simulate-elite-experts

Simulate Elite Experts

中文介绍

simulate-elite-experts 用于处理高复杂度、高不确定性或较难逆转的决策问题。它不是普通的“专家角色扮演”，而是一个带硬约束的四视角推理框架：选择两位与问题直接相关、可从公开作品推断方法论的真实人物，再加入一个领域专家抽象角色和一个全知智能体抽象角色，让它们围绕同一决策进行立场陈述、交叉质询、修正和综合。

适合使用的场景：

• 需要比较多个强观点，而不是只要一个直接答案。
• 决策涉及速度、质量、风险、成本、组织、伦理等多重约束。
• 用户希望知道“应该找哪些顶级人物/视角来讨论这个问题，以及他们可能会怎么说”。

不适合用于简单事实查询、单一正确答案、低成本可逆选择或纯情绪支持。输出必须区分事实、假设和推测；对真实人物的观点只能基于公开资料做模拟推断，不能伪造直接引语、私人意图或未验证事实。

Applicability Pre-Check (Gate 0)

Before running the full framework, evaluate 3 dimensions to decide whether this tool is appropriate:

1. Complexity: Does the problem involve multiple stakeholders, competing constraints, or non-obvious tradeoffs?

   • Yes (score 1) → proceed.
   • No (score 0) → consider skipping; a direct answer may suffice.

2. Reversibility: Is the decision hard to reverse once made?

   • Hard to reverse (score 1) → proceed; structured deliberation is valuable.
   • Easy to reverse (score 0) → consider skipping; low-cost decisions rarely need multi-lens analysis.

3. Information ambiguity: Is there significant uncertainty about facts, outcomes, or stakeholder preferences?

   • High ambiguity (score 1) → proceed; the framework surfaces hidden assumptions.
   • Low ambiguity (score 0) → consider skipping; the answer may already be clear.

Decision rule:

• Total >= 2 → use the full framework.
• Total = 1 → offer the user a choice between micro (lightweight) and full framework.
• Total = 0 → recommend skipping; inform the user this question is better answered directly.

If the user explicitly requests the framework regardless of score, proceed but note the pre-check result.

Execution Mode Selection

Choose the execution mode before building the roster:

• one-shot (default): Produce the complete output in one response. Use this when the user asks for an analysis, report, comparison, or direct recommendation and has not asked to co-design the panel.
• interactive: Produce Section 1 only, ask the user to confirm or swap the roster, then wait before running dialogue rounds. Use this when the user explicitly asks to choose experts, workshop the panel, or collaborate step by step.
• blind-spot-gated (opt-in overlay): Run the candidate deliberation, then perform a clean-context outside audit across 5 failure categories (hidden assumptions, wrong framing, omitted alternatives, second-order effects, simpler paths) evaluated by a materiality gate. If material: true, trigger a single-shot re-entry to revise the decision before final delivery. Use this only when explicitly requested (e.g. --blind-spot-gated or asking for "gated audit").