Under a cap, slow new tests — don't cut winners

What happened

Simulation: 65.1% ROAS throttling tests vs 60.2% cutting working ad sets.

Build the lesson into your workflow.