This guide runs every check Redact has, from fast unit tests to a browser run against real local inference. Run commands from the redact directory.

​
Static checks and unit tests

Run typecheck and build one after the other, not at the same time: both write Next.js generated files. npm test runs 87 Vitest tests and needs no models.

​
Browser tests

Playwright needs a browser. Install one once:
The browser is about 550 MB and took about five and a half minutes to install in testing. If you see Executable doesn't exist at …/ms-playwright/…, this step was skipped. The manual-workflow tests expect no local service: they check the screens Redact shows when analysis is unavailable. Stop qvac:service first, start the app, then run them:
Expect 6 passed and 2 skipped. The skipped test is the live one below, which is opt-in.

​
Live browser test with real inference

This test uploads a file, waits for real suggestions, accepts and rejects them, saves the copy, and checks a pixel in the download. It needs the models, the service and the app.
  1. Start the service: npm run qvac:service.
  2. Start the app on port 3101.
  3. Run:
The first run includes model loading and takes up to about 1.5 minutes. The test allows 240 seconds. For the same test with the service sandboxed, see How to run Redact with the network blocked.

​
Labelled evaluation

Runs four labelled pages through OCR and reasoning and exits non-zero if any falls below the thresholds. About 2 minutes. See Evaluation reference.