Vocabulary Bridge

Terminology translator.

The Living Framework uses precise terminology that does not always map directly to common AI industry vocabulary. This page bridges the two: find the term you know, and see the concept the framework uses to describe that phenomenon — with more rigour.

Industry terms are often underspecified — they describe what is happening without explaining the structural mechanism that causes it. The Living Framework terms name the mechanism. The translations below show why the concepts are related, where they differ, and which papers define the precise meaning.

Context and memory
Industry term
Living Framework concept
Paper
Context engineering
How you structure the information given to an LLM
Context Engineering / Context Architecture
Paper 01 documented this as a structural reliability discipline in 2024. The concept gained mainstream traction in 2025–2026. The LF frame emphasises what happens to context over time — not just at input, but through the duration of a collaboration.
Paper 01 (2024) Paper 02 (2024)
Context window management
Techniques for staying within token limits
Context Integrity
The LF concept covers more than token management. Context integrity refers to whether the shared understanding between human and AI remains accurate and undegraded — across time, not just at any one moment. Token management is one tactic; context integrity is the structural condition it serves.
Paper 01 (2024) Paper 02 (2024)
AI memory
Persistent storage of information across sessions
Memory Governance
Industry discussions focus on what gets stored. The LF concept asks: how is memory structured, verified, updated, and kept coherent? Storage is only one dimension of a governed memory architecture. Memory that is stored but unverified, or stored but structurally inconsistent, introduces reliability risks that raw persistence does not address.
Paper 01 (2024) Paper 02 (2024)
RAG / Retrieval-Augmented Generation
Retrieving external knowledge at inference time
Memory Governance (retrieval layer)
RAG names a technical mechanism. The LF frame asks whether retrieved content is reliable enough to be incorporated into the active context without introducing integrity failures. Retrieval quality and coherence with existing context are governance questions, not engineering questions alone.
Paper 02 (2024)
Context drift
When the LLM seems to forget earlier instructions
Context Drift (Failure Mode 1)
This is a named failure mode in the Living Framework — not just a symptom. The LF describes its mechanism (gradual degradation of established parameters over extended sessions), its signals, and its structural causes. It was documented in Paper 03 in 2024 before the term was widely used in practitioner writing.
Paper 03 (2024) Failure Library
Reliability and verification
Industry term
Living Framework concept
Paper
LLM reliability
How consistently an AI system produces correct outputs
Reliability Architecture
The LF treats reliability as an architectural property of the human-AI system — not a property of the model alone. A reliable architecture can produce reliable outputs even from a model that is individually imperfect. The framing shifts responsibility from the model to the system design.
Paper 01 (2024) Paper 06 (2025)
Grounding
Connecting AI outputs to verifiable facts or sources
Verification Layer
Grounding describes what we want: outputs connected to reliable sources. The Verification Layer describes the structural mechanism that ensures this happens — a dedicated layer in the architecture, separate from generation, whose function is to check outputs before they are incorporated or acted on. Paper 08 (2025) argued this layer is the most consistent structural gap in current AI reliability architectures.
Paper 08 (2025)
Hallucination
When an AI confidently states something false
Trust Deficit / Numerical Error Propagation
"Hallucination" is a single label for two distinct structural problems. Numerical Error Propagation (Failure Mode 3) describes false quantitative claims that enter calculations and amplify. Trust Deficit (Failure Mode 5) describes the broader collapse of reliable collaboration that follows repeated undetected errors. Each has different mechanisms, different signals, and different structural responses.
Paper 03 (2024) Failure Library
AI safety
Preventing AI from causing harm
Reliability Engineering (operational scope)
AI safety in the public discourse typically addresses existential or systemic risk from advanced AI systems. The Living Framework addresses operational reliability — the structural failure modes that cause real harm in current AI collaborative workflows. These are related but distinct concerns. The LF does not address existential AI risk; it addresses the reliability failures already occurring in deployed systems.
Paper 01 (2024) Research Overview
Collaboration and agency
Industry term
Living Framework concept
Paper
AI collaboration
Working alongside an AI system on tasks
Distributed Cognitive System
The LF frames human-AI collaboration not as human-using-tool but as a distributed cognitive system — two processors with different capabilities and failure modes sharing cognitive work. This reframe has practical consequences: it changes where you look for reliability problems (in the system, not just the model) and what architectural decisions matter.
Paper 07 (2025)
Agentic AI / AI agents
AI systems that take multi-step actions autonomously
Distributed Cognition / Verification Layer
Agentic AI raises the stakes for the reliability architecture because errors compound across action sequences. The LF concepts are directly applicable: distributed cognition provides the frame for understanding the human-AI division of responsibility in agentic systems, and the verification layer is even more critical when actions cannot be easily reversed.
Paper 07 (2025) Paper 08 (2025)
Prompt engineering
Crafting inputs to get better AI outputs
Instruction Architecture / Language Reliability
Prompt engineering describes a practice. The LF concepts describe the structural properties that make instruction-giving reliable: how instructions are structured to maintain integrity, how linguistic patterns in AI output signal reliability problems (Paper 05), and how language choices affect the collaborative architecture rather than just individual outputs.
Paper 01 (2024) Paper 05 (2025)
Long-context AI
AI models with large context windows
Context Integrity over extended sessions
A large context window is a capability; context integrity is the reliability condition. Long-context AI makes extended collaboration technically possible — but it does not automatically make it reliable. The LF research focuses precisely on what breaks down over extended sessions, even as raw context capacity increases.
Paper 01 (2024) Paper 02 (2024)
Failure and recovery
Industry term
Living Framework concept
Paper
Knowledge loss / forgetting
When important context disappears from the collaboration
Reconstruction Failure (Failure Mode 6)
The LF names this as a structural failure mode with a specific mechanism: the inability to rebuild a reliable working state after a session break or context loss event. It is distinct from context drift (which is gradual degradation) and addresses the recovery problem specifically — what architecture enables reliable reconstruction, and what prevents it.
Paper 03 (2024) Failure Library
AI errors / mistakes
Incorrect outputs from an AI system
Six named structural failure modes
The LF does not treat AI errors as a single category. Paper 03 (2024) documented six structurally distinct failure modes, each with a different mechanism, different signals, and different architectural response. Treating all errors as the same kind of event leads to responses that address the symptom without the cause.
Paper 03 (2024) Failure Library
Domain specialisation
AI maintaining expertise in a specific subject area
Domain Erosion (Failure Mode 4)
Domain Erosion describes what happens when domain-specific precision degrades over a session — the AI gradually reverts to generalist language as specialised context is crowded out or deprioritised. It is the failure condition that domain specialisation is trying to prevent, and understanding its mechanism informs how to structure domain context for durability.
Paper 03 (2024) Failure Library
On terminology

Why the distinctions matter

The Living Framework uses precise terminology not for novelty but because imprecise terms make it harder to identify and address structural problems. When "hallucination" covers both numerical propagation errors and systemic trust collapse, a team that addresses one thinks it has solved the other. When "context engineering" describes any act of structuring an LLM prompt, it obscures whether the intervention addresses context integrity over time — which is the reliability concern.

The translations on this page are not corrections of industry vocabulary. Industry terms are useful for the phenomena they name. The purpose here is to bridge the two vocabularies — so that a practitioner who knows what context drift feels like can find the structural account of what is causing it, and what changes the architecture needs to prevent it.

If a term you use regularly is missing from this page, or if a translation seems imprecise, the contact on the Advisory page is the right channel.