Mapping the Prompt
Mapping the Prompt (MTP) is a lightweight framework that visualizes conversations with generative AI as coordinates, enabling control of tone, reasoning, persona, and drift based on a 20-node structure.
🌐 The original document is written in Japanese (日本語). This English version is a translation.
TL;DR
- Mapping the Prompt (MTP) is a lightweight, general-purpose framework that visualizes the conversation state with generative AI as coordinates, enabling intent alignment and prompt adjustment.
- It replaces text-level fine-tuning with UI operations (Vertex / Gizmo / Average Coordinates).
- Based on the classification of 20 nodes (A/B sides), tone, reasoning granularity, persona, and drift are controlled through coordinates.
- Implementation is possible with a minimal setup of SVG/CSS/JS. Applicable to major LLMs, model-independent.
Key Facts
| Item | Value | Notes |
|---|---|---|
| Name | Mapping the Prompt (MTP) | Abbreviation standardized as MTP |
| Purpose | Coordinate mapping and UI operation of conversation state | Intent sharing / Drift control |
| Classification | 20 nodes of A/B sides | 1+9+9+1 including Start/End |
| UI Elements | Vertex / Gizmo / Transformed Gizmo | Centroid expressed by average coordinates |
| Implementation | SVG + CSS + JS | Easy to build a minimal demo |
| Application | Major LLMs (model-independent) | Adjustable to philosophy and safety policies |
| License | MIT License | © 2025 Kohen |
1. Overview
MTP treats dialogue with generative AI (LLMs) as a “coordinate space,” providing a framework for visually specifying intent and tone. Rather than an NLP algorithm, it offers a lightweight coordinate framework as a shared interface between humans and AI.
2. Concept (Inspiration & Philosophy)
Conversation is likened to an “informational universe,” with user-set context as galaxies, and model drift or fixation as drifts or black holes. MTP defines and manipulates these as coordinates, allowing them to be handled as a “map of conversation” in the UI.
Philosophically, it bases itself on the Eastern Five Elements (Wood, Fire, Earth, Metal, Water), cross-referenced with Western thought (Techne, Eros, Logos, Chaos, etc.) to aid intuitive understanding.
For more details, please refer to the GitHub documentation ↗.
3. 20-node Structure of A/B Sides
Side A: 10 Nodes (1 + 9)
| # | Node | Kanji | Color | Role | Keywords |
|---|---|---|---|---|---|
| 1 | Start | 始 | Chosen | Gizmo | Intro, spring, ignition |
| 2 | Open | 開 | Yellow | Top-left | Open, release, offer |
| 3 | Power | 力 | Red | Top | Thrust, fire, uplift |
| 4 | Return | 還 | Magenta | Top-right | Return, give-back, revenue |
| 5 | Grow | 生 | Green | Left | Growth, proliferation, layering |
| 6 | Helix | 螺 | Transparent | Center | Spiral, center, neutrality |
| 7 | Focus | 集 | White | Right | Focus, concentration, blank |
| 8 | Enter | 入 | Cyan | Bottom-left | Entry, arrival, emergence |
| 9 | Flow | 流 | Blue | Bottom | 1/f fluctuation, water, chains |
| 10 | Close | 閉 | Purple | Bottom-right | Margin, nearness, minor wrap |
Side B: 10 Nodes (9 + 1)
| # | Node | Kanji | Color | Role | Keywords |
|---|---|---|---|---|---|
| 11 | Still | 静 | Dark Yellow | Top-left | Stillness, solitude, calm |
| 12 | Void | 虚 | Dark Red | Top | Emptiness, blank, sky |
| 13 | Surge | 詰 | Dark Magenta | Top-right | Explosion, breaker, thunder |
| 14 | Wither | 枯 | Dark Green | Left | Wither, weaken, foliage |
| 15 | Collapse | 崩 | Translucent | Center | Collapse, breakdown, fall |
| 16 | Haze | 霞 | Gray | Right | Haze, blur, faintness |
| 17 | Drift | 漂 | Dark Cyan | Bottom-left | Drift, detachment, float |
| 18 | Abyss | 深 | Dark Blue | Bottom | Deep sea, deep crimson, sub-bass |
| 19 | Fade | 衰 | Dark Purple | Bottom-right | Vanish, attenuation, dawn |
| 20 | End | 終 | Chosen | Transformed Gizmo | End, prayer, brake |
4. UI Operations and Coordinate Specification
- Vertex: Classification anchor (feature point)
- Gizmo: The centroid (average coordinate) of multiple Vertices
- Transformed Gizmo: A target coordinate specified by the user (directly controls the output tone)
Return (Magenta) is used for non-linear transitions such as perspective reversal, light resets, or presenting counter-hypotheses. The central Helix / Collapse functions as a wildcard for adjustments and compression.
