Write Methods Section

Generate or update the methods section based on actual scripts in the pipeline. Links code directly to reproducible documentation.

When to Use

• After completing a script in scripts/

• When methods.md is out of sync with code

• Before writing results section

• As part of documentation review

Prerequisites

• At least one script exists in scripts/

• Scripts have docstrings/documentation

• dvc.yaml reflects the pipeline structure (if using DVC)

Common Mistakes to Avoid

Vague descriptions

• ❌ "Samples were analyzed using standard methods"

• ✅ "Samples were analyzed using high-performance liquid chromatography (HPLC; Agilent 1260 Infinity II) with a C18 column (4.6 × 250 mm, 5 μm particle size)"

Missing statistical details

• ❌ "Data were analyzed statistically"

• ✅ "Group differences were analyzed using two-tailed independent samples t-tests with Welch's correction. Effect sizes were calculated using Cohen's d."

Incomplete software information

• ❌ "Analysis was performed in R"

• ✅ "Analysis was performed in R version 4.2.1 using the lme4 package (version 1.1-30) for mixed-effects models"

Mixing Methods with Results

• Methods: What you planned to do

• Results: What happened

• Keep strictly separate!

Insufficient sample size justification

• Always include power analysis or rationale

• State actual vs. planned sample sizes if different

Execution Steps

1. Gather Context

Read these files:

• .research/project_telos.md

• Aims and methodology notes

• dvc.yaml

• Pipeline stages (if exists)

• params.yaml

• Parameters used

• scripts/*.py or *.R

• All scripts

• manuscript/methods.md

• Existing draft (if any)

2. Analyze Pipeline Structure

Extract from each script:

• Purpose (from docstring)

• Inputs and outputs

• Key parameters

• Dependencies/libraries used

• Algorithms/methods applied

3. Methods Section Structure

Methods

Overview

<!-- Brief summary of analytical approach -->

Data

Data Sources

<!-- Where data comes from, access info -->

Data Preprocessing

<!-- Cleaning, filtering, transformation -->

Analysis Pipeline

[Stage 1 Name]

<!-- What, why, how - link to script -->

[Stage 2 Name]

<!-- Continue for each stage -->

Statistical Analysis

<!-- Tests, thresholds, multiple testing correction -->

Software and Reproducibility

<!-- Tools, versions, how to reproduce -->

4. Writing Standards for Methods

Level of Detail Test: Could another researcher reproduce your work?

Include Exclude

Parameter values used Step-by-step code walkthrough

Software versions Obvious standard operations

Algorithm choices with rationale Every library imported

Thresholds and why they were chosen Trivial data wrangling

Sample sizes at each step Code syntax details

Reporting Checklist:

For computational methods:

• Software name and version

• Key parameters and settings

• Random seeds (if applicable)

• Hardware specs (if relevant)

For statistical analyses:

• Test name and implementation

• Significance threshold (e.g., α = 0.05)

• Multiple testing correction method

• Effect size measures used

• Assumptions checked

For machine learning:

• Model type and hyperparameters

• Training/validation/test split

• Cross-validation strategy

• Evaluation metrics

• Feature selection method

5. Script-to-Methods Mapping

Create a mapping table:

Code-to-Methods Reference

6. Generate Methods Draft

Methods

<!-- Draft generated by Research Assistant on [DATE] Based on scripts in: scripts/ Pipeline defined in: dvc.yaml

⚠️ VERIFY:

• All software versions are current
• Parameters match params.yaml
• Nothing critical is missing -->

Overview

This study employed [brief approach summary] to [achieve aim]. The analysis pipeline was implemented in Python [version] and managed using DVC for reproducibility.

Data

Data Sources

[Describe data sources from project_telos.md or scripts]

Data Preprocessing

Script: scripts/01_preprocess.py

[Auto-generated from script docstring and code analysis]

Analysis Pipeline

[Stage Name]

Script: scripts/02_analysis.py

[Description from docstring]

Parameters:

• Parameter 1: value (rationale)
• Parameter 2: value (rationale)

Statistical Analysis

[If applicable - describe tests used]

Software and Reproducibility

All analyses were performed using:

• Python [version] with [key packages]
• R [version] with [key packages] (if applicable)
• DVC [version] for pipeline management

The complete analysis pipeline is available at [repository] and can be reproduced using:

dvc repro

Code-to-Methods Reference

Script
Section
Params

[auto-generated table]

Missing Documentation Flags

[List any scripts without docstrings or unclear purposes]

### 7. Validation Checks

After generating, verify:

Methods draft created. Validation checklist:

✓ All scripts/*.py referenced
✓ Parameters extracted from params.yaml
? Check: Are all parameter values correct?
? Check: Are software versions current?
? Check: Can a reader reproduce this?

Flagged issues:
⚠️ Script preprocessing.py lacks docstring
⚠️ params.yaml has parameters not mentioned in methods

Run /review_script preprocessing.py to add documentation.

## Reporting Guidelines by Study Type

Follow the appropriate reporting guideline:

| Study Type | Guideline | URL |
|------------|-----------|-----|
| Randomized Controlled Trials | CONSORT | consort-statement.org |
| Observational Studies | STROBE | strobe-statement.org |
| Systematic Reviews | PRISMA | prisma-statement.org |
| Diagnostic Accuracy | STARD | stard-statement.org |
| Animal Studies | ARRIVE | arriveguidelines.org |
| Machine Learning | TRIPOD+AI | tripod-statement.org |

## Related Skills

- `review-script` - Add documentation to undocumented scripts
- `write-results` - Next manuscript section (after figures exist)
- `next` - Get suggestions

## Notes

- Methods should be written in past tense (you did the work)
- Be specific: "threshold of 0.5" not "appropriate threshold"
- Link to code repository in software section
- Update methods when pipeline changes