# How the `/thesis-tracker` Skill Enforces Post-Buying Discipline for Investment Thesis Validation

> Learn how the /thesis-tracker skill enforces post-buying discipline. Convert research into living documents for objective actions and hypothesis validation.

- Repository: [Xbt Lin/ai-berkshire](https://github.com/xbtlin/ai-berkshire)
- Tags: how-to-guide
- Published: 2026-06-26

---

**The `/thesis-tracker` skill enforces post-buying discipline by converting static research notes into a structured, quarterly-reviewed living document that triggers objective actions based on pre-defined red-lines and hypothesis validation.**

Investing successfully requires more than picking winners; it requires the discipline to sell when the thesis breaks. The `/thesis-tracker` skill in the `xbtlin/ai-berkshire` repository eliminates emotional decision-making by systematizing post-buying discipline for investment thesis validation through a rigorously structured workflow defined in [`skills/thesis-tracker.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/thesis-tracker.md).

## From Static Notes to Living Documents

The skill operates on the principle that an investment thesis must be a **testable hypothesis** rather than a narrative. It enforces discipline through two primary operational modes that transform initial research into an immutable tracking system.

### Mode A: Pre-Defined Sell Conditions

When you invoke `/thesis-tracker` for a new company lacking a `{公司名}-thesis.md` file, the skill enters **Mode A** (建立投资论文). This guided workflow forces you to define concrete exit criteria before committing capital, including:

- A **5-sentence core thesis** that fits inside `reports/{公司名}-thesis.md`
- A **hypothesis table** (核心假设清单) pairing each assumption with verification methods and quarterly review frequencies
- A **red-line list** (红线清单) of deal-breaker conditions
- **Valuation anchors** validated automatically by calling `tools/financial_rigor.py verify-valuation`

By requiring these elements upfront—documented in lines 34-95 of the specification—the skill ensures you cannot enter a position without explicit exit criteria, preventing retrospective rationalization.

### Mode B: Quarterly Validation Workflow

Once established, the thesis enters **Mode B** (季度检查), triggering a systematic review process that pulls latest financials and events to update hypothesis statuses. Each hypothesis receives a discrete status marker:

- 🟢 **成立** (Holding)
- 🟡 **弱化** (Weakening)
- 🔴 **受损** (Damaged)
- ⚫ **破裂** (Broken)

This visual tracking system, defined in the **核心假设清单** section at lines 57-67, removes ambiguity about whether your original assumptions still hold, forcing you to confront evidence that contradicts your position.

## Red-Line Guardrails and Forced Actions

The most critical discipline mechanism is the **red-line table** (红线清单) located at lines 71-80 of [`skills/thesis-tracker.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/thesis-tracker.md). Unlike gradual hypothesis degradation, red-lines represent existential threats—such as management fraud or sustained revenue decline—that trigger immediate, predefined actions.

When a red-line condition is met, the skill mandates a concrete recommendation (e.g., **立即清仓** / immediate liquidation), removing the emotional temptation to "wait and see" or "average down" on a broken thesis.

## Quantitative Health Scoring

After checking hypotheses and red-lines, the skill calculates a **论文健康度** (Thesis Health Score) from 0-10 using the weighted formula found at lines 76-86. This score maps directly to disciplined portfolio actions:

| Score Range | Action | Meaning |
|-------------|--------|---------|
| 8-10 | **加仓** | Add to position |
| 6-7 | **持有** | Hold current position |
| 4-5 | **减仓** | Reduce position size |
| 0-3 | **清仓** | Exit completely |

By translating qualitative concerns into a quantitative score, the system forces objective decision-making even when sunk-cost bias pressures you to hold losing positions.

## Immutable Audit Trails

Every review appends a structured record to the thesis file using the **追踪记录表** format at lines 94-102:

```

检查日期 | 健康度 | 核心变化 | 动作建议

```

This append-only log creates an **immutable review trail** that makes it impossible to ignore past warnings or rationalize away deteriorating fundamentals. The historical record lives directly in `reports/{公司名}-thesis.md`, ensuring continuity across quarters and preventing the "what was I thinking?" memory lapses that plague discretionary investors.

