A user has published a detailed prompt titled "Decision-First Algorithm v2.5" designed to address common frustrations with AI models, such as performing unrequested actions, reopening settled issues, or deviating from the intended path.
The algorithm instructs the AI to apply a structured decision-making procedure before answering requests that require judgment, analysis, or problem-solving. Key steps include identifying the actual outcome needed, building the minimum necessary logical structure, and independently validating governing conditions. The prompt emphasizes locking confirmed decision structures to prevent unnecessary re-evaluation and prioritizing counterexamples and competing hypotheses over mere relevance.
The author notes that this approach worked better than expected for themselves and others who tested it, suggesting it helps users interact with AI more effectively by enforcing stricter logical discipline.