“It opened for me” is not a test. A test connects a requirement to an action and an expected result. For a small product, ten carefully run manual checks are more useful than a README that says “tested” when no one ran anything.
Four levels of verification
- Function: empty fields, valid submission, repeated click, network error.
- Integration: did the record really save; was the API response handled correctly?
- Interface: readability, keyboard use, focus, errors, and loading states.
- Responsive layout: clipping, horizontal scroll, or covered buttons on a phone.
Automate repetitive critical checks after writing the manual scenario. Playwright is one browser automation option; it does not replace trying the site on a real phone or asking a person to use it.
| ID | Condition | Expected behavior |
|---|---|---|
| 1 | Valid fields | One request and one confirmation |
| 2 | Name is empty | Submission stops; the field is identified |
| 3 | Contact is empty | Submission stops; the hint is understandable |
| 4 | Double click | No duplicate request is created |
| 5 | Server returns 500 | No false success; input stays available |
| 6 | Very long text | The interface remains usable; input has a sensible limit |
| 7 | Keyboard only | Fields and button are reachable; focus is visible |
| 8 | 360 px viewport | No horizontal scroll or clipped action |
Check states, not just a screenshot of the happy path. If a database is connected, verify a real test record and a denied attempt to read another user’s data.
Record a defect before fixing it
Capture reproduction steps, expected and actual behavior, device/viewport, error details, and the last changed area. Ask the agent to fix that behavior only. Repeat the original scenario and nearby regressions after the fix.
Practice
- Fill in the manual test table.
- Run it on a phone or emulated viewport, then on a wide screen.
- Check keyboard-only use.
- File one reproducible defect, fix it, and repeat the scenario.
- Note which checks would be worth automating before the next release.
Save: test results and one fixed defect with evidence that it no longer reproduces.
Self-check
- Did you test failure and duplicate-action states?
- Do you know what automation covers and what it does not?
- Can you show that the bug stopped reproducing?