Meta Series 03: Atomic Protocol — How to Calculate Truth_Score

Why Do We Need a Score?

In Meta 02, we introduced the Calibration Triangle — three questions: Is the logic sound? Can it be applied? And how complete is it?
But “how complete” still feels vague. Can we turn that into a more precise number?
Yes. That’s the purpose of the Atomic Protocol and Truth_Score.

Truth_Score is a 0–100 score that quantifies the “real-world credibility” of a product, a piece of information, or a decision. It doesn’t aim for 100% absolute truth (which almost never exists). Instead, it helps you quickly decide in an imperfect world: “Is this worth my time, money, or trust?”

The Three Dimensions of the Atomic Protocol

We break any “claim” or “product” down into three fundamental atomic layers:
- Material: What is the physical composition or data source?
- Process: How was it made or generated?
- Connection: What are its real-world interaction records?

These three dimensions correspond to what we discussed earlier: material, process, and connection. We can score each dimension from 0 to 100 based on evidence quality, then calculate a weighted result.

The Truth_Score Formula (Practical Version)

Truth_Score = (Material_Score × 0.6) + (Process_Score × 0.3) + (Connection_Score × 0.1)
The weights can be adjusted depending on the context. For most everyday products and information, 0.6 / 0.3 / 0.1 is a solid starting point.

How to Score Each Dimension:

Material Score
- 0–20: No verifiable source (e.g. “natural ingredients” with no origin or list).
- 21–60: Basic description (e.g. ingredient list but no traceability).
- 61–90: Detailed supply chain information (traceable to origin and batch).
- 91–100: Third-party verification or blockchain record (organic certification, lab reports).

Process Score
- 0–20: Completely opaque (black box).
- 21–60: General process description only (manual but no time/energy data).
- 61–90: Traceable time, location, and steps (production logs, factory photos).
- 91–100: Process can be independently reproduced or has been third-party audited.

Connection Score
- 0–20: Almost no real feedback.
- 21–60: Some reviews, but low credibility (e.g. all 5-stars with no text).
- 61–90: Large volume of genuine Google reviews, on-site photos, delivery records.
- 91–100: Multiple independent sources cross-verify with no major contradictions.

Real Case: Evaluating an “AI Interview System”

A company is selling an AI automated interview system to a hotel, claiming “95% accuracy”.

• Material: No training data source provided, no explanation of how “accuracy” is defined → 15
• Process: Black-box algorithm, no details on handling service industry emotions or improvisation → 10
• Connection: No calibration with the hotel’s historical data, no independent third-party testing → 5
  

Truth_Score = (15 × 0.6) + (10 × 0.3) + (5 × 0.1) = 12.5

A score of 12.5 means extremely low credibility. The hotel owner’s decision to reject it was perfectly rational.

L1–L3 Tiered Verification

You can choose different depths of verification based on risk level:
- L1 (Quick Scan): ≤10 minutes — only check Material and Connection. Suitable for daily shopping or quick news verification.
- L2 (Standard Calibration): 30–60 minutes — full Truth_Score calculation. Suitable for most business decisions.
- L3 (Deep Audit): 2+ hours — independent verification of original evidence. Suitable for high-risk investments or legal matters.

Reality Check’s private Tier 3 service provides this L3-level deep calibration. "For L3 deep audits where the stakes are high, our Reality Check framework provides the professional scrutiny needed."

The Limits of the Atomic Protocol

Truth_Score only works for things with physical or verifiable anchors. Pure art, emotion, and creativity are hard to score.
If data is completely blocked (e.g. state secrets), calculation becomes impossible.
It is a helpful tool, not a final verdict. Human intuition and tolerance for uncertainty remain irreplaceable.

Conclusion

The Calibration Triangle tells you: when truth is unavailable, ask three questions.
The Atomic Protocol turns one of those questions (completeness) into a calculable score.
You don’t need to run a full calculation every time. But when facing high-risk, high-value decisions, this framework helps turn vague “feelings” into comparable “data”.

In Meta Series 04, we will discuss: Synchronization Rate — the real threshold and practice of human-AI collaboration.

Next in the Series
Meta Series 04: Synchronization Rate — The Threshold and Practice of Human-AI Collaboration
Meta Series 05: Stance Evolution Theory — No Right or Wrong, Only Generation and Resolution (in preparation)

"© 2026 Reality Check Framework. Designed for Human-AI Synergy."

