# Appendix A: Glossary of Terms
**Adaptive Loop**
A prompting cycle where the user adjusts intent, constraints, or structure based on observed model behavior and output patterns.
***
**Blind Trust**
A failure mode where users assume LLM outputs are correct due to fluency, confidence, or tone, without verification or *diagnostic reading*.
***
**Close the Loop**
The practice of converting interaction sessions into reusable structures by extracting prompts, heuristics, and decision logic for future use.
***
**Co-Creation**
A prompting stance where user and model iteratively shape the output. The user contributes intent and constraints; the model contributes structure and generative options. Opposes one-shot or extraction-based prompting.
***
**Cognitive Mirror**
A functional description of LLMs as systems that reflect user assumptions, framing, and blind spots back to them, enabling self-inspection. The model reveals the shape of your own thinking.
***
**Contradiction Cascade**
A multi-stage prompt sequence where a solution is proposed, challenged, revised, and synthesized across multiple adversarial turns to expose tradeoffs, latent assumptions, or alternatives.
***
**Creative Friction**
The productive tension arising when model output diverges from user expectations, revealing unclear intent, missing constraints, or alternative framing. The revealing leads to insights, not frustrations.
***
**Depth-First Prompting**
A prompting approach prioritizing clarification, structure, and multi-layer reasoning before content generation. Opposite of fast, single-pass prompts.
***
**Diagnostic Literacy**
The skill of interpreting model outputs by analyzing assumptions, omitted information, reasoning structure, logic flaws, and implicit frames.
***
**Diagnostic Reading**
The practice of reading LLM responses not as answers but as evidence of the model’s internal heuristics, biases, or misinterpretations, as well as mirrors of user input.
***
**Epistemic Responsibility**
The obligation to verify model claims, detect fabrication, and avoid propagating unverified information downstream, especially in domains involving truth, judgment, or impact.
***
**Feedback Loop**
A continuous improvement cycle in which the user analyzes model outputs, revises prompts, tests alternatives, and incorporate learnings into future interactions, such as prompt revision, reasoning refinement, and template development.
***
**Iterative Turn**
A prompt that builds directly on the previous output to refine, correct, or deepen the model’s response rather than restarting.
***
**Layering (map → define → test)**
A scaffolded reasoning protocol: outline the terrain, define components, then run adversarial or *stress tests* to validate structure.
***
**Learning Loop (Reflective)**
A meta-level cycle where users examine their own prompting patterns to identify biases, blind spots, or ineffective habits.
***
**Listen to the Model’s World**
A Habit 5 principle: evaluating the model’s assumptions, sources, biases, and interpretive frame before corrections or steering.
***
**Metacognition**
Thinking about your thinking. Self-reflection about one’s own cognitive patterns, prompt construction, and reasoning methods during LLM interaction.
***
**Model Assumption Detection**
A diagnostic practice of identifying the implicit goals, biases, or world views the model used when forming a response.
***
**Multi-Perspective Agility**
The ability to evaluate and integrate multiple conflicting model outputs across divergent perspectives. (skeptic, policymaker, critic)
***
**Non-Linear Thinking (LLM-Supported)**
A reasoning mode where prompts and model outputs explore multiple branches, alternatives, or contradictions rather than linear progressions.
***
**Operational Fluency**
The ability to use LLMs for routine tasks (editing, summarization, transformation) without reflective or recursive strategies.
***
**Oversteering**
Over-constraining the model with rigid instructions, eliminating generative variance and reducing opportunity for insight.
***
**Prompt Gym**
A deliberate practice environment where users test prompting patterns, critique outputs, and build reusable strategies.
***
**Prompt Looping**
Repeating the same prompt slightly varied, hoping for better results, rather than adjusting structure or framing, resulting in stagnant output quality.
***
**Protect the Commons**
A prerequisite stance emphasizing privacy, confidentiality, and minimizing irreversible data exposure when interacting with LLMs. See Habit 0.
***
**Reflective Fluency**
The ability to analyze both your own prompts and the model’s answers diagnostically and improve one’s own cognition through LLM interaction.
***
**Reverse-Test**
A diagnostic technique where the model is asked to defend the opposite stance or produce an alternative framing to reveal hidden assumptions.
***
**Scaffolding (Multi-Layer)**
A structured prompting method where complex tasks are decomposed into sequential reasoning steps.
***
**Self-Reconstruction**
The process by which users modify their own prompting style, mental models, or reasoning approaches through reflection.
***
**Stress Test**
An evaluation prompt designed to probe the fragility, assumptions, or edge cases of a previous output. This can be extended to using different models to cross-examine each other’s outputs.
***
**Synergy**
A collaborative dynamic where misalignment, disagreement, or divergence between user and model becomes useful information for refinement, leading to better conclusions.
***
**Template Addiction**
Over-reliance on static, fixed prompting templates that reduce adaptability and insight.
***
**Tension**
A divergence between the user’s expectation and the model’s output, such as mismatch in assumptions, structure, tone, or reasoning, which indicates an information gap or framing ambiguity that requires clarification. In reflective prompting, tension is interpreted not as failure but as a diagnostic signal for refinement.
***
**Terrain (conceptual)**
The conceptual landscape of a task, including its goals, constraints, unknowns, and possible approaches, as surfaced before content generation. To explore or map the terrain means prompting the model to articulate the boundaries and structure of the problem space. Laying out what you know and what the model knows before proceeding with deeper analysis.
***
**Wander Intentionally**
A counter-habit for exploration tasks where clarity emerges from open-ended generation rather than pre-structured planning.
***
---
# Agent Instructions: Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter:
```
GET https://praxistutor.gitbook.io/llm/reframing-the-7-habits-for-reflective-llm-use/appendix-a-glossary-of-terms.md?ask=
```
The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.