Testing Lessons

An API tool that fetches one document by repository path was given a guard: the path had to look like a

Automated stress testing of intake payloads between 20,000 and 60,000 characters requires end-to-end length verification

A pull request that **only added one lesson file** came back red with:

A check that is green no matter what the code does is worse than no check, because it occupies the place

While triaging a crash, the reporter ran **two rounds, saw one crash**, and concluded the failure

`lessons/contrib/assertion-that-cannot-fail.md` covers assertions that pass whatever the

Two bugs survived a fully green compile and shipped in a release whose notes claimed both were

Monitoring "did the restart produce a new crash signature in the log" produced a distorted result:

Production code adds a new attribute access (e.g. `self.paths.config_path`), and every test that mocks that object with

Cache-invalidation tests pass on developer machines and fail on fresh CI runners. The tests forced expiration by overwri

Automating an Ant Design `` with Playwright, `page.keyboard.press('Escape')` did **not**

CI skipped the tests that "cannot run locally" with a selection expression like:

A pytest fixture declared with `scope="module"` handed the **same mutable clone** to three sandbox tests. A mutator test

A batch verifier checked one decision procedure across 61 samples. Its final line read:

改前端导航/布局——把左侧 `Menu` 换成顶部 `Tabs`——时，只改了**一个** UI 自动化脚本的选择器，其余几个还在找旧的 CSS 类名（如 `.ant-menu-item`）。

To align a host platform's plugin against its slot / extension point names, the reporter

Upgrading `vitest` and `@vitest/coverage-v8` from 3.x to 4.x dropped branch coverage from 81.30% to 74.96%, causing the

Retrospective validation of a scoring / prediction tool — feed samples with known outcomes back in