work-analysis

name	Work Analysis
description	Execute statistical analysis pipeline, generate tables and figures

Work Analysis Sub-Skill

Overview

Run the analysis pipeline from data to outputs. Every result must be reproducible -- fixed seeds, documented transformations, and a reproducibility manifest. This sub-skill is invoked by the work router -- it does not run standalone.

Agent Delegation

Deploy analysis agents:

• Analyst: .claude/agents/drl/analyst.md -- executes regressions and generates outputs
• Data-quality-reviewer: validates data integrity before analysis
• Robustness-checker: runs alternative specifications and sensitivity checks

Methodology

Step 1: Understand the Task

1. Read the task description and the research plan's model equations
2. Identify which analyses are required (descriptive, main regressions, robustness)
3. Check bd show <epic> for the variable operationalization table

Step 2: Load and Clean Data

1. Load raw data using src/data/loaders.py
2. Apply cleaning rules from src/data/cleaners.py
3. Validate against the operationalization table
4. Write analysis-ready datasets

Step 3: Run Analysis Pipeline

Execute in order, outputting to paper/outputs/tables/ and paper/outputs/figures/:

1. Descriptive statistics (src/analysis/descriptive.py): summary stats, correlations, distributions
2. Main regressions (src/analysis/econometrics.py): implement each model equation per the plan, follow variable operationalization exactly
3. Robustness battery (src/analysis/robustness.py): alternative specifications, sensitivity analyses

Step 4: Log All Decisions

Log ALL methodological decisions to docs/decisions/ using the ADR template:

• Variable exclusions or transformations
• Winsorization or outlier handling choices
• Sample restrictions applied
• Any deviation from the original plan

Use /drl:decision for guided logging.

Step 5: Update Reproducibility Manifest

uv run python -m src.orchestrators.repro

Verification Gate

• Reproducibility manifest updated
• Output tables exist in paper/outputs/tables/
• Output figures exist in paper/outputs/figures/
• All analyses from the plan are executed (no skipped models)

Scope

• Data pipeline: src/data/
• Analysis code: src/analysis/
• Visualization: src/visualization/
• Outputs: paper/outputs/tables/, paper/outputs/figures/