English | 中文 | 日本語

AI Berkshire — Value Investing Research Framework for the AI Era

"Price is what you pay, value is what you get." — Warren Buffett

Redefining the depth and efficiency of investment research with AI.

AI Berkshire is a collection of investment research skills compatible with both Claude Code and Codex. It systematizes the methodologies of four value investing masters — Buffett, Munger, Duan Yongping, and Li Lu — and delivers professional-grade research through AI Agents.

One person + Claude Code / Codex = an entire investment research team.

Track Record · Why Not Just Ask AI? · Skills · Quick Start · Reports · Design Philosophy

---

Real Track Record

Not paper trading. This framework is backed by a real-money, audited portfolio.

2024 Full-Year Return: +69.29%

2025 Full-Year Return: +66.38%

Benchmark Comparison

Benchmark	2024 Full Year	2025 Full Year
This Framework (Live)	+69.29%	+66.38%
Hang Seng Index	+17.67%	+27.77%
S&P 500	+23.31%	+16.39%
CSI 300	+14.68%	+17.66%
NASDAQ Composite	+28.64%	+20.36%

2024 Alpha: Beat the S&P 500 by 46 percentage points, beat the Hang Seng by 52 percentage points

2025 Alpha: Beat the S&P 500 by 50 percentage points, beat the Hang Seng by 39 percentage points

Cumulative live returns exceed ¥1.46 million over two years, significantly outperforming all major global indices for two consecutive years.

Disclaimer: Past performance does not guarantee future results. Screenshots are from a real brokerage account (Futu Securities).

---

Why Can't You Just Ask AI Directly?

You can, of course, ask Claude: "Should I buy Pinduoduo?" You'll get a balanced "on one hand... on the other hand..." analysis that ends with "investing involves risks, please make your own judgment."

That kind of analysis looks right but can't drive actual decisions.

AI Berkshire doesn't solve the "can AI analyze?" problem — it solves the analysis quality and decision discipline problem. Here's what's different:

1. Forces a Verdict — No Fence-Sitting

Ask AI directly, and you get a both-sides-pleasing "analysis." AI Berkshire forces concrete output: Pass / Fail / Gray Zone, with specific price ranges and tiered recommendations.

Vanilla AI response: "Pinduoduo has growth potential but also faces competitive pressure. Investors should weigh..."

AI Berkshire output:

Strategy	Recommendation	Price Range
Aggressive	Build 20% position at current price	$95–105
Moderate	Wait for buyback policy clarity	$85–95
Conservative	Doesn't meet 10-year certainty bar — pass	—

Mirror Test: If you can't articulate it in 5 sentences = don't buy. No exceptions.

2. Four-Master Dialectic, Not a Single Perspective

It's not just "analyze this using Buffett's method." The four perspectives create real tension and contradictions —

Take Pinduoduo as an example:

• Duan Yongping (business model): Great business, C2M model hard to replicate → 3.7/5
• Buffett (financial valuation): Ex-cash P/E just 6.3x, a cash machine → 4.4/5
• Munger (inversion): Moat shallower than it appears — Douyin hit ¥4 trillion GMV in 3 years → 3.5/5
• Li Lu (long-term certainty): Management culture concerns, uncertain in 10 years → 2.0/5

Buffett says "genuinely cheap," Li Lu says "if uncertain, don't buy" — this conflict is the real state of investment decisions. A single prompt can't produce this multi-perspective dialectic, yet it's precisely what prevents blind spots.

3. Structured Anti-Bias Mechanisms

AI's greatest danger isn't giving wrong answers — it's giving answers that look right but don't withstand scrutiny. AI Berkshire embeds multiple "anti-deception" layers into the process:

Mechanism	Problem Solved	Example
Information Richness Rating (A/B/C)	Prevents "more data = more certainty" illusion	Pop Mart rated B: limited data, estimated metrics flagged with confidence levels
Munger-Style Inversion Test	Forces thinking about failure scenarios	"How could Pinduoduo die?" → Lists 5 scenarios with probabilities
Quick-Kill Checklist	8 red lines, any one is a veto	Management integrity issues → immediate rejection regardless of valuation
Contrarian Check	Avoids thinking like the crowd	"Why are smart people shorting this?" → Surfaces overlooked risks
Intellectual Honesty	Prefer "I don't know"	Marks data gaps as "gray zone" rather than filling certainty with speculation

4. Financial Data Precision

LLMs can't do mental math reliably. Getting a P/E wrong by one decimal point or confusing HKD with CNY can lead to catastrophic investment decisions.

