# Core Principles of AI Berkshire Investment Research: The Multi-Master Framework

> Discover the core principles of AI Berkshire investment research. Learn how AI, dual-source verification, and checklist gating create a disciplined, data-driven approach to investment decisions.

- Repository: [Xbt Lin/ai-berkshire](https://github.com/xbtlin/ai-berkshire)
- Tags: deep-dive
- Published: 2026-06-27

---

**AI Berkshire investment research combines the philosophies of Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu into a disciplined, AI-augmented workflow that enforces data integrity through dual-source verification and checklist-gated decision making.**

The `xbtlin/ai-berkshire` repository codifies a systematic approach to value investing that merges traditional Berkshire Hathaway principles with modern AI capabilities. At its core, the framework is defined in structured markdown "skills" that guide autonomous agents through rigorous data collection, multi-dimensional analysis, and disciplined decision gates. The methodology specifically addresses the risks of AI hallucination by embedding human-level skepticism and verification protocols directly into the research pipeline.

## The Four-Master Methodology

The foundation of AI Berkshire investment research rests on a **七个模块** (seven-module) workflow that sequentially applies four investment masters' lenses. As defined in [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md) (lines 7-9), each module filters potential investments through distinct analytical frameworks:

- **Buffett's Moat Analysis**: Evaluating sustainable competitive advantages and economic castle durability.
- **Munger's Inversion**: Applying reverse thinking to identify failure modes and risks before seeking gains.
- **Duan Yongping's "Good Business" Criteria**: Assessing business essence, culture, and long-term durability.
- **Li Lu's Civilizational Trends**: Analyzing macro-level industry shifts and secular growth patterns.

This multi-master approach ensures that no single bias dominates the evaluation, requiring consensus across distinct investment philosophies before capital allocation.

## Data Integrity and Verification Protocols

AI Berkshire treats data quality as a first-class constraint, implementing multiple layers of verification to prevent "LLM-hand-calc" errors.

### AI Bias Awareness and Information Richness

Every analysis begins with an **AI偏见自觉** (AI bias awareness) check that grades the **信息丰富度** (information richness) as A, B, or C level. According to [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md) (lines 13-26), this rating determines inference limits:

- **A-Level**: Ample data available; standard analytical methods apply.
- **B-Level**: Moderate data requiring careful estimation and explicit uncertainty flags.
- **C-Level**: Sparse data demanding strict first-principles questioning and restricted model confidence.

### Dual-Source Verification and Financial Rigor

The framework mandates a **dual-source rule** requiring all financial figures to originate from two independent providers (e.g., macrotrends + stockanalysis for US equities). Any discrepancy exceeding 1% triggers immediate flags per [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md) (lines 35-41).

After data collection, the **`financial_rigor`** tool ([`tools/financial_rigor.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/financial_rigor.py)) programmatically validates market capitalization, valuation ratios, and cash-flow calculations. This automated cross-validation eliminates arithmetic errors before narrative generation begins.

### Automated Report Auditing

Before publication, **[`report_audit.py`](https://github.com/xbtlin/ai-berkshire/blob/main/report_audit.py)** executes a random sampling audit, selecting 15% of numerical claims and re-fetching them from prescribed sources. As specified in [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md) (lines 112-135), reports only clear for release when every sampled datum falls within 1% of the source value.

## The Modular Investment Workflow

The research process flows through six gated analytical phases, each linked to specific master principles:

1. **Business Essence Analysis** (Duan): Defining the fundamental economic model and cultural durability.
2. **Moat Assessment** (Buffett): Quantifying competitive advantages and entry barrier sustainability.
3. **Reverse Thinking & Risk List** (Munger): Compiling inversion-based failure scenarios and negative confirmations.
4. **Management Appraisal** (Duan & Buffett): Evaluating capital allocation discipline and integrity markers.
5. **Industry & Civilizational Trends** (Li Lu): Mapping secular tailwinds and technological disruption risks.
6. **Valuation & Safety Margin** (Buffett & Duan): Calculating intrinsic value and required discount thresholds.

