Scientific Clarity Checker

Scientific Clarity Checker

Table of Contents

• Core Principles
• Workflow
• Analysis Frameworks
• Common Issues
• Guardrails
• Quick Reference

Core Principles

1. Claims must match evidence: Every conclusion needs explicit support

2. Precision over vagueness: Quantify wherever possible

3. Hedging matches certainty: Strong claims need strong evidence

4. Logic must flow: Arguments should be traceable step by step

5. Terminology must be consistent: Same concept = same word

6. Mechanistic clarity: The "how" should be explained, not just "what"

Workflow

Copy this checklist and track your progress:

Clarity Check Progress:
- [ ] Step 1: Identify core claims and hypotheses
- [ ] Step 2: Structural logic review (argument flow)
- [ ] Step 3: Claims-evidence audit
- [ ] Step 4: Quantitative precision check
- [ ] Step 5: Terminology consistency audit
- [ ] Step 6: Hedging calibration
- [ ] Step 7: Mechanistic clarity check

Step 1: Identify Core Claims

List all major claims, conclusions, and hypotheses in the document. These are what the author wants readers to believe after reading. Every claim needs to be evaluated. See resources/methodology.md for claim extraction.

Step 2: Structural Logic Review

Map the argument structure: What premises lead to what conclusions? Are all logical steps explicit? Are there gaps in the reasoning chain? See resources/methodology.md for logic mapping.

Step 3: Claims-Evidence Audit

For each claim: What evidence supports it? Is the evidence presented in this document or only cited? Does the evidence actually support the claim? Flag overclaiming. See resources/template.md for audit format.

Step 4: Quantitative Precision Check

Look for vague quantifiers ("some", "many", "significant increase"). Check for missing statistics, n values, confidence intervals. Flag qualitative descriptions that should be quantitative. See resources/template.md for checklist.

Step 5: Terminology Consistency Audit

Check that terms are used consistently throughout. Verify abbreviations are defined before use. Ensure technical terms are appropriate for audience. See resources/methodology.md for audit process.

Step 6: Hedging Calibration

Match hedge strength to evidence strength. "Demonstrates" needs strong evidence; "suggests" allows weaker evidence. Flag overclaiming (strong words, weak evidence) and underclaiming (weak words, strong evidence). See resources/methodology.md for calibration.

Step 7: Mechanistic Clarity Check

Where explanations of "how" are needed, are they provided? Are mechanisms speculative or evidence-based? Is the level of mechanistic detail appropriate? Validate using resources/evaluators/rubric_clarity.json. Minimum standard: Average score ≥ 3.5.

Analysis Frameworks

Claim-Evidence Chain

For each major claim, trace the chain:

CLAIM: [What the author asserts]
    ↓
EVIDENCE TYPE: [Data/Citation/Logic/Authority]
    ↓
EVIDENCE: [What supports this claim]
    ↓
EVALUATION: [Strong/Moderate/Weak/Missing]
    ↓
ISSUES: [If any - overclaiming, logical gap, etc.]

Logic Flow Assessment

Map argument structure:

PREMISE 1: [Starting assumption or fact]
    +
PREMISE 2: [Additional assumption or fact]
    ↓
INFERENCE: [Logical step taken]
    ↓
CONCLUSION: [What follows from inference]
    ↓
VALIDITY CHECK: [Does conclusion follow from premises?]

Common logical issues:

• Gap: Missing premise needed for conclusion
• Leap: Conclusion doesn't follow from premises
• Assumption: Unstated premise that may not hold
• Circularity: Conclusion assumed in premise

Quantitative Precision Matrix

Hedging Calibration Scale

Common Issues

Overclaiming

Pattern: Strong conclusion words with weak evidence

Examples:

• ❌ "This proves X" (based on correlation)
• ❌ "We have demonstrated Y" (single experiment, no mechanism)
• ❌ "This establishes Z" (preliminary data only)

Fix: Match hedge strength to evidence or add qualifying statements

Logical Gaps

Pattern: Conclusion requires unstated premise

Examples:

• "Protein X is elevated in disease Y; therefore, X causes Y" (missing: causation ≠ correlation)
• "Our model predicts Z; therefore, Z is true" (missing: model validation)

Fix: Make implicit premises explicit or acknowledge limitations

Vague Quantification

Pattern: Qualitative language where numbers exist

Examples:

• "Expression was significantly increased" (what p-value? what fold-change?)
• "Most patients improved" (what percentage?)
• "The treatment worked well" (by what metric?)

Fix: Replace with specific numbers

Terminology Drift

Pattern: Same concept, different words (or vice versa)

Examples:

• Alternating "subjects", "participants", "patients" for same group
• Using "expression" and "levels" interchangeably
• Abbreviation used before definition

Fix: Standardize terminology; create consistency table

Missing Mechanism

Pattern: "What" without "how"

Examples:

• "Treatment X reduces disease Y" (how does it work?)
• "Mutation Z causes phenotype W" (through what pathway?)

Fix: Add mechanistic explanation or acknowledge it's unknown

Guardrails

Key requirements:

1. Point out gaps, not fabricate support: Identify missing evidence without inventing it
2. Preserve author intent: Flag issues without rewriting meaning
3. Audience-appropriate: Technical detail depends on target readers
4. Document-appropriate: Standards differ for abstracts vs. full papers
5. Constructive feedback: Identify problems with suggestions for improvement

What this skill does NOT do:

• ❌ Check factual accuracy of citations (can't verify papers)
• ❌ Assess experimental design quality (would need methods expertise)
• ❌ Verify statistical analysis (specialized skill)
• ❌ Judge scientific importance (subjective)

Focus areas:

• ✅ Internal logic and consistency
• ✅ Claims vs. evidence alignment
• ✅ Clarity and precision of language
• ✅ Appropriate hedging
• ✅ Terminology consistency
• ✅ Argument structure

Quick Reference

Key resources:

• resources/methodology.md: Claim identification, argument mapping, terminology audit, hedging guide
• resources/template.md: Claims-evidence matrix, precision checklist
• resources/evaluators/rubric_clarity.json: Quality scoring

Quick checks:

• Can I identify every major claim?
• Does each claim have explicit evidence?
• Are there logical gaps in the argument?
• Are numbers used instead of vague quantifiers?
• Is terminology consistent throughout?
• Does hedge strength match evidence strength?
• Are mechanisms explained where needed?

Red flags to look for:

• "This proves/demonstrates/establishes" + weak evidence
• "Significant" without p-values
• Conclusions that don't follow from premises
• Same concept with multiple names
• "How" questions left unanswered

Time estimates:

• Quick scan (major issues): 10-15 minutes
• Standard review (full checklist): 30-45 minutes
• Deep analysis (comprehensive audit): 1-2 hours

Inputs required:

• Scientific document (any type)
• Context (audience, purpose)
• Specific concerns (if any)

Outputs produced:

• Annotated document with issues flagged
• Summary of clarity issues by category
• Recommendations for improvement