Idea Evaluation Pipeline — Data skill for Claude Code

Research Idea Evaluation Pipeline

A structured pipeline for iteratively evaluating, pivoting, and defending research ideas until they reach top-3 finance journal quality (JF, JFE, RFS) or top-5 economics journal quality.

Designed for PhD students in finance and economics. Works with any AI coding assistant (Claude Code, Cursor, Codex, Windsurf, etc.) or manually by copy-pasting prompts.

---

How It Works

Your idea goes through up to 8 steps. The pipeline loops until the idea scores 7/10 or higher:

1. EVALUATE IDEA  →  2. REVIEW EVALUATION
                          │
                     Critique unfair? → Re-run Step 1
                          │
                     Score >= 7? ─── Yes ──→ 5. LITERATURE REVIEW
                          │                        ↓
                          No               6. VERIFY LIT REVIEW
                          ↓                        ↓
                     3. PIVOT IDEA          7. FINAL VERDICT
                          ↓                        ↓
                     4. EVALUATE PIVOT      8. REVIEW FINAL VERDICT
                          │                        │
                     Score < 7? → Back to 3   Score >= 7? → DONE
                     Score >= 7 → Step 5      Score < 7  → Back to 3

---

Quick Start

1. Create your idea folder

mkdir my_idea

2. Add your idea file

Copy `idea_template.txt` into your folder and fill it in:

cp idea_template.txt my_idea/idea.txt

The template requires:

• Research question and hypothesis with expected sign
• Identification strategy (shock, instrument, or natural experiment)
• Specific data sources with variable names and sample periods
• 3 closest papers with full citations, URLs, and how your idea differs
• Proposed regression equation (if possible)

**The 3 closest papers are critical.** Pick the papers a referee would immediately cite against you — not broadly related work, but