Test First

The iron law

No production code without a failing test first.

Not "usually". Not "unless it's small". The rule has no exception clause, because every exception anyone has proposed was a case where the test was skipped and the bug shipped.

The cycle

RED — write one failing test

One test, for one behavior, with a name that says what the behavior is. Use real code, not mocks of the thing you are testing. A mock of your own function tests that you can configure a mock.

Before you write it, name the production change that would make this test fail. If you cannot name it, you do not yet know what you are testing.

Verify RED — mandatory

Run it. Watch it fail. Then check why it failed:

• It must fail, not error. An import error is not a red test.
• It must fail because the behavior is missing, not because of a typo, a missing fixture, or a wrong path.
• If it passes, you are testing behavior that already exists. Stop and find the real gap.

Skipping this step is where the whole cycle stops working, because a test that never failed proves nothing when it passes.

GREEN — the minimum

Write the least code that makes it pass. No speculative parameters, no "we'll need this later", no configuration for a case nobody asked for.

Verify GREEN

Run it and watch it pass. Then run the surrounding suite: your change must not have broken anything. Read the output — not the exit code alone.

REFACTOR — only on green

Improve names and structure with the tests passing. Run them again after.

If you wrote code first

Delete it. Do not keep it in a scratch file, do not adapt it, do not look at it while writing the test. Code you can see is code the test will be shaped to fit, which produces a test that passes for the wrong reason.

Rationalizations

Thought	Reality
"It's a one-line change"	One-line changes ship bugs. That is why they are worth a test.
"I'll add the test after"	The test after is shaped by the code. It tests what you built, not what was needed.
"It's just a config value"	If a wrong value breaks things, a test can catch a wrong value.
"There's no framework here"	Then the first task is the framework, and it is a real task.
"I already know it works"	Then the test passes immediately and costs a minute. Write it.
"The test would be trivial"	Trivial tests catch typos, which are the most common defect.
"This is exploratory"	Exploration is a spike, and its code is throwaway. Label it and delete it.

Regression tests

A bugfix needs a test that fails before the fix and passes after. Prove it both ways: write the test, watch it fail, apply the fix, watch it pass, revert the fix and watch it fail again, restore. A regression test you never saw fail is a regression test you cannot trust.