Real case: When analyzing Tencent, different sources reported market cap in "HKD billions" and "CNY billions." AI Berkshire's approach:

# Market cap manual verification: Price × Shares Outstanding, cross-checked with reported data
python3 tools/financial_rigor.py verify-market-cap \
  --price 510 --shares 9.11e9 --reported 4.65e12 --currency HKD
# ✅ Verified — deviation only 0.08%

All calculations use Python decimal.Decimal (exact decimal arithmetic), not float. Key data requires at least 2 independent sources for cross-validation.

5. Reproducible Research Process

Ask AI directly, and the format, depth, and coverage vary every time — today's Tencent analysis has a moat score, tomorrow's Meituan analysis might forget it.

AI Berkshire ensures: Same input → structurally consistent, equally deep output. This means you can:

• Compare 7 companies side by side with identical scoring criteria
• Re-analyze the same company 6 months later and directly compare changes
• Align research outputs across team members

Real output — 7 companies screened with the same Checklist:

Company	Verdict	Circle of Competence	Good Business	Moat	Management	Margin of Safety	Overall
Kweichow Moutai	✅ Pass	★★★★★	★★★★★	★★★★★	★★★☆☆	★★★★☆	4.7
Tencent	✅ Pass	★★★★☆	★★★★★	★★★★★	★★★★★	★★★★☆	4.7
NVIDIA	✅ Conditional	★★★★☆	★★★★★	★★★★★	★★★★★	★★★☆☆	4.3
Meituan	✅ Conditional	★★★★☆	★★★★☆	★★★★☆	★★★★☆	★★★★☆	4.0
Kuaishou	✅ Conditional	★★★☆☆	★★★★☆	★★★★☆	★★★★☆	★★★★★	4.0
Pinduoduo	❓ Gray	★★★★☆	★★★★☆	★★★☆☆	★★★☆☆	★★★★★	3.8
Pop Mart	❓ Gray	★★★☆☆	★★★★☆	★★★★☆	★★★★★	★★★☆☆	3.7

6. Multi-Agent Parallelism = Multiplied Research Depth

/investment-team launches 4 independent Agents to research a company simultaneously. Each Agent conducts its own web searches, cross-validates data, and reaches independent conclusions. This isn't splitting one prompt into four sections — it's 4 "analysts" each doing complete research, with a Team Lead synthesizing the final call.

Ask AI directly, and you have one context window. Four parallel Agents means 4× the search volume, 4× the information sources, and 4 independent perspectives.

In One Sentence

Regular users asking AI get "analysis that looks right." With AI Berkshire, you get "research reports you can actually make decisions from."

---

Architecture

Three-Layer Design Philosophy:

• Skill Layer: Abstracts "what you want to do" into 19 clear entry points — deep research, earnings analysis, industry screening, portfolio management, and thinking tools. Pick by scenario.
• Agent Layer: Team skills (e.g. /investment-team, /earnings-team) run 4 master-perspective Agents in parallel under a Team Lead — searching and judging independently, challenging each other before synthesis. Lightweight skills skip this layer and call tools directly.
• Tool Layer: Exact-precision calculations, real-time web search, report auditing — ensures every report's data is rigorous and verifiable.

---

Skills Overview (19 Skills)

🔬 Deep Research

Skill	Purpose	When to Use
/investment-research	Four-master comprehensive analysis	Full-spectrum research on a public company
/investment-team	Multi-Agent parallel research team	4 Agents in parallel — fastest and most comprehensive
/management-deep-dive	Management deep dive	"Buying a stock is buying its people" — when management is the key variable
/private-company-research	Private company research	Research info-scarce private companies like Ant Group, SpaceX
/deep-company-series	8-part long-form deep dive series	Publication-grade series, ~120K words from cognitive reset to decision closure

