Test Execution

DAGZ picks which tests to run, schedules them across workers, and runs them efficiently. Workers are scheduled to balance load, share fixture setup, and steal work when they finish early. Coverage is collected at function and line granularity with near-zero overhead. Flaky tests are detected and retried. Parallel workers each get their own database, so real DB tests run concurrently without colliding.

This section covers each piece:

• Test Selection: baselines, dependency graphs, and what "affected" means.
• Scheduling: why actual durations drift from the plan, and how to tune for it.
• Coverage: semantic-unit and line coverage, exporting to standard tooling.
• Runtime Behavior: multi-node coordination, retries, progress logs, and other things DAGZ does on its own.