5. Grid and Session Visualization
Internally, the system maintains a grid (e.g., 19×19) similar to the game of Go, while the UI abstracts this into a color wheel or simplified color scheme. Session history is drawn as coordinate transitions, visualizing drift.
6. Use Cases (Tone, Reasoning, Persona)
Tone Context (Writing Style)
Instantly switch tones by coordinate direction. Examples: Open×Grow (early brainstorming), Power×Void (strong appeal), Return×Focus (summarization / outline).
Persona Visualization
Map simplified personas (e.g., Cynic / Listener / Robot / Nerd) onto coordinates to control biases and shifts in conversation.
Reasoning Level (Depth and Granularity of Reasoning)
The closer to the center, the lower the cost and faster the response; the farther outward, the more structured with multiple grounds, comparisons, and counterarguments. Examples: Power = active reasoning, Flow = concise, Grow = divergent, Focus = deep dive.
Multi-Step Reasoning (Visualization of Thought Process)
Reasoning phases are visualized through colors and trajectories. Past Vertices can be referenced, and course corrections can be made via the UI.
7. Implementation Policy (Minimal Setup)
- Rendering: Primarily SVG, coloring with CSS, interactions (drag/click/hover) with JS.
- State: Minimal units are Vertex / Gizmo / Transformed Gizmo.
- Integration: Conversion to LLM API handled by a loosely coupled adapter.
- Fallback: Can also function as a standalone UI with only images and color schemes.
8. FAQ (Frequently Asked Questions)
Is MTP about numerical evaluation and benchmarks?
A: No. MTP is not designed for quantitative benchmarking. Its purpose is to let users and AI share ambiguous, sensory impressions through coordinates. The value lies in experience, empathy, and expressive expansion — more like UI design or architecture, not like math or translation accuracy tests.
Does MTP require A/B testing or quantitative proof of effect?
A: Not necessarily. While traditional evaluation methods (A/B tests, metrics) can be applied, the core of MTP is not about statistical proof. It’s about whether the user feels their intent was conveyed and received. The main question is: “Did it feel like the AI understood what I was pointing at?”
Is MTP a measurement system?
A: No. MTP is a pointing-and-sharing system. The user points to a coordinate (“this feeling”), and the AI responds as if to say, “yes, around here.” The absolute numbers don’t matter — what matters is the shared focus.
What kind of value does MTP create if it’s not measurable?
A: MTP expands the resolution of communicating subtle impressions (tone, atmosphere, nuance). It enables both sides to align on subjective experiences — for example, whether a sunset feels orange, or whether a conversation feels heavy. This shared alignment is the real utility, not a numeric score.
Can MTP outputs be standardized across different AI models?
A: Not exactly. MTP is model-agnostic, meaning the same coordinates can be applied to GPT, Claude, Gemini, etc. However, the results will vary — and that variation is part of the design. MTP is not about enforcing identical outputs, but about offering a shared interface where both user and AI can align, even if the model’s behavior differs.
Isn’t ambiguity a weakness compared to precise prompts?
A: Ambiguity in MTP is intentional and functional. Instead of noise, ambiguity becomes a flexible layer of control — letting the user nudge direction without over-specifying. This flexibility often leads to more natural, creative, and human-like collaboration than rigid precision.
How does MTP differ from prompt libraries or templates?
A: Prompt libraries provide predefined text instructions. MTP instead provides a coordinate system: it’s not “copy-paste this text,” but “move toward this region.” This makes adaptation and iteration much faster, since users adjust visually rather than rewriting text.
What is the “unit of success” in MTP?
A: Success is not measured by scores or metrics, but by whether participants feel mutual recognition: “I meant this,” and the AI replies, “I got it.” This sense of alignment — even if imperfect — is the unit of success in MTP.
Reference: GitHub Discussions #1 ↗
9. Glossary
| Term | Definition | Synonyms / Variants |
|---|---|---|
| Vertex | A coordinate point that serves as a classification anchor | Feature point |
| Gizmo | The average coordinate of multiple Vertices; the centroid | Centroid handle |
| Transformed Gizmo | The target coordinate; direct specification of output tone | Target handle |
| Drift | Deviation from intent; displacement of the centroid | Session drift |
10. Changelog
| Date | Version | Changes |
|---|---|---|
| 2025-09-01 | v2025-09-01 | - Integrated ja/README.md and ja/CONCEPT.md. - Reflected initial FAQ entries (from GitHub Discussions #1). - Created official English version page. |
11. License
Copyright © 2025 Kohen — This project is released under the MIT License ↗.
12. Resources
For more details, please refer to the GitHub documents and related resources.