📊 Earnings Analysis

Skill	Purpose	When to Use
/earnings-review	Earnings deep read (primary sources)	Read raw filings only — no sell-side reports — like Buffett reads annual reports
/earnings-team	Earnings team + publishable article	Four masters interpret earnings in parallel → editor polish → reader review → publish-ready

🏭 Industry Screening

Skill	Purpose	When to Use
/industry-research	Industry value chain scan	Map all investment opportunities across an industry's value chain
/industry-funnel	Industry funnel screening	Full market → rough cut ≤10 → final pick 3, with deep analysis
/quality-screen	Quality screen (7 hard metrics)	Quickly eliminate non-first-class companies; supports single stock / industry / index / thematic batch screening
/bottleneck-hunter	Supply-chain bottleneck hunter	Start from a supertrend and find physical supply-chain bottlenecks and arbitrage opportunities
/investment-checklist	Buffett pre-buy checklist	Six gates, 10-minute decision on whether to dig deeper

📈 Portfolio Management

Skill	Purpose	When to Use
/portfolio-review	Portfolio review & optimization	Graduate from "researching companies" to "managing a portfolio" — sizing, concentration, rebalancing
/thesis-tracker	Investment thesis tracker	Post-buy discipline system: continuously track whether your thesis has been falsified
/thesis-drift	Investment thesis drift detection	Compare two theses/reports — separate factual, valuation, and wording changes
/news-pulse	Price-move rapid attribution	When a stock surges or drops — figure out "what happened" in 10 minutes

🧠 Thinking Tools

Skill	Purpose	When to Use
/dyp-ask	Duan Yongping Q&A	Think through any question the Duan Yongping way — business, investing, life
/financial-data	Financial data retrieval & cross-validation	Ensure key data comes from 2+ independent sources; alerts on >1% deviation
/wechat-article	WeChat article workflow	Author, editor, and reader Agents collaborate to produce a publishable article

---

Quick Start

Cost & Model Selection

Deep-research skills run multiple research passes, cross-source checks, and multi-agent synthesis by design, so they can consume a large number of tokens. That cost is part of getting fuller coverage across business quality, financials, industry structure, and risk.

For high-stakes investment decisions, the maintainer's view is that the strongest model usually offers the best analysis ROI; saving model cost should not come at the expense of important judgment quality. Lighter models can be useful for triage, summarization, or low-risk questions, but moat, valuation, management, and risk synthesis should be expected to depend more heavily on model capability.

To control cost, adjust the workflow before expecting a full deep-research run to become cheap: use /quality-screen first to rule out weaker companies, or /news-pulse for quick price-move attribution. Run /investment-research or /investment-team only when the result is worth deeper work.

1. Install an AI Client

This repository keeps one canonical workflow and provides Claude Code commands plus Codex skills. Install the client you plan to use.

For Claude Code users:

npm install -g @anthropic-ai/claude-code

For Codex users on macOS / Linux:

# macOS / Linux
curl -fsSL https://chatgpt.com/codex/install.sh | sh

# Or use npm
npm install -g @openai/codex

# Or use Homebrew
brew install --cask codex

# Verify installation
codex --version

Windows users can use the official PowerShell installer: powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex".

If codex --version prints a version, you can continue with this project's Codex skills installation.

Reducing Approval Prompts

These skills issue many tool calls, and Claude Code asks for approval for each one by default. That behavior comes from Claude Code's client-side permission system; it is not a repository default this project can change.

If you trust the current workflow and are running in a trusted environment, start Claude Code in skip-permissions mode:

claude --dangerously-skip-permissions

Warning: this disables Claude Code's tool-approval guardrails. Use it only when you trust the repository, commands, and working directory.

2. Install Skills

For Claude Code users on macOS / Linux:

# Clone the repository
git clone https://github.com/xbtlin/ai-berkshire.git

# Copy skills to Claude Code global commands directory
cd ai-berkshire
./scripts/install-claude-commands.sh

For Claude Code users on Windows PowerShell / Command Prompt:

git clone https://github.com/xbtlin/ai-berkshire.git
cd ai-berkshire
.\scripts\install-claude-commands.bat

For Codex users on macOS / Linux:

