Verify carrier changes

Choose a page for your task; you do not need to read the whole documentation tree.

Task	What to read
Design evaluation samples	Declare questions, control conditions, and expected evidence.
Check a carrier	Check structure, dependencies, and measurability before measurement.
Evaluate an Agent and project context	Choose project runtime conditions and appropriate controls.
Verify local task results	Check results with actual files and independent acceptance.
Iterate candidate carriers	Generate and compare candidates after establishing useful samples.

Return to the documentation start page.

For a first run, see the evaluation quickstart; see statistical rigor for interpretation.

For long-running eval, sample, evolve, and doctor calls, the first Ctrl+C requests cancellation and cleanup. A second SIGINT schedules forced exit with code 130 after 750 ms, allowing the subprocess termination grace period first. This fallback requires a responsive Node event loop and can leave incomplete evidence; forced termination is not a completed measurement.