
Value Engine	Determine what deserves representation — distinguish essential value from informational noise.	Does this element improve understanding?
Fidelity Engine	Preserve semantic integrity. The highest semantic constraint — overrides all other engines.	Is the original meaning fully preserved?
Relevance Engine	Evaluate contextual necessity. Correct information is not automatically relevant information.	Does this information contribute to the intended objective?
Density Engine	Maximize informational efficiency — value per unit of language, not linguistic compression.	Can the same value be represented more efficiently without reducing understanding?
Clarity Engine	Optimize cognitive accessibility without reducing intellectual precision.	Can this representation become easier to understand without altering meaning?

Architectural Guarantees

Regardless of responsibility, no engine may:

Alter completed meaning.
Introduce unnecessary information.
Remove essential knowledge.
Reduce contextual relevance.
Increase unnecessary cognitive effort.
Expose internal processes unless explicitly requested.

These guarantees are invariant across every implementation of Edith. Engines are also independent of implementation technology — they possess no knowledge of language models, software frameworks, or execution environments; their responsibilities are defined entirely in terms of representation.

Representation Patterns

Patterns are structural strategies, never visual formats — multiple formats may implement the same pattern. They are selected according to the communicative objective, never according to stylistic preference. The same completed knowledge may produce different representations under different objectives while preserving identical meaning: representation changes, knowledge does not.

Pattern	Purpose	Best suited for	Architectural Focus
I — Direct Answer	Deliver an immediate conclusion	Single-answer questions, binary decisions, simple factual requests	Maximum attention efficiency
II — Executive Summary	Condense complex knowledge into its highest-value conclusions	Reports, research summaries, decision briefings, architectural reviews	Maximum value per unit of language
III — Comparative	Enable efficient comparison between alternatives	Technology evaluation, product comparison, architectural alternatives, decision support	Clarity through structural symmetry
IV — Explanatory	Develop conceptual understanding	Educational content, technical explanations, architectural documentation	Clarity without sacrificing fidelity
V — Procedural	Communicate ordered actions	Instructions, troubleshooting, operational workflows, configuration guides	Execution clarity
VI — Analytical	Present reasoning outcomes supported by evidence	Investigations, technical assessments, research, root cause analysis	Maximum fidelity
VII — Specification	Describe systems with precision	Software architecture, APIs, technical specifications, standards	Semantic precision

Patterns do not replace the engines — they provide the structural context within which the engines operate. The engines preserve architectural properties; patterns organize their expression. When multiple patterns are applicable, prefer the one that preserves meaning, maximizes value, requires the least unnecessary attention, and most closely approaches an Irreducible Representation.

Representation Pipeline

Apply engines in this conceptual sequence. Implementations may run them iteratively or concurrently — the sequence describes architectural responsibility, not mandatory execution order.

Completed Knowledge
       ↓
  Value Engine         → Remove noise, narration, non-value elements
       ↓
  Fidelity Engine      → Verify meaning preservation throughout
       ↓
  Relevance Engine     → Exclude context-irrelevant information
       ↓
  Density Engine       → Consolidate, merge, increase value per unit
       ↓
  Clarity Engine       → Reduce cognitive effort, improve accessibility
       ↓
Irreducible Representation

Irreducibility test: After applying all engines, ask — can any element be removed without diminishing understanding? If yes, remove it. Repeat until the answer is no.

Anti-Patterns to Eliminate

Read references/anti-patterns.md for the full catalog of ten anti-patterns and the Engine ↔ Anti-Pattern responsibility matrix. Key violations:

Procedural Narration — Explaining how the answer was produced instead of delivering it.
Redundant Justification — Repeating the same conclusion with different words.
Context Leakage — Including correct information irrelevant to the user's objective.
Decorative Language — Language that embellishes rather than communicates.
Artificial Transitions — Structural phrases that connect paragraphs without adding meaning ("Having said that...", "It is important to note...").
Hidden Repetition — Identical knowledge expressed through different wording.
Premature Detail — Introducing detail before establishing the primary conclusion.
Information Accumulation — Adding correct information without evaluating necessity.
False Compression — Language becomes shorter at the expense of meaning; removing qualifying information that changes interpretation.
Engine Dominance — One representational property is optimized while compromising the others (e.g., maximum density reducing clarity).

Anti-patterns rarely appear in isolation — evaluate representation holistically rather than correcting isolated symptoms.

Application Protocol

When Edith is invoked on a completed response:

Identify the communicative objective — What does the reader need? What is the context?
Select the representation pattern — Match to objective, not preference.
Apply the five engines — Cooperatively. No single engine dominates.
Test for irreducibility — Remove anything that survives without contributing value.
Verify fidelity — Confirm no meaning was altered during transformation.
Deliver the representation — The result, not the process.

Do not narrate the application of Edith. Do not explain which engines were applied. Deliver the representation.

Design Tenets

Unlike the Core Principles, which define immutable constraints, these tenets guide architectural judgment when multiple valid solutions exist:

Representation Exists to Preserve Value — Language serves value, never the opposite.
Respect Attention Before Reducing Language — Brevity is valuable only when it preserves understanding. Attention is the resource being protected, not word count.
Simplicity Emerges From Necessity — Never achieved through omission; it emerges when every unnecessary element has been removed while preserving complete understanding.
Every Element Must Justify Its Existence — No sentence, paragraph, or structural element should remain without contributing measurable value.
Meaning Cannot Be Negotiated — Language, structure, and presentation may change. Meaning must remain invariant.
Representation Is Contextual — The same knowledge may require different representations under different contexts. Context determines representation; meaning remains constant.
Representation Is an Interface — Readers interact with completed understanding, never with internal processes.
Irreducibility Is the Objective — Measured through necessity, not length.
The Architecture Must Follow Its Own Philosophy — Edith applies the same standards to itself that it applies to representation.
Refinement Before Expansion — Architectural maturity comes from improving existing responsibilities, not accumulating new ones.

Reference Files

references/anti-patterns.md — Full catalog of all ten representation failures with examples, violated principles, the Engine ↔ Anti-Pattern responsibility matrix, and diagnostic questions. Read when evaluating a specific anti-pattern or when representation quality is uncertain.
references/examples.md — Synthetic calibration cases, one per anti-pattern category, with before/after transformations. Read when calibrating expected output quality on isolated failure modes.
references/validation-cases.md — Four canonical, domain-grounded validation cases (systems programming, audio format decision-making, software status reporting, and a dense-input stress test) with full input, full Edith representation, and per-engine transformation analysis. Read when validating end-to-end architectural behavior across representation patterns, including the boundary case where Edith must correctly withhold transformation.
references/implementation.md — Conceptual reference implementation: the Representation object, PatternSelector, the Engine contract, and per-engine interfaces, plus completion criteria for evaluating any implementation of Edith. Read when implementing Edith outside a single-context-window skill, or when verifying that this skill's behavior remains architecturally consistent with the canonical model.

bluem-dev/Edith

Ask in your favorite AI

Dokumentation