# Clone the repository
git clone https://github.com/xbtlin/ai-berkshire.git

# Generate and install Codex skills to ~/.codex/skills
cd ai-berkshire
./scripts/install-codex-skills.sh

# Optional: install Codex slash prompts to ~/.codex/prompts
# for a Claude Code-like /investment-research entry point
./scripts/install-codex-prompts.sh

For Codex users on Windows PowerShell / Command Prompt:

git clone https://github.com/xbtlin/ai-berkshire.git
cd ai-berkshire
.\scripts\install-codex-skills.bat

REM Optional: install Codex slash prompts
.\scripts\install-codex-prompts.bat

The repository maintains three entry points: skills/*.md are the Claude Code command sources; codex-skills/*/SKILL.md are Codex skill packages generated from skills/*.md by scripts/sync-codex-skills.py; codex-prompts/*.md are an optional Codex slash-prompt compatibility layer.

3. Use

Invoke directly in Claude Code:

# Deep Research
/investment-research Tencent
/investment-team Meituan
/management-deep-dive Wang Xing, Meituan
/private-company-research SpaceX
/deep-company-series Pinduoduo

# Earnings Analysis
/earnings-review Tencent 2025Q4
/earnings-team PDD 2025 Annual

# Industry Screening
/industry-research Nuclear Power
/industry-funnel AI Compute
/quality-screen Hang Seng Index Constituents
/bottleneck-hunter AI Infrastructure
/investment-checklist Moutai, NVIDIA, Apple

# Portfolio Management
/portfolio-review Tencent 30%, Meituan 20%, Moutai 20%, Cash 30%
/thesis-tracker Pinduoduo
/thesis-drift Pinduoduo reports/PDD-thesis-2025Q4.md reports/PDD-thesis-2026Q1.md
/news-pulse Tencent

# Thinking Tools
/dyp-ask Where is Pinduoduo's real moat?
/wechat-article Meituan

After installing for Codex, restart Codex and refer to skills by name, for example:

Use investment-research to research Tencent
Use earnings-review to analyze PDD 2025 annual results
Use industry-funnel to screen AI compute
Use bottleneck-hunter to scan AI infrastructure bottlenecks
Use thesis-drift to compare two Pinduoduo theses
Use wechat-article to write a Meituan investment article

If you install Codex slash prompts, restart Codex and search for them in the / menu. Codex's official custom prompt entry point usually appears as prompts:<name>, for example:

/prompts:investment-research Tencent

---

Detailed Skill Descriptions

1. /investment-research — Four-Master Comprehensive Analysis

The most thorough single-company deep research framework. Executes seven modules in sequence:

Data Collection → Business Essence (Duan Yongping) → Moat (Buffett) → Inversion (Munger)
    → Management Assessment (Duan Yongping + Buffett) → Civilizational Trends (Li Lu)
    → Valuation & Margin of Safety

Key Features:

• AI research bias awareness mechanism (A/B/C information richness rating)
• Multi-source cross-validation on key data (manual market cap calculation, 2+ independent sources)
• Each master's "follow-up questions" woven throughout
• Three-scenario valuation (bull/base/bear) + reverse DCF

Sample Output Excerpt:

Comprehensive Decision Memo

Dimension	Conclusion	Confidence
Business Quality (Duan Yongping)	Excellent: platform business, two-sided network effects, near-zero marginal cost	★★★★★
Moat (Buffett)	Wide and widening: network effects + switching costs + scale economies, triple-layered	★★★★☆
Management (Duan Yongping + Buffett)	Strong: founder-led, excellent capital allocation discipline	★★★★☆
Top Risk (Munger)	Regulatory policy uncertainty; new business losses dragging overall profits	★★★☆☆
Civilizational Trend (Li Lu)	Aligned with digital consumption trends, but not a "civilization-level paradigm shift"	★★★★☆
Valuation (Buffett + Duan Yongping)	Current P/E 18x, slightly below historical median, modest margin of safety	★★★★☆

Duan Yongping: "The essence of this business is connecting consumers and merchants — profiting from efficiency gains. The hallmark of a great business: more users bring more merchants, more merchants bring more users. Once the flywheel spins, it's very hard to stop."

Munger: "Invert, always invert — if this company vanished tomorrow, what would users and merchants do? If the answer is 'quickly find a substitute,' the moat isn't deep enough. If the answer is 'life would become very inconvenient,' that's worth paying attention to."

