/research-refine

Use the experiment-report skill to refine an existing report.

Refinement Dialogue Flow

When user requests refinement, engage in structured dialogue:

Step 1: Initial Assessment Questions

Ask the user:

1. "Which report do you want to refine?" (path)
2. "What is your primary concern?" (structure/rigor/clarity/all)
3. "Is this for internal review or external sharing?"

Step 2: Focus Area Selection

Based on user response, prioritize one of:

Structure Focus:

• Executive Summary length and completeness
• Logical flow between sections
• Redundancy elimination

Scientific Rigor Focus:

• Claim-evidence alignment
• Fact/interpretation separation
• Confidence language calibration

Clarity Focus:

• Conciseness
• Terminology consistency
• Figure references

Step 3: Claim-Evidence Audit

For each finding, verify:

• Claim is stated clearly
• Evidence table is complete (notebook path, figure path, statistics)
• All paths exist and are accessible
• Uncertainty level is stated
• Alternative interpretations are considered

Questions to identify gaps:

• "For Finding X, what is the primary evidence?"
• "Is there a figure that supports this claim?"
• "What confidence level is appropriate here?"
• "What alternative explanations should we mention?"

Step 4: Confidence Language Calibration

Check each claim against evidence strength:

Evidence Strength	Appropriate Language
Very strong (multiple lines)	"demonstrate", "show", "establish"
Strong (validated)	"indicate", "suggest strongly"
Moderate (single line)	"suggest", "consistent with"
Weak (preliminary)	"may", "could", "possible"
Speculative	"we hypothesize", "one possibility"

Ask user:

• "Are there any claims that feel overconfident?"
• "Are there any claims that are too hedged?"

Step 5: Quality Gate Verification

Apply checklist from references/refinement-guide.md:

Structure:

• Executive Summary: 3-5 sentences
• All sections present
• No redundancy

Evidence:

• All figure paths verified
• All notebook paths verified
• Statistics include test, p-value, effect size

Rigor:

• Facts separated from interpretation
• Confidence language calibrated
• Alternative explanations considered
• Limitations acknowledged

Writing:

• Concise and clear
• Consistent terminology
• Figures referenced in text

Step 6: Implementation

After identifying issues:

1. Prioritize by severity (Critical > Important > Minor)
2. Present proposed changes to user
3. Implement with user approval
4. Verify improvements

Minimal Refinement (Quick Pass)

If user wants fast refinement, focus on these 3 critical checks:

1. Claim-Evidence Links: Is every claim backed by evidence?
2. Fact/Interpretation Separation: Are observations in Findings, interpretations in Synthesis?
3. Confidence Calibration: Does language match evidence strength?

See references/refinement-guide.md for detailed criteria.