Test setup in a clean context

Your current Mac contains years of hidden state. A setup script can appear complete while depending on an old PATH entry, cached credential, or manually installed tool. On the machine where you wrote it, those dependencies are invisible — everything is already there, so nothing fails.

The only honest test runs where none of that state exists.

Get a clean context

Use a fresh local account or disposable test machine. A new macOS user account is the cheap option: System Settings, Users & Groups, add a standard user named setuptest. It shares the OS but starts with an empty home directory — no dotfiles, no Homebrew PATH wiring, no agents, no caches. For higher-stakes verification, a wiped spare Mac or a macOS virtual machine is closer to the real “day one” experience.

Two rules protect the test’s honesty and your data. Follow only the written prerequisites — if you have to remember something, the document failed. And run the installer without secrets: a clean-context test needs no production credentials, and do not use a production account or irreplaceable data for the test.

Run it like a stranger

Log into the fresh account, open Terminal, and follow your runbook top to bottom. Record every prompt or missing assumption as you hit it:

## Test run — 2026-08-03, fresh account

- step 2: `git clone` asked for auth — doc assumes SSH key already exists
- step 4: `brew` not found in new shell — shellenv line missing from doc
- step 6: installer paused on a license prompt the doc never mentions

Each line is a bug in the documentation, found for free.

Verify outcomes, not vibes

“It seemed to work” is not a result. Verify commands, runtime versions, Git identity, and a sample project build:

command -v brew git node
node --version          # matches the project's .nvmrc?
git config --get user.email
cd ~/dev/sample-project && npm install && npm test

The realistic finding — and you should expect one — is a hidden dependency: some tool you installed manually in 2024 that no document mentions, which the sample build needs. That is the test succeeding. Add the missing piece to the setup, wipe the test account, and run it again until a stranger could finish it alone.