John Barrios · Yale School of Management Claude Code for Accounting Research

Labs

Five labs, one per build module, each 80 minutes inside the module session. Every lab page carries the full instructions, the tiered task menu (base / standard / stretch), and the verify checklist you self-grade against — pass/fail, spot-checked in the debrief. Starter packs are small zips: templates and checklists, never data and never credentials. You download data yourself from public sources, and Module 4’s WRDS pull runs on your own account (a public-data fallback track covers anyone whose access isn’t live yet).

Lab	Builds	Starter
Module 1 — Install, project setup, one verified task	Working environment, first CLAUDE.md, one externally verified number	zip
Module 2 — The empirical loop on SEC FSDS	Download → clean → figure → replication README, with paper trail	zip
Module 3 — 10-K text panel + your first skill	EDGAR pipeline with quality report, validity protocol, SKILL.md	zip
Module 4 — WRDS pipeline (live or fallback track)	Filtered funda extract → Parquet/DuckDB → metadata table (+ .dta)	zip
Module 5 — Capstone: harden a pipeline	One earlier pipeline made reproducible, verified, and hygienic against the rubric	zip

The capstone rubric on the Module 5 page is the course’s summative standard: reruns from the README, paper trail cross-checked against commits, no hand-typed numbers, credentials audited clean. Module 6 has no lab — its deliverable is the position memo described on the seminar page.