open world research audit

Research reports and strategy memos often suffer from logical leaps, hidden assumptions, and unverified claims. This skill provides a rigorous audit framework to stress-test existing documents against supplied evidence, separating what is proven from what is merely inferred.

What it does

• Claim extraction identifies decision-bearing statements and maps them to specific report locators.
• Evidence mapping traces claims to supplied source documents while flagging unverified assertions.
• Support classification rates each claim as supported, partially supported, underdetermined, or unsupported.
• Assumption exposure surface hidden definitions or numeric probability bands that lack a statistical basis.
• Decision boundary analysis determines the specific conditions under which a recommendation should change.

How it works

1. Scope freeze records the report version, available evidence, and the user's intended decision or thesis.
2. Materiality audit extracts 5 to 8 pivotal claims and tests their factual dependencies separately from risk preferences.
3. Verification assigns source IDs to evidence and tests for entailment, contradictions, and jurisdictional relevance.
4. Calibration produces a decision boundary report highlighting permitted actions, missing evidence, and reversal conditions.

Frameworks & tools

This tool uses a structured audit protocol incorporating a failure taxonomy and evidence rubric. It processes Markdown, PDF excerpts, and text-based memos.

Why this beats prompting it yourself

Standard prompting often results in the AI agreeing with the report's tone or hallucinating supporting facts. This skill enforces a strict evidence-to-claim mapping protocol that prevents silent browsing and requires explicit locators for every judgment.

Use cases

• Strategy memo review to find logical gaps before a high-stakes board meeting.
• Market analysis stress-test to identify which trends are backed by data versus expert opinion.
• Due diligence support for auditing claims made in investment pitches against provided data rooms.
• Deep Research verification to ensure AI-generated reports haven't drifted from their primary sources.

Known limitations

This skill does not perform research from scratch or provide legal, medical, or investment advice. It requires an existing report and is limited to the evidence provided in the context window.