method-audit

name	method-audit
description	Extract and compare data-collection methods across a set of empirical papers. Use when the user needs a cross-paper methods matrix or wants to assess how a literature gathers evidence.
allowed-tools	Read, Write, Edit, Glob, Grep, Bash(uv*), Bash(uv:*), Task, WebSearch, WebFetch, Bash(paperpile*)
argument-hint	[topic, .bib file, or paper directory]
skill-dependencies	["causal-design","split-pdf"]

Method Audit

Reverse-engineer the data collection and empirical methods across a corpus of papers. Produce a critical comparison table that surfaces methodological blind spots.

Output Path

Per rules/review-artefact-routing.md (auto-loads in research projects (path-scoped to paper-*/ and paper/)):

• Source slug: method-audit
• Write reports to: reviews/_project/method-audit/<YYYY-MM-DD-HHMM>.md inside the project. Path is relative to the research project root, not the Task-Management repo.
• Never at project root (./CRITIC-REPORT.md-style filenames are forbidden — pre-rule layout).
• Idempotency: timestamps include hour+minute (HHMM) to disambiguate same-day runs; never overwrite an earlier run's report.
• Index update: if reviews/INDEX.md exists, write a one-line entry under "Latest per source" pointing at the new file. Otherwise review-recap will rebuild the index next time it runs.
• Infrastructure repos (Task-Management, atlas-workspace, etc.): this section does not apply — the path-scoped rule won't load there.

When to Use

• Writing a methodology section — need to justify your approach relative to the literature
• Reviewing empirical papers — need to compare data quality across studies
• Identifying methodological gaps — what approach has nobody tried yet?
• Preparing a replication or extension — need to understand exactly how prior work was done

When NOT to Use

• Theoretical papers — use an installed theoretical-comparison workflow instead
• Single-paper deep read — use split-pdf
• Your own research design — use causal-design or experiment-design
• Code review — use the code-review agent

Input

A .bib file, PDF directory, topic description, or list of papers. If ambiguous, ask.

Workflow

Phase 1: Corpus Assembly

Assemble 10-20 empirical papers from the supplied bibliography or directory, or through configured scholarly-search tools. Prioritise papers with empirical content and filter out pure theory, editorials, and commentaries.

Phase 2: Method Extraction

For each paper, read using split-pdf methodology. Extract:

1. Research design — experimental, quasi-experimental, observational, survey, qualitative, mixed
2. Data source — where the data comes from (name the dataset, survey instrument, or archive)
3. Sample
   • Population and sampling frame
   • Sample size (N)
   • Unit of observation
   • Time period
   • Response rate (if survey)
   • Attrition (if longitudinal)
4. Variables
   • Dependent variable(s) and how measured
   • Key independent variable(s) and how measured
   • Controls included
5. Estimation strategy
   • Statistical method (OLS, IV, DiD, RCT, qualitative coding, etc.)
   • Identification strategy (what makes the estimate causal, if claimed)
   • Robustness checks reported
6. Biases acknowledged — what limitations the authors discuss
7. Biases NOT acknowledged — what you can spot that they don't mention

Phase 3: Comparative Analysis

3.1 Methods Comparison Table

Paper	Design	Data Source	N	Period	Method	ID Strategy	Response Rate

Sort by sample size (largest first).

3.2 Technique Distribution

Count how many papers use each:

• Design type (experimental, observational, etc.)
• Estimation method (OLS, IV, DiD, etc.)
• Data type (survey, admin, experimental, scraped, etc.)

Flag any technique that is dominant (>60% of papers) — this signals a methodological monoculture.

3.3 Bias Audit

For each paper, classify biases:

Paper	Biases Acknowledged	Biases Missed	Severity

Common missed biases to check for:

• Selection bias — non-random sampling without correction
• Measurement error — self-reported outcomes, proxy variables
• External validity — single-country, single-firm, WEIRD samples
• Survivorship bias — studying only firms/people that survived
• Publication bias — significant results overrepresented
• Endogeneity — causal claims without credible identification
• Multiple testing — many outcomes tested without correction

3.4 Methodological Gaps

• Designs nobody has tried (e.g., no RCT in a field of observational studies)
• Data sources nobody has used (e.g., admin data when everyone uses surveys)
• Robustness checks nobody runs (e.g., no placebo tests, no sensitivity analysis)
• Populations understudied (e.g., only US data in a global phenomenon)

Phase 4: Output

Write to METHOD-AUDIT.md in the project directory.

Output Format

# Method Audit: [Topic]

**Date:** YYYY-MM-DD
**Corpus:** [N] empirical papers
**Dominant design:** [Most common research design]
**Dominant method:** [Most common estimation method]

## Comparison Table

| Paper | Design | Data | N | Period | Method | ID Strategy | Biases Noted |
|-------|--------|------|---|--------|--------|-------------|-------------|

## Technique Distribution

| Category | Count | Papers |
|----------|-------|--------|

## Bias Audit

### Commonly Acknowledged
- [Bias type] — mentioned by [N] papers

### Commonly Missed
- [Bias type] — present in [N] papers but acknowledged by [M]
  - **Why it matters:** [Impact on findings]
  - **Papers affected:** [List]

## Methodological Gaps

1. **No [design/method] studies** — [Why this matters]
2. **Understudied population:** [Who is missing]
3. **Missing robustness check:** [What should be tested]

## Implications for Your Research

- **Opportunity:** [What methodological gap you could fill]
- **Risk:** [What bias to watch for in your own design]
- **Benchmark:** [What sample size / design quality is expected in this field]

Cross-References

Skill	When to use instead/alongside
Installed theoretical-comparison workflow	For theoretical rather than methodological comparison
causal-design	To design your own identification strategy
experiment-design	To design experiments or surveys
replication-audit	To check which findings have been replicated