noahnan-max/governed-dcf-skill/dcf-valuation-governance

Governed DCF Valuation

Treat valuation as an auditable research system, not a one-cell formula.

中文介绍

这个 Skill 的目标不是快速给出一个“目标价”，而是把估值变成一套任何关键数字都能追溯、复算、质疑和更新的研究系统。

• 适合谁：需要搭建或审计 DCF、复核卖方估值、解释市场隐含预期，或建立标准化估值工作簿的研究者。
• 解决什么：避免自由现金流凭空输入、WACC 与永续增长率失真、终值占比过高、股权桥遗漏和公式错误。
• 标准产出：方法路由、三情景预测、假设档案、FCFF 估值桥、敏感性矩阵、反向 DCF、独立重算和质量门状态。
• 核心价值：结果必须落在 PASS / DRAFT_REVIEW / BLOCKED 之一；缺失数据保持缺失，不用零值或乐观假设掩盖问题。

Non-negotiable output state

Return one of:

• PASS: required inputs, explicit assumptions, model checks, sensitivity, and independent recalculation all pass.
• DRAFT_REVIEW: the model runs, but one or more evidence or review gates remain open.
• BLOCKED: required facts or structural inputs are missing; do not fill them with invented values.

Never turn missing data into zero. Never describe a DRAFT_REVIEW model as final.

Workflow

1. Route the valuation family. Read references/valuation-routing.md.

• Ordinary operating company: FCFF DCF.
• Bank, insurer, or broker: FCFE, DDM, residual income, or fair-PB.
• Property, resources, or strong segment economics: NAV or SOTP.

1. Freeze the valuation date and units.

• State valuation date, reporting cut-off, model currency, per-share currency, diluted shares, and bridge items.

1. Build the evidence and assumption dossiers.

• Historical facts must trace to primary disclosures.
• Separate fact, calculation, and judgment.
• Cover revenue, margin, reinvestment, risk/terminal value, and equity bridge.
• Give every material assumption a base view, bear/base/bull range, evidence, counter-evidence, and falsifier.

1. Build integrated operating forecasts before valuation.

• Forecast business drivers and complete statements or equivalent schedules.
• Derive FCFF from operations; do not type a free-standing cash-flow series with no bridge.

1. Run three scenarios.

• Probabilities must be explicit and sum to 100%.
• Each scenario must carry its own WACC, terminal growth, and forecast path.

1. Run valuation checks.

• Require WACC > terminal growth.
• Reconcile enterprise value to equity value and diluted per-share value.
• Report terminal-value share of enterprise value.
• Produce WACC × terminal-growth sensitivity and reverse DCF.

1. Apply workbook and governance gates.

• Read references/workbook-contract.md and references/governance-gates.md.
• Recalculate independently, scan formula errors, and bind outputs to a run manifest and hashes.

1. Communicate uncertainty.

• Present a range and the variables that change it.
• State evidence gaps, falsifiers, and what would move the valuation.
• Add “educational analysis, not investment advice.”

Public reference CLI

The bundled CLI runs a synthetic or user-provided three-scenario FCFF case. It is a transparent reference implementation, not a substitute for an integrated production workbook.

python scripts/dcf_cli.py validate --input ../../../examples/synthetic-consumer-case.json
python scripts/dcf_cli.py run --input ../../../examples/synthetic-consumer-case.json
python scripts/dcf_cli.py reverse \
  --input ../../../examples/synthetic-consumer-case.json \
  --scenario base \
  --target-price 42

The run output must include:

• enterprise value, equity value, and value per diluted share;
• explicit-period PV and terminal-value PV;
• terminal-value share;
• probability-weighted expected value;
• base-case WACC × terminal-growth sensitivity;
• warnings and a machine-readable status.

Required answer structure

1. Conclusion and status — valuation range, probability-weighted value, PASS/DRAFT_REVIEW/BLOCKED.
2. Method routing — why this valuation family fits the company.
3. Source and cut-off — disclosures, dates, units, currency.
4. Operating model — business drivers and statement links.
5. Assumption dossier — bear/base/bull plus evidence and falsifiers.
6. DCF bridge — FCFF, WACC, terminal value, EV-to-equity, per share.
7. Sensitivity and reverse DCF — what the current price implies.
8. Quality gates and limitations — checks passed, open gaps, independent recalculation.

Stop rules

Stop and return BLOCKED when:

• the valuation date, share count, or bridge cannot be established;
• a material assumption has no evidence or explicit judgment label;
• WACC <= terminal growth;
• probabilities do not sum to 100%;
• formula errors or independent recalculation differences remain;
• the method is structurally wrong for the company.

Keep research and execution separate. A valuation result is not a trade instruction.