claim-ledger

Claim Ledger

When to use

Use this skill when the user expresses the following intents:

• "First build a research framework/evidence chain, then write the content"
• "Determine whether a viewpoint is publishable and whether it holds up"
• "Organize a pile of materials into publishable conclusions"
• "Have the AI produce the Ledger first, then the thread/long-form article"

Core definition

The Claim Ledger breaks each viewpoint into three parts:

• Claim: the claim you want readers to believe
• Evidence: the evidence supporting that claim
• Boundary: the conditions under which it holds and the conditions under which it fails

Goal: make every conclusion traceable, refutable, and reviewable.

Minimum fields (9 required items)

One row per Claim, and it must include:

1. Claim ID：C01/C02/...
2. Claim: a testable statement
3. Type：事实/解释/预测/建议
4. Confidence：高/中/低
5. Evidence：E01,E02...(at least 2)
6. Source Grade：S/A/B/⚠️
7. Counter: at least 1 piece of counterevidence/counterexample; if none, write "not found, risk = high"
8. Boundary: conditions under which it holds + conditions under which it fails
9. Publishability：可发/需改写/不可发

Evidence Card specification

Each piece of evidence must be a reviewable card, not a generalized description. Fields:

• Source: title + author/institution + date
• Link: URL or document location
• Quote/Data: key original wording or numbers
• Supports: which part of the Claim it supports
• Reliability：S/A/B/⚠️ + rating rationale
• Freshness：是/否(whether outdated)

Rules of thumb:

• Factual Claim: at least 1 S or A
• Explanatory Claim: at least 2 cross-source corroborations
• Predictive/advisory Claim: must have historical comparison or failure conditions, otherwise downgrade to low confidence

Six-step process (from material pile to publishable)

1. First write 5 candidate Claims (first define what is to be proven)
2. Label each Claim with its type (fact/explanation/prediction/advice)
3. Add 2-4 pieces of evidence to each Claim (prefer primary sources)
4. Force the addition of counterevidence/counterexamples (if none, record that explicitly)
5. Write the boundaries (conditions under which it holds and conditions under which it fails)
6. Assign confidence and publishability (downgrade wording if necessary)

Quality gate (must be executed)

If any of the following is not satisfied, entry into content generation is prohibited:

• Any Claim has fewer than 2 pieces of evidence
• Any Claim lacks a Counter
• Any Claim lacks a Boundary
• 预测/建议 Claim has no failure conditions

Handling rules:

• Insufficient evidence: downgrade to "guess (low confidence)"
• Tone requirement: do not use certainty wording

Mapping to content production

Thread mapping (recommended)

• 3 个 Claim × 4 条推 = 12 条线程
• Each group of 4 in order:
  1. Claim
  2. Evidence
  3. Boundary
  4. Action

Long-form mapping (recommended)

Each Claim as one section:

• Conclusion -> Evidence -> Opposing risk -> Boundary -> Summary/checklist

Default output format

1. Claim Ledger table (at least 3 Claims)
2. Evidence Cards(aligned with the Ledger's E#)
3. Gate Check(list pass/fail items one by one)
4. Only if all Gates pass, generate the target content (thread/long-form article/script)

Execution instruction template

先输出 Claim Ledger（至少 3 条 Claim，每条 >=2 条证据 + >=1 条反证 + 边界）。
未满足质量闸门前，不允许生成正文。
若证据不足，必须降级为“猜测（低置信度）”，并避免确定性语气。

Reference templates and examples

When you need to directly apply tables and cards, read:

• references/templates.md

When you need to automatically verify the Ledger quality gate, run:

• scripts/validate_ledger.py --file <ledger.md>