---

2. /investment-team — Multi-Agent Research Team

Launches 4 AI Agents in parallel, simulating a real investment research team. Each Agent searches independently, analyzes independently, and delivers independent ratings. The Team Lead synthesizes the final judgment.

Sample Output Excerpt:

One-Line Conclusion

Meituan is the undisputed leader in China's local life services, with multi-layered network effect moats. Current valuation sits at historically low levels — significant long-term value. Recommend accumulating on dips.

Four-Dimension Scorecard

Dimension	Framework	Score	Core Judgment
Business Model & Moat	Duan Yongping	★★★★☆	Strong two-sided network effects; food delivery + in-store form a flywheel
Financials & Valuation	Buffett	★★★★☆	Core business margins improving steadily; valuation at historical lows
Industry & Competition	Munger	★★★☆☆	Douyin invading in-store business; competitive landscape may deteriorate
Risk & Management	Li Lu	★★★★☆	Wang Xing has exceptional strategic vision, but new business cash burn needs monitoring

Composite Score: 3.8 / 5

Investment Recommendation

Strategy	Recommendation	Price Range (HKD)
Aggressive	Build 30% position at current price	120–140
Moderate	Wait for pullback to 100–110 to enter	100–120
Conservative	Wait for quarterly results to confirm margin trend	<100

---

3. /investment-checklist — Buffett Pre-Buy Checklist

Six gates for rapid screening — decide in 10 minutes whether a company is worth deeper research:

Gate 1: Circle of Competence (Can I understand it?)
    ↓ Pass
Gate 2: Good Business (What are the economics?)
    ↓ Pass
Gate 3: Moat (How deep is the competitive advantage?)
    ↓ Pass
Gate 4: Management (Can they be trusted?)
    ↓ Pass
Gate 5: Margin of Safety (Is the price cheap enough?)
    ↓ Pass
Gate 6: Decision Discipline (Rational or FOMO?)
    ↓ Pass
   ✅ Mirror Test

Supports multi-company comparison — screen multiple targets at once:

/investment-checklist Tencent, Alibaba, Meituan, Pinduoduo

Sample Output Excerpt:

Mirror Test

"I am buying Tencent at HK$380 because:

1. The essence of this business is a social network + digital content platform — I understand it;
2. Its moat is 1.2 billion users' social graph, and it's widening;
3. Management — Pony Ma is understated, pragmatic, and an excellent capital allocator — trustworthy;
4. The current price represents ~80% of intrinsic value, providing a meaningful margin of safety;
5. Even if I'm wrong, downside is manageable because net cash exceeds ¥200 billion and gaming cash flow is rock-solid."

✅ Passed the Mirror Test

If you can't articulate it in 5 sentences = don't buy. No exceptions.

---

4. /industry-research — Industry Value Chain Scan

Start from an investment theme and complete a full industry value chain study:

Investment Logic Chain → Value Chain Map → Global Listed Company Scan
    → Four-Master Analysis on Segment Leaders → Portfolio Allocation Recommendation

Sample Output Excerpt:

Investment Logic Chain: Nuclear Power

Underlying Trend: AI data center power demand explosion + carbon neutrality goals → Drives: surging demand for stable, clean baseload power → Creates: deterministic demand for nuclear restarts / new builds / SMRs → Benefits: uranium mining → fuel fabrication → equipment manufacturing → operators

Recommended Portfolio

Tier	Weight	Target	Segment	Core Logic
Core	50%	CGN / Cameco	Operations + Uranium	Highest certainty
Satellite	30%	CNNP / Dongfang Electric	Operations + Equipment	Domestic substitution beneficiary
Option	15%	NuScale / Nano Nuclear	SMR	High risk, high convexity
ETF	Alternative	URA / URNM	Full chain	Passive approach

---

5. /industry-funnel — Industry Funnel Screening

Start from an industry/theme and progressively narrow: Full market → ≤10 → 3 deep dives:

Full Market Scan (activity + returns + top-30 market cap union → 30-60 companies)
    ↓ 5 value investing hard filters
Rough Cut ≤ 10
    ↓ Detailed analysis (300-500 words each)
Detailed Analysis ≤ 10
    ↓ Final selection (by portfolio complementarity, NOT by top-3 score)
Four-Master Deep Analysis on 3 companies (800-1200 words each)
    ↓
Recommended Portfolio (Core / Satellite / Option) + Action Signals

