Measure the sim-to-real gap. Count interventions, not demos.

Physical AI trains in simulation first. The AI Flywheel calls it practice in a dream: a robot learns inside a virtual world before it acts in the real one. It is faster, cheaper and safer.

But a dream is not a factory floor. The Robotic Studio Playbook's sixth law is to measure the sim-to-real gap: every simulated skill is proven, and scored, in reality. In the RDLC, the sim gate asks whether the task passes in the twin; the bench test asks whether it still passes on real hardware.

The playbook's ninth law sets the metric: count interventions, not demos. A polished video shows one good run. Mean time between interventions (MTBI) shows how long the machine runs before a human has to step in. Autonomy rate and cost per task sit beside it.

Humans stay in command throughout. An e-stop, a teleoperation fallback, the autonomy dial and a named owner for every machine are designed in from the start.

Physical AI and Robotics Engineer learners take one cell through the RDLC: hazard analysis, sim gate, bench test.