local-testinglisted

# Local testing — prove it at runtime, not in the suite - **Can:** run the suite, the build, the shipped artifact and the real flow — L1 through L5. - **Must:** compute the expected value *before* looking at the screen, and declare the level you actually reached. - **Cannot:** treat a green suite as proof, or edit the implementation to make a check pass. **A green suite is not proof that it works.** The test runs in one environment; the product runs in another. > Measured in this kit: **21 tests passing while the git audit had four holes**, including one that > let an executor overwrite any file the human was already editing without a word. Every one of those > tests started from a **clean working tree**, which is not what a repository in use looks like. The > suite was not lying — it was answering a question nobody had asked it. The levels are **cumulative**: L3 without L1 is theater; L1 without L3 is faith. | Level | What it proves | Cost | |---|---|---| | **L1** | the logic is right and the module does not use the wrong environment's APIs | seconds | | **L2** | the production artifact builds and opens without errors | ~1 min | | **L3** | the real flow, with real data, shows the right result | ~5 min | | **L4** | it works in the genuinely hostile environment (another engine, another device, a host page) | ~10 min | > 🔒 If the project has **no staging**, the levels stop being a recommendation and become a > **condition for shipping**. Skipping one requires an explicit