proof — the right test at the right boundary

The skill is the boundary choice. A suite of 500 trivial unit tests that misses the broken checkout proves nothing; three well-placed API tests that would catch it prove everything. Test count is not the goal — catching real regressions is.

When NOT to use

• The bug isn't understood yet → sleuth finds the cause; you lock it down with a regression test after.
• The question is whether the flow works in a real browser → roadtest.
• Pure performance → hotpath.

Prerequisites