Each test says who to sign in as, what to do, and what should happen. Mark it Pass, Fail or Blocked (couldn't run it), and add a note when something's off. Your results stay in this browser. When you're done, Copy results and paste them to Claude.
The same sheet is used for every round. Claude adds new tests to the list inside this file (each with a stable number, like CA-03), and when it needs you to test something it sends a link naming just those tests, for example test-sheet.html?tests=CA-03,CA-05 or one area, ?area=PP. Start a new run clears your marks for a fresh round; when the tests change, the sheet's version moves on and old marks are set aside automatically. New modules get their own letters (OR = Ordering, FS = Food safety, and so on) as they're added.
These run from scripts on every database change, so you don't need to test them by hand: each database step's own checks (checks/R37-*.sql), the security test that tries every cross-company and cross-venue read and write (tests/isolation.py), the no-hard-coding count (tools/check-hardcoded.py), the shared venue code check (tools/check-venue-block.py), and the design scan. This sheet is for what only a person using the screens can judge.