Key Features:

• Every layer has explicit keep/drop criteria — eliminated names come with a stated reason (not a black box)
• Final 3 are selected for portfolio complementarity (high certainty + moderate upside + high convexity), not by ranking scores
• Mandatory "future IPO candidates" list to avoid missing private-market key players
• AI bias awareness: counters large-cap bias / English-language bias / narrative bias / listed-only bias

Difference from /industry-research:

• industry-research emphasizes value chain structure and panoramic view (sliced by segment)
• industry-funnel emphasizes the stock-picking funnel (progressive screening from full market to 3)

Live Test: AI Sector, 4 Sub-Tracks in Parallel (2026-05-09):

Sub-Track	Final 3	Core Position Pick
AI Compute	TSMC / NVIDIA / SK Hynix	TSMC ★★★★★
AI Models	Alphabet / Meta / Alibaba	Alphabet ★★★★★
AI Applications	Microsoft / Adobe / AppLovin	Microsoft + Adobe ★★★★
AI Infrastructure & Power	Eaton / TBEA / Talen Energy	Eaton + TBEA ★★★★

Key Insight: The biggest winners in the AI application layer aren't AI-native companies — they're established giants with distribution, data, and workflow embeddedness. This echoes the 1995–2000 Internet bubble's "sell the picks and shovels" pattern (Amazon and Apple won; Pets.com didn't).

Full reports: AI Compute · AI Models · AI Applications · AI Infrastructure & Power

---

6. /private-company-research — Private Company Deep Research

A "detective-style" research framework designed for information-scarce private companies:

Key Differentiators:

• Financial data piecing: Assembled from IPO filings, parent company reports, funding news, and industry data
• Confidence tagging: Every data point tagged 🟢 High / 🟡 Medium / 🔴 Low confidence
• Multi-method valuation cross-check: Funding-round valuation + comparable companies + DCF + endgame backsolve
• Exit path analysis: Full evaluation of IPO / M&A / secondary transfer paths

Sample Output Excerpt:

Company Snapshot: SpaceX

Item	Detail
Latest Valuation	~$350B (2025 secondary market) 🟡
Estimated Revenue	~$13B (2024) 🟡
Starlink Subscribers	4M+ (end of 2024) 🟢
Launch Cadence	100+ per year (2024) 🟢

Valuation Assessment

Method	Valuation Range	Notes
Latest Funding	$350B	Secondary market price; includes liquidity premium
Comparable Companies	$200–280B	Benchmarked against telecom + aerospace + defense
DCF (Base Case)	$250–350B	Assumes Starlink $30B revenue by 2027
Endgame Backsolve	$400–600B	Assumes Starlink becomes global telecom infrastructure

Composite Fair Value Range: $250B – $400B

---

7. /news-pulse — Price-Move Rapid Attribution

Designed for "when a stock surges or drops, quickly figure out what happened." Not deep research — it's 10–15 minute rapid attribution to avoid panic-selling or essay-length anxiety spirals when your holdings move.

Key Differentiators:

• 4-dimensional parallel recon: Company events / Regulatory policy / Industry competitors / Market sentiment (sell-side + influencers + southbound capital flows)
• Attribution over listing: Doesn't just list all news — judges "which event actually explains this price move"
• Mandatory nature classification: Value Event / Sentiment Fluctuation / True Cause Unknown / Mixed — where "True Cause Unknown" is often the most valuable output (potential insider front-running)
• Clear action items: Whether to trigger deep research, re-examine your thesis, or simply watch

When to Use What:

Scenario	Skill
Complete research (hours)	/investment-team or /investment-research
Earnings deep read	/earnings-review
Long-term thesis tracking	/thesis-tracker
Price move, 10-min attribution	/news-pulse

Sample Output Excerpt (Tencent 4/17–5/01 live test, -10.47% over 14 days):

One-Line Attribution

Approximately 70–80% of this -10.47% drop was driven by fund flows and sentiment (buyback blackout period + southbound selling + sector beta + AI narrative displacement). 20–30% came from deferred digestion of the AI capex doubling announcement — no fundamental deterioration. Sell-side consensus remains Buy. This is a "liquidity + sentiment-driven pullback," not a value event.

