The summary.
WRITE TESTS IN THE LANGUAGE YOUR DEVELOPERS ALREADY USE
Tests then get maintained by whoever changes the code, rather than lagging behind it.
Keep tests alongside the code they test — separating them guarantees drift.
TESTS SHOULD READ AS SCENARIOS, WITH NO SELECTORS, WAITS OR LOW-LEVEL DETAIL
Interface changes must touch one layer, not every test. And page objects hold actions, not assertions.
NEVER ADJUST A TEST UNTIL IT PASSES WITHOUT UNDERSTANDING THE FAILURE
It removes the detection without removing the problem. In review, flag any assertion relaxed or removed — weakening a test to pass a change is quiet and common.
PARALLELISM REVEALS HIDDEN DEPENDENCIES AS NEW FLAKINESS
Fix the dependency rather than reducing parallelism. Every test creating its own data with generated identifiers is what makes it possible.
EVERY FAILURE REPORT NEEDS ACTUAL AND EXPECTED, PLUS APPLICATION LOGS
The test sees symptoms; the application logs the cause.
BUDGET MAINTENANCE TIME EXPLICITLY
Otherwise it happens only once the suite is already unusable.