Meta Series 03：原子協議——如何計算 Truth_Score

為什麼需要一個分數？

在 Meta 02 裡，我們提出了「校準三角」——三個問題：邏輯通不通？能不能應用？完善度多少％？

但「完善度多少％」聽起來還是比較模糊。我們能不能給出一個更具體的數字，讓判斷變得更快、更可比較？
可以。這就是 原子協議 和 Truth_Score 的目的。

Truth_Score 是一個 0–100 的分數，用來量化一個產品、一段資訊、或一個方案的「真實可信度」。它不追求 100% 絕對真實（那幾乎不可能），而是幫助你在資訊不完美的世界裡，快速決定「這個東西值不值得投入時間、金錢或信任」。

原子協議的三個維度

我們把任何一個「宣稱」或「產品」拆解成三個最基本的原子層級：材料、過程、連繫。
- 材料：它的物理成分或數據來源是什麼？
- 過程：它是如何被製造或產生的？
- 連繫：它與真實世界的交互記錄是什麼？

這三個維度對應你之前提到的「材料、過程、連繫」。每一維度我們都可以根據證據質量給分（0–100），再加權計算。

Truth_Score 公式（實戰版）

Truth_Score = (Material_Score × 0.6) + (Process_Score × 0.3) + (Connection_Score × 0.1)

權重可以根據場景調整（例如醫療或法律決策，可以把 Material_Score 權重提高到 0.8）。但對大多數日常產品和資訊來說，0.6 / 0.3 / 0.1 是很好的起點。

如何給每個維度打分？

材料分數（Material_Score）
如果完全沒有可驗證的來源，例如只說「天然成分」卻沒有產地或成分表，那分數就會很低，大概在 0–20 分之間。
如果有基本的成分表，但無法追溯供應鏈，則落在 21–60 分。
有詳細供應鏈資訊、可追溯到產地和批次，能拿到 61–90 分。
如果還有第三方驗證或區塊鏈存證，則可以達到 91–100 分。

過程分數（Process_Score）
過程完全不透明、像黑盒一樣，基本上只有 0–20 分。
只有一般性流程描述，沒有時間或能量數據，大概 21–60 分。
如果有可追溯的時間、地點和步驟記錄，例如生產日誌或工廠照片，就能到 61–90 分。
過程可以被獨立複現或經過第三方審計，則是 91–100 分。
連繫分數（Connection_Score）
完全沒有客戶反饋或第三方評價，只有 0–20 分。

有少量評價，但看起來可信度低（例如全是五星卻沒有文字），大概 21–60 分。
如果有大量真實的 Google 評論、現場照片、配送記錄，就能達到 61–90 分。
當有多個獨立來源交叉驗證，而且沒有明顯矛盾時，就可以拿到 91–100 分。

案例：用 Truth_Score 評估「AI 自動面試系統」

假設某公司向酒店推銷一套 AI 面試系統，宣稱「準確率 95%」。

材料方面，他們沒有提供訓練數據來源，也沒說明如何定義準確率，只能給 15 分。
過程是黑盒算法，完全沒有說明如何處理服務業的情緒與應變能力，只有 10 分。
連繫方面，沒有該酒店歷史數據的校準，也沒有獨立第三方測試報告，只有 5 分。

計算下來：(15×0.6) + (10×0.3) + (5×0.1) = 12.5 分。
Truth_Score 只有 12.5 → 極低可信度。酒店老闆拒絕是非常理性的決定。

L1–L3 分層驗證

你可以根據風險程度選擇不同驗證深度：
- L1（快速掃描）：10 分鐘以內，只看材料和連繫，適合日常購物或判斷新聞真偽。
- L2（標準校準）：30–60 分鐘，完整計算 Truth_Score，適合大多數商業決策。
- L3（深度審計）：2 小時以上，獨立驗證每一維度的原始憑證，適合高風險投資或法律相關事項。
當真實變得昂貴，你需要的是一套可計算的邏輯。

針對風險極高、不容有失的 L3 深度審計，我們的 Reality Check 架構能提供所需的專業精密校準。

原子協議的極限

Truth_Score 只適用於有物理或可驗證錨點的事物。純粹的藝術、情感、創意很難打分。如果數據完全被封鎖（例如國家機密），也無法計算。
它是一個輔助工具，不是最終判決。人類的直覺和對不確定性的容忍度，仍然不可替代。

結語

校準三角告訴你：當真實不可得時，問三個問題。
原子協議則把其中一個問題（完善度）變成一個可計算的分數。
你不需要每次都做複雜計算。但當你面對高風險、高價值的決策時，這個框架能幫助你把模糊的「感覺」轉化為可比較的「數據」。

在 Meta Series 04，我們將討論：同步率——人機協作的真正門檻與練習方法。

後續預告
Meta Series 04：同步率——人機協作的門檻與練習
Meta Series 05：立場演化論——沒有對錯，只有生成與解決（籌備中）

© 2026 Reality Check Framework. 專為人機協作共生而設計。