## Automation Hooks for Bias Reduction

The skill integrates with [`tools/financial_rigor.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/financial_rigor.py) to automate valuation anchor verification during both initial creation and quarterly checks. By calling `verify-valuation` automatically, the system reduces manual bias in interpreting whether price multiples still offer adequate margin of safety, ensuring that "cheap" is determined by algorithmic checks rather than emotional attachment.

## Practical Usage Examples

Invoke the skill using the following syntax patterns:

```text
/thesis-tracker 腾讯                # First-time creation (Mode A)

/thesis-tracker 腾讯 建立论文      # Force thesis rebuild after major pivot

/thesis-tracker 腾讯 季度检查      # Run quarterly validation (Mode B)

```

A typical quarterly check outputs a structured report like this:

```markdown

# 腾讯 – 投资论文追踪报告（2026-04-09）

一、论文健康度评分：7/10  
二、核心假设检查  
| # | 假设                     | 上次状态 | 最新证据               | 当前状态 | 变化 |

|---|--------------------------|----------|------------------------|----------|------|
| 1 | 收入增速≥15%            | 🟢 成立   | Q4收入增速12%          | 🟡 边际弱化 | ⚠️   |
| 2 | 毛利率≥60%               | 🟢 成立   | 毛利率61.2%            | 🟢 成立   | —    |

三、红线检查  
| # | 红线条件                | 是否触发 | 证据 |

|---|------------------------|:--------:|------|
| 1 | 管理层诚信问题          | ❌ 未触发 | —    |
| 2 | 核心业务连续2季下滑    | ❌ 未触发 | —    |

四、本期关键变化：收入增速放缓至12%，但利润率提升。  
五、估值更新：PE从18×降至16×，安全边际仍在30%上。  
六、结论与行动建议：**持有**，关注收入增速继续走低的风险。  
七、下次检查重点：下一季报的收入增长率与管理层持股变动。  

（记录已追加至 `reports/腾讯-thesis.md`）

```

## Summary

- **Pre-defined exit criteria**: Mode A forces documentation of red-lines and valuation anchors before purchasing, eliminating retrospective rationalization.
- **Quarterly hypothesis validation**: Mode B systematically checks assumptions against fresh data, updating statuses with explicit icons (🟢🟡🔴⚫).
- **Quantitative health scoring**: A 0-10 weighted score translates thesis strength into actionable portfolio decisions (加仓, 持有, 减仓, 清仓).
- **Immutable audit trails**: Append-only logs in `reports/{公司名}-thesis.md` prevent ignoring historical warnings.
- **Automated validation**: Integration with `tools/financial_rigor.py verify-valuation` reduces manual bias in valuation checks.

## Frequently Asked Questions

### What triggers the red-line alert mechanism in the thesis tracker?

Red-line alerts trigger when pre-defined deal-breaker conditions documented in the **红线清单** (lines 71-80 of [`skills/thesis-tracker.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/thesis-tracker.md)) are breached. These include existential threats like management fraud, sustained revenue decline, or business model disruption. When triggered, the skill forces a concrete action recommendation—typically **立即清仓** (immediate liquidation)—removing emotional discretion from the exit decision.

### How does the thesis health score determine portfolio actions?

The **论文健康度** score (0-10) calculated by the weighted formula in [`skills/thesis-tracker.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/thesis-tracker.md) maps directly to disciplined actions: scores of 8-10 recommend **加仓** (adding to the position), 6-7 indicate **持有** (hold), 4-5 suggest **减仓** (reduce), and 0-3 mandate **清仓** (full exit). This quantitative translation prevents subjective justification for holding deteriorating positions.

### Where is the investment thesis data stored and how is it structured?

Thesis data lives in `reports/{公司名}-thesis.md` files within the repository. Each file contains structured sections including the 5-sentence core thesis, a hypothesis verification table (**核心假设清单**), a red-line conditions table (**红线清单**), valuation anchors, and an append-only tracking log (**追踪记录表**) that records every quarterly check with dates, health scores, and action recommendations.

### Can the thesis tracker integrate with external financial data sources?

Yes, the skill calls `tools/financial_rigor.py verify-valuation` to automate valuation anchor verification during both initial thesis creation and quarterly reviews. While the specification references `tools/websearch` for retrieving current prices, financials, and news events, the actual data fetching is implemented through the tool layer, allowing the thesis tracker to validate hypotheses against real-time market data without manual input.