Prompt engineering is fundamentally a specification problem: every implicit assumption left out of the prompt is a degree of freedom the model may resolve differently than intended.
satisfied(req) when clarity(t) > ambiguity * req.strictness
compliance = satisfied / total
- Requirement nodes — individual constraints the prompt asks the model to satisfy (format, tone, length...).
- Specification density — how many explicit constraints are packed per instruction — higher links more requirements together.
- Instruction clarity — how unambiguously each constraint is worded, raising activation toward compliance.
- Ambiguity tolerance — how much vague wording the model is allowed before a requirement silently fails.
Well-specified prompts reduce the search space of valid completions, which is why explicit output schemas and examples reliably outperform vague instructions.