Each phase in [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md) (lines 100-110) builds upon the previous, with hard stops preventing progression if fundamental questions remain unanswered.

To execute this workflow from the command line, the repository provides illustrative skill invocations:

```bash

# Run a full investment-research report for a ticker (e.g., 阿里巴巴)

opencode skills/investment-research "阿里巴巴"

# Run the Buffett-style checklist for multiple tickers

opencode skills/investment-checklist "腾讯, 茅台, NVDA"

```

Both commands internally invoke the markdown skill files and automatically launch the required `Task` agents, data-gathering pipelines, and the `financial_rigor` validation tool.

## Checklist-First Decision Gates

Before deploying deep analytical resources, investments must pass the **巴菲特价值投资买入前 Checklist** defined in [`skills/investment-checklist.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-checklist.md) (lines 44-64). This six-gate system (六关) tests:

- Business clarity ("能说清楚怎么赚钱")
- Economic durability
- Moat depth
- Management integrity
- Price safety margin
- Discipline adherence

Failure on any hard-stop criterion immediately aborts deeper analysis, preventing emotional commitment to marginal opportunities.

### The Mirror Test

The final decision discipline requires the **镜子测试** (mirror test). As codified in [`skills/investment-checklist.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-checklist.md) (lines 75-84), analysts must produce a concise five-sentence justification under 200 words. If the investment thesis cannot withstand this brevity constraint, the opportunity is rejected regardless of quantitative attractiveness.

## Transparency of Confidence

AI Berkshire strictly separates **AI研究置信度** (AI research confidence) from **投资确定性** (investment certainty). The former derives from data richness ratings (A/B/C), while the latter emerges from business fundamentals. This dual-confidence system prevents over-reliance on data-abundant but economically weak companies, and vice versa, as noted in [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md) (lines 7-12).

## Summary

- **AI Berkshire investment research** integrates Buffett, Munger, Duan, and Li Lu's philosophies through a seven-module workflow defined in structured markdown skills.
- **Data integrity** is enforced via dual-source requirements, the `financial_rigor` validation tool, and [`report_audit.py`](https://github.com/xbtlin/ai-berkshire/blob/main/report_audit.py) sampling with 1% tolerance thresholds.
- **Decision discipline** relies on a six-gate checklist and the 200-word mirror test to prevent emotional investing.
- **Information richness ratings** (A/B/C) determine allowable inference levels and AI confidence disclosures.
- **Modular analysis** progresses sequentially from business essence through valuation, with hard stops for unresolved risks.

## Frequently Asked Questions

### What is the Four-Master Methodology in AI Berkshire investment research?

The Four-Master Methodology is a seven-module workflow ([`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md), lines 7-9) that sequentially applies investment principles from Warren Buffett (moat analysis), Charlie Munger (inversion thinking), Duan Yongping (business essence), and Li Lu (civilizational trends). This structure ensures multi-dimensional validation before capital allocation.

### How does AI Berkshire ensure data accuracy in financial reports?

The framework implements a dual-source rule requiring all figures to originate from two independent providers with discrepancies >1% flagged automatically. The `financial_rigor` tool validates calculations programmatically, while [`report_audit.py`](https://github.com/xbtlin/ai-berkshire/blob/main/report_audit.py) performs pre-publication random sampling of 15% of numerical claims to verify they fall within 1% of source values.

### What is the Mirror Test and why is it required?

The Mirror Test (镜子测试) is a final discipline gate requiring a five-sentence, sub-200-word investment justification ([`skills/investment-checklist.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-checklist.md), lines 75-84). If the thesis cannot be expressed this concisely, the investment is rejected. This constraint prevents complexity bias and ensures crystal-clear understanding before capital commitment.

### How does the Information Richness Rating affect investment analysis?

Companies receive A, B, or C ratings based on data availability ([`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md), lines 19-26). A-Level permits standard inference, B-Level requires explicit uncertainty flags and estimation, while C-Level restricts model confidence and demands first-principles questioning. This rating directly constrains the AI's allowable certainty in the final report.