Start with an input small enough that you can calculate the result yourself. Call the function in a test and compare its result with that answer. For a subtotal function, three numbers are enough to check the usual case before testing an empty list or an invalid value.
Put logic where you can call it
Separate a calculation or transformation from terminal prompts and file access where practical. A function that receives a list and returns a summary is easier to test than code that immediately reads a user's Downloads folder and prints everything.
This separation also makes the program easier to explain. One part reads input, one validates and transforms it, and one writes the result. Each boundary gives you a place to ask a clear question.
Use known outcomes
Suppose a function totals accepted counts. Test two small positive values, an empty list, and a rejected negative value if your contract forbids negatives. The expected answers should come from your stated rule, not from calling the same function to calculate its own expected result.
Python's unittest module is enough for these first checks. A failed assertion should tell you which behavior changed. Name tests after that behavior, such as test_empty_input_returns_empty_summary, rather than an arbitrary sequence number.
Control files and network context
Use temporary directories for tests that write files. Avoid depending on a particular file already existing on your computer. Keep fixture data small and synthetic so the test remains understandable and safe to share.
A public website is not a reliable unit-test fixture. For networking practice, a controlled loopback listener provides a known local condition. For a feed parser, a bundled synthetic RSS or Atom document can test parsing without making a publisher request.
Test failures that matter
Verify that an invalid input is rejected, that a failed collection does not erase previous good data, and that an existing output is not overwritten when the program promises exclusive creation. Those cases protect useful guarantees.
Mocks help simulate conditions that are hard to reproduce, but label their role honestly. A mocked timeout does not prove how every operating system or network will behave. Keep actual environment evidence separate from simulated failure cases.
Document what your test suite actually checks
Record the Python version, operating system, source revision, and test result for a release. Passing tests supports the behaviors those tests cover. It does not prove universal compatibility, a successful fresh-machine installation, or a correct conclusion from real-world data you have never inspected.
If the tests work in a terminal but your editor can’t find them, troubleshoot pytest discovery in VS Code.