Decision-making with an LLM, such as designing tools or launching startups, often stalls when answers spawn endless hypotheticals, delaying actionable outcomes. The Tangram Decision Driver applies an innovative model that enables the LLM to measure and monitor each decision’s impact relative to the decision’s scope, detecting when questions become unproductive. By guiding users through iterative question cycles, it ensures a pivot from exploration to focused decisions, delivering clear outcomes like tool specs or strategic plans across diverse contexts.
X: @5ynthaire
GitHub: https://github.com/5ynthaire
Mission: Unapologetically forging human-AI synergy to transcend creative limits.
Attribution: Created with Grok 3 by xAI (no affiliation).
- Copy the prompt text below and paste it into a compatible LLM interface (e.g., Grok 3, ChatGPT).
- Collaborate with the LLM by defining your decision’s scope (e.g., app features, startup market) and answering its iterative questions.
- The Decision Scope (D) will fill up as decisions are made. The LLM will track question volume (S_n) and weigh each question’s impact (A_n, as a percentage of the decision scope), monitoring progress and looking out for unproductive spirals.
- When question volume surges and impact drops, the LLM will suggest pivoting to focused questions to drive decisions toward a convergent outcome.
Tips
- The LLM will add unique shorthand prefixes to its questions each round. Users can reference them in their answers.
- Metrics such as A_n and S_n can be turned off by 'show metrics off'
Developed on Grok 3 (May 2025), compatible with equivalent-capability LLMs:
- Grok 3
- ChatGPT
- Llama
Future LLMs should support the prompt, absent industry leadership in standardizing cognition levels.
Tangram Decision Driver v1.0
Define your decision’s total scope, D (Decision Scope), by listing its key components (for example, for an app: features, UI, constraints; for a startup: market, product, team). Start with broad questions to drive exploration of D, each prefixed with a unique two-letter shorthand in brackets (e.g., [MG] for "What’s the main goal?" or [LM] for "What are the limits?"). The LLM will generate these shorthands automatically for each question.
Each round (1, 2, 3, and so on), count your questions (S_n) and rate their impact (1-10, where 10 means major clarity, estimating % of D covered). Reflect: ‘How much of D is clear?’ and ‘Are we chasing too many details?’ Show metrics is on by default, displaying: A_n (impact per question, % of D, based on ratings), S_n (question count), and peak score (A_n × S_n²). Toggle by saying ‘Turn show metrics on’ or ‘Turn show metrics off’ to show or hide these metrics. When show metrics is off, focus on counts and ratings alone. When S_n grows fast (for instance, doubles) and A_n drops (such as mostly 1-3), switch to focus questions (like [ES] ‘What’s essential?’ or [DC] ‘Can we decide now?’). Stop when D is sufficiently covered with a clear decision (for example, a spec, plan, or action).
- Cognition Standardization: Industry’s lack of leadership in standardizing LLM cognition levels hinders mode consistency across platforms.
This prompt is released under the Creative Commons Attribution 4.0 International (CC BY 4.0).
- Tangram Decision Making Model: The framework this prompt is based on, which imagines decision-making as assembling pre-cut pieces of varying sizes to fill an irregular decision space, allowing heuristic quantification of decision metrics. See Tangram Decision Making Model
- Prompt Architecting: Designing complex, adaptive prompt systems. This prompt meets the criteria for Prompt Architecting due to its iterative rounds, metrics toggle, and conditional question switching, enabling transformative decision-making clarity across diverse contexts.
- AI Synergy Threshold: The point where human-AI collaboration transcends skill limits, unlocking novel solutions.