Apply

A timed, single-sitting assignment on a real, unfamiliar codebase — a slice of our production repo. You get 120 minutes from the moment you start.

Two parts: a warm-up set of small backend bugs caught by failing unit tests (~30 min), and a genuinely tangled payments investigation (~90 min). Then a short video at the end.

You will not finish everything — that's intended. We care about how far you get and how you think. Everything you need is in the README.md inside the download.

You need Docker and Node + pnpm. Part 2 boots a local Postgres in Docker, so install Docker and have it actually running before you begin. The clock does not pause while you install things.

pnpm install                           # dependencies
cd ASSIGNMENT && docker compose up -d  # Postgres for Part 2
pnpm test                              # the failing tests are Part 1

You do not need Supabase or any cloud credentials. Nothing talks to a real server.

We read your Claude Code / Codex transcripts as part of grading — how you prompted and whether you verified what came back matters more to us than the diff. Capture is already wired into the repo, so there is no setup beyond pnpm install.

Before you submit, run this once and commit the folder:

pnpm logs:capture
git add .claude-logs && git commit -m "session logs"

Then open the folder and check your sessions are actually in there. The script reads from local Claude/Codex storage and can misbehave on an unusual setup. If it does, debugging it is fair game — that's part of the task.

Once you've pushed your work, record a ~10 minute screen recording and upload it here with your repo link. Two things:

1. 01 · Introduce yourself

   ~2 min — who you are, what you've built, how you normally work with AI tools.

2. 02 · Walk through the work

   ~8 min — share your screen and show the actual code: what you fixed in Part 1 and why, then how you traced the payments problem in Part 2 and what you concluded.

Unedited and unscripted is fine — we're not grading production value. QuickTime, Loom, OBS, or Cmd+Shift+5 all work. MP4, MOV, or WebM, up to 2 GB. Leave yourself ten minutes at the end for this.

Push your work to a public GitHub repo — keep it public so we can read it without any access dance. Then come back to the assignment page and submit both your repo URL and your video.

Don't rewrite git history or backdate commits — we read the real timeline, and an honest messy history beats a suspiciously clean one.