scientific-claim-tightening

Scientific Claim Tightening

Read-only diagnostic workflow for finding and tightening overbroad scientific claims. Do not edit files in this skill.

When To Use

Use when a passage contains or may contain words and constructions such as:

• only, first, novel, best, general, scalable, efficient, reusable
• outperforms, demonstrates, proves, validates, guarantees
• existing approaches, all approaches, no work, none, always, never
• result summaries in the abstract, introduction, discussion, or conclusion
• reviewer comments about overclaiming, unsupported claims, or scope

Required Inputs

• Target passage and location.
• Immediate context when available.
• Relevant evidence source, or permission to search the manuscript.

Reference Context

Before diagnosing, read these files when relevant:

• .agents/context/claim-ledger.md
• .agents/context/do-not-say.md
• .agents/context/manuscript-context.md
• .agents/context/contributions.md
• .agents/context/evaluation-claims.md
• .agents/context/terminology.md

Use manuscript files as the current source of truth. For implementation-dependent claims, invoke .agents/agents/paper-implementation-checker.md.

Workflow

1. Identify every high-risk claim in the target passage.
2. For each claim, classify risk:
   • Scope risk: claim broader than evaluated scenarios or reviewed literature.
   • Evidence risk: claim not supported by cited results or manuscript context.
   • Causality risk: causal wording where experiments show only association.
   • Terminology risk: ambiguous or drifting technical terms.
   • Implementation risk: claim depends on code behavior and needs verification.
3. Compare each claim against the claim ledger and do-not-say list.
4. Report the supported scope and a direction for safer wording.
5. Do not write a full replacement paragraph unless the user explicitly asks; use /write-to-intent, /technical-paragraph-rewrite, or /address-reviewer-comment for drafting.