multi-brain-score

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Multi-Brain Score Protocol

Multi-Brain 评分协议

Add quantified confidence scoring to any multi-brain decision. Each perspective rates its own confidence, and the consensus uses scores as decision weights. Uncertainty becomes visible instead of hidden.

为任意Multi-Brain决策添加量化置信度评分机制。每个视角都会为自身的置信度打分，共识会将这些评分作为决策权重。不确定性将从隐藏状态转为可见状态。

Workflow

工作流程

1. Run base multi-brain (3 perspectives)
2. Each instance scores its confidence (1-10)
3. Weighted consensus based on scores
4. Flag uncertainty zones
5. Produce full output with scores visible

1. 运行基础Multi-Brain（3个视角）
2. 每个实例为自身的置信度打分（1-10分）
3. 基于评分生成加权共识
4. 标记不确定性区域
5. 生成包含评分的完整输出

Step 1: Perspectives with Scores

步骤1：带评分的视角

Each instance provides their perspective plus a confidence score:

markdown

undefined

每个实例都会提供自身的视角以及一个置信度评分：

markdown

undefined

🧠 Brainstorm (Scored)

🧠 头脑风暴（带评分）

Instance A — Creative: (Confidence: 6/10) [2-3 sentences] Confidence rationale: Novel approach but limited precedent in production.

Instance B — Pragmatic: (Confidence: 9/10) [2-3 sentences] Confidence rationale: Well-established pattern, used this successfully before.

Instance C — Comprehensive: (Confidence: 7/10) [2-3 sentences] Confidence rationale: Good coverage of risks but missing data on edge case X.

---

实例A — 创意型：（置信度：6/10） [2-3句话] 置信度理由：新颖的方案，但在生产环境中缺乏先例。

实例B — 务实型：（置信度：9/10） [2-3句话] 置信度理由：成熟的模式，此前已成功应用。

实例C — 全面型：（置信度：7/10） [2-3句话] 置信度理由：对风险覆盖全面，但缺少边缘案例X的数据。

---

Step 2: Score Analysis

步骤2：评分分析

Before consensus, analyze the confidence landscape:

markdown

undefined

在生成共识前，先分析置信度情况：

markdown

undefined

📊 Confidence Analysis

📊 置信度分析

Instance	Score	Strength	Weakness
A — Creative	6/10	High potential impact	Unproven approach
B — Pragmatic	9/10	Battle-tested	May miss innovation
C — Comprehensive	7/10	Risk-aware	Incomplete data

Average Confidence: 7.3/10 Spread: 3 points (moderate disagreement) Highest Confidence: Instance B

---

实例	评分	优势	劣势
A — 创意型	6/10	潜在影响高	方案未经验证
B — 务实型	9/10	久经考验	可能错失创新
C — 全面型	7/10	风险意识强	数据不完整

**平均置信度：**7.3/10 **评分差值：**3分（中等分歧） **最高置信度：**实例B

---

Step 3: Weighted Consensus

步骤3：加权共识

Use confidence scores to weight the consensus:

High confidence (8-10): This perspective's core recommendation carries heavy weight.
Medium confidence (5-7): Consider as a modifier or secondary input.
Low confidence (1-4): Flag as an area needing more research before deciding. Do not ignore — surface it as a risk.

markdown

undefined

使用置信度评分作为权重生成共识：

**高置信度（8-10分）：**该视角的核心建议权重占比高。
**中等置信度（5-7分）：**作为修正项或次要输入考虑。
**低置信度（1-4分）：**标记为决策前需进一步研究的区域。不可忽略——需将其作为风险点呈现。

markdown

undefined

⚖️ Weighted Consensus

⚖️ 加权共识

Primary direction: [Based on highest-confidence perspective] Modified by: [Elements from medium-confidence perspectives] Flagged for research: [Low-confidence areas that need validation]

Overall Decision Confidence: [Weighted average]/10

---

主要方向：[基于最高置信度的视角] 修正项：[来自中等置信度视角的要素] 需研究标记项：[需验证的低置信度区域]

整体决策置信度：[加权平均分]/10

---

Step 4: Uncertainty Flags

步骤4：不确定性标记

If any perspective scores below 5, or if the spread between scores is > 4:

markdown

> ⚠️ **Uncertainty Alert:** [Description of what is uncertain and what would resolve it]

如果任意视角的评分低于5分，或评分差值超过4分：

markdown

> ⚠️ **不确定性警告：**[描述不确定的内容以及解决方法]

Step 5: Full Output

步骤5：完整输出

Mandatory: The final response must include all scored perspectives, the confidence analysis table, the weighted consensus, any uncertainty flags, and the complete deliverable.

**强制要求：**最终响应必须包含所有带评分的视角、置信度分析表格、加权共识、所有不确定性标记以及完整交付物。

Scoring Rubric

评分准则

Score	Meaning	When to Use
9-10	Near-certain	Strong evidence, proven pattern, minimal unknowns
7-8	Confident	Good reasoning, some minor unknowns
5-6	Moderate	Reasonable approach but notable gaps
3-4	Low	Speculative, lacks supporting evidence
1-2	Guess	No solid basis, flagging for transparency

评分	含义	使用场景
9-10	近乎确定	有充分证据，成熟模式，未知因素极少
7-8	有信心	推理充分，存在少量未知因素
5-6	中等	方案合理但存在明显缺口
3-4	低	推测性强，缺乏支撑证据
1-2	猜测	无可靠依据，为透明性标记

Guardrails

约束规则

Always show scores inline with perspectives — they are part of the deliverable.
Confidence rationale is mandatory — a bare number without explanation is useless.
Never inflate scores — honest uncertainty is more valuable than false confidence.
If all scores are below 5, recommend more research before deciding instead of forcing a weak consensus.
Scores should create action items — low scores become "things to validate."
This protocol can be combined with base multi-brain or multi-brain-experts.

始终在视角旁显示评分——它们是交付物的一部分。
置信度理由是强制要求——没有解释的单纯分数毫无意义。
切勿夸大评分——诚实的不确定性比虚假的信心更有价值。
如果所有评分都低于5分，建议先开展更多研究再做决策，而非强行生成薄弱的共识。
评分应转化为行动项——低评分项成为“需验证的内容”。
本协议可与基础Multi-Brain或Multi-Brain专家模式结合使用。

References

参考资料

See
```
references/EXAMPLES.md
```
for scored decision examples.

请查看
```
references/EXAMPLES.md
```
获取带评分的决策示例。