brian-ren-pro/econ-empirical-paper-skills/top-journal-style-benchmark

Top Journal Style Benchmark

Overview

Use this skill to turn a local manuscript revision into a benchmarked top-journal writing pass. The workflow compares the manuscript against 1-3 close exemplar papers, diagnoses section-level language and structure gaps, revises a copy, and validates whether the revision landed.

Core Rules

• Work from the latest source manuscript, not rendered PDF/DOCX unless no source exists.
• Never overwrite the user's current manuscript unless explicitly asked; create a clearly named revision copy.
• Prefer local PDFs, Zotero files, and project literature manifests before web search.
• Do not export or reproduce full copyrighted article text. Use local extraction for analysis, but report only short quotes, paragraph roles, and paraphrased structure.
• Match exemplars by economic object and writing job, not just keyword overlap.
• Keep empirical claims fixed unless the user explicitly asks to change results.
• Run an independent validation pass with a subagent when the user requests it or when the revision changes a high-stakes section.

Workflow

1. Lock The Section And Source

Identify:

• source manuscript path;
• target section and boundaries;
• current output format needed by the user;
• whether the user allows direct edit or requires a copy;
• relevant project rules for routing drafts and outputs.

If the section is embedded in a larger draft, extract only the target section for analysis, but make revisions in a full-copy manuscript so numbering and context remain intact.

2. Select Benchmark Articles

Choose 1-3 benchmark articles using this hierarchy:

1. Same identification setting or institutional shock.
2. Same economic mechanism or market.
3. Same data source and empirical object.
4. Same journal style or field audience.

For each benchmark, record:

• citation, journal, and year;
• local PDF path or manifest source;
• why it is close;
• which sections are useful for comparison.

If no close local benchmark exists, search locally first with rg/find; browse only when local sources are insufficient or the user asks.

3. Extract Structure, Not Full Text

Use PDF/text extraction to inspect benchmark sections. Save or report:

• section headings;
• paragraph sequence and rhetorical job;
• short phrase-level examples when necessary;
• what the benchmark does not include.

Avoid copying full paragraphs from benchmark articles into new files or responses.

4. Compare Paragraph By Paragraph

For each manuscript paragraph or sentence group, compare:

• rhetorical job;
• claim specificity;
• mechanism continuity;
• evidence density;
• causal language;
• journal-style concision;
• whether it sounds like a result list, literature list, or process note.

Use references/section_benchmark_rubric.md when a detailed checklist is needed.

5. Revise A Copy

Create a new revision copy with a suffix that states the pass, such as:

• P3-abstract-top-journal
• P4-data-benchmark
• P5-introduction-jf-style

Revision principles:

• Preserve facts, coefficients, table numbers, and empirical scope.
• Replace process language with manuscript language.
• Prefer one continuous mechanism over many disconnected result claims.
• Put limitations in the right place; do not let the abstract or opening paragraph become defensive.
• If revising DOCX output too, regenerate from the revised source and verify archive validity.

6. Validate

Before calling the work complete:

• diff the revised section against the prior version;
• check table/figure numbering if the full manuscript was regenerated;
• search for process traces such as script, screened, filtered, audit, sample-selection, and outdated data-source names;
• verify DOCX/PDF outputs if generated;
• run a subagent review when requested.

Subagent prompt pattern:

Read the revised section at <path>. Compare it against these benchmark papers and the stated revision goal. Do not edit files. Score whether the revision landed, identify inconsistencies or overclaims, and give at most 3 required fixes.

Output Pattern

For a normal pass, produce:

• benchmark selection memo;
• section-by-section comparison;
• revised manuscript copy;
• optional DOCX/PDF export;
• validation notes;
• independent review summary if a subagent was used.

Keep user-facing summaries concise; link the files.