Course: Diagnosing failures — 1. Course overview
A hands-on course on finding out why a run failed: the evidence a run keeps, the failure class it gets, the AI’s suggestion when no rule can name the failure, and what a failing shared component means for every test that uses it. Each module uses a small demo application that fails in known ways, so you always know the right answer.
Modules
Section titled “Modules”| Module | What you learn | Time |
|---|---|---|
| 1. This page | What the course covers, and what you need | 5 min |
| 2. Logs, video and trace | What the application reported around a failed step; the video and trace; secrets in recordings; how long evidence is kept | 1 h |
| 3. Failure classification | Failures the application caused and failures on the test’s side; the investigation report; how history moves the confidence | 1 h |
| 4. AI diagnosis | The AI’s suggestion for Unclassified failures, what it costs, telling it it was wrong, and turning suggestions into rules | 1 h |
| 5. Component maintenance | Which shared components need a person, and why the failure class decides between fixing the component and looking at the application | 1 h |
Do the modules in order: module 4 uses the demo application of module 3.
Who it is for
Section titled “Who it is for”- Testers who read failed runs every day: modules 2 and 3.
- Test managers who decide whether a failure blocks a release, and who own shared components: all modules.
- Administrators who set how long evidence is kept and whether the AI may be used: modules 2 and 4.
What you need
Section titled “What you need”- Your own installation (Deployment). The demo applications run on your computer, and the browser that runs your tests has to reach them.
- The Admin role in a practice project.
- For module 4, an Anthropic API key in the installation’s
.envfile. The module makes four real AI calls (a few cents).
Learning objectives
Section titled “Learning objectives”By the end you can:
- tell an application failure from a test failure using the evidence a run keeps;
- read a failed step’s console and network logs, video and trace;
- name the failure class of a run, and explain the likely causes and their confidence;
- read and judge the AI’s suggestion for a failure no rule matches, and mark it Helpful or Wrong;
- turn repeated Unclassified causes into new rules;
- use the component maintenance queue to decide what to fix first.