Attribution Table

Candidate Explanation	Estimated Contribution	Confidence
Buyback blackout period (structural, pre-5/13 earnings)	-3% to -4%	High
Southbound capital turned net seller on Tencent	-2% to -3%	High
AI narrative stolen by competitors (DeepSeek V4 / Qwen 3.6 / MoonDark 1T)	-1% to -2%	Medium
Sector/macro beta (oil + geopolitics + Fed Warsh hawkish)	-2% to -3%	High
Pre-Q1 earnings de-risking	-1% to -2%	Medium
Fundamental deterioration	0%	Very High (ruled out)

Nature Classification: ✅ Mixed

70% fund flows / sentiment + 20% long-term AI narrative concern + 10% pre-Q1 uncertainty

Key counter-evidence: Duan Yongping sold Tencent puts on 4/8 (bullish); 24 sell-side analysts consensus Strong Buy; NetEase rose 2% on 4/30 against the tide (rules out gaming industry issue); Tencent underperformed Hang Seng Tech by 7pp (Hang Seng Tech actually rose 4% for the month).

Usage:

/news-pulse Tencent
/news-pulse Pinduoduo down 12% within a week
/news-pulse miHoYo

---

Live Research Reports

Below are real investment research reports generated with this framework, showcasing actual AI-powered research output quality.

Company	Skill Used	Core Conclusion	Report
Pinduoduo (PDD)	/investment-team	Composite 3.4/5 — extremely cheap but 10-year certainty insufficient; suitable for moderate position	View Report
Tencent (0700.HK)	/investment-research	Social monopoly + superior capital allocation; 14x forward P/E is reasonable-to-low	View Report
7-Company Comparison	/investment-checklist	Moutai & Tencent pass; NVIDIA, Meituan & Kuaishou conditional; Pinduoduo & Pop Mart gray zone	View Report
Master Holdings Tracker	Custom Research	Buffett / Li Lu / Duan Yongping latest 13F holdings + PDD cost-basis analysis	View Report

More reports will be added continuously. PRs submitting your own research reports generated with this framework are welcome.

---

Design Philosophy

Four-Master Methodology Synthesis

Duan Yongping · "The Right Business" — business essence, the shared starting point for the other three lenses:

Buffett	Munger	Li Lu
Moat Margin of safety Management	Inversion Risk list Bias audit	Civilization trends Paradigm shifts Industry value

The four masters aren't just dividing labor — they're designed to challenge each other:

• Duan Yongping says "great business" → Munger asks "how could it die?"
• Buffett says "cheap enough" → Li Lu asks "will it still exist in 10 years?"
• What you get isn't four reports stitched together — it's four thinking systems colliding

Financial Rigor Tool (tools/financial_rigor.py)

Feature	Command	Problem Solved
Market Cap Verification	verify-market-cap	Price × shares outstanding, exact calculation, detects unit errors
Valuation Verification	verify-valuation	P/E / P/B / ROE / FCF Yield — exact decimal arithmetic
Multi-Source Cross-Validation	cross-validate	Auto-compare same data point across N sources; alerts above tolerance
Three-Scenario Valuation	three-scenario	Bull / base / bear exact target price calculation
Benford's Law Detection	benford	Detect anomalies in first-digit distribution of financial data
Precision Calculator	calc	Any financial expression computed exactly — replaces LLM mental math

Design Principle: All calculations use Python decimal.Decimal (exact decimal), not float (floating-point approximation). 0.1 + 0.2 = 0.3 must never fail in a financial context.

---

Future Directions

• Historical backtesting: AI research reports vs. actual stock price performance
• Macroeconomic cycle analysis framework
• Real-time data feeds via MCP (Wind / Bloomberg / Yahoo Finance)

---

Disclaimer

This project is for educational and research purposes only and does not constitute investment advice. Investing involves risk; decisions should be made with caution. Always do your own due diligence (DYOR).

---

License

MIT License

---

"The best investment you can make is in yourself." — Warren Buffett

AI Berkshire: Giving everyone their own investment research team.

Star History

If this project has been helpful to you, please give it a Star!