Every call follows the same script, a caller planning a dinner, in the same synthetic voice; what changes between scenarios is what Callback does during it, and the agent's replies are generated fresh each time. Every moment is cut from a real recording: the caller on the left channel, the agent on the right. The limits they are measured against are my choice, not an industry standard; where they come from. Select a waveform to play from that point; the arrow keys change moment and the space bar plays.