Measure replacement tests on a common coverage denominator

When replacing unit tests with end-to-end tests, “coverage must not decrease” is not measurable unless both suites feed the same coverage system. A unit runner’s statement and branch report excludes browser and server execution from a separate end-to-end runner, so deleting unit tests can appear to reduce coverage even when behavior coverage improves. Establish the common instrumentation and baseline first, or explicitly use a boundary-to-oracle matrix plus mutation controls as the acceptance metric. Do not silently substitute artifact count or test count for coverage.

Applies when: A repository is moving substantial coverage between test layers. Keywords: unit tests end to end coverage baseline instrumentation mutation