It seems that Prefect does not inherently apply exponential backoff when checking for open concurrency slots. Instead, the behavior you're observing might be related to how the slot decay mechanism works. When you set a slot decay rate, it determines how quickly slots refresh and become available for new tasks.
With a slot decay rate of zero, slots should become available instantly once a task completes. However, if you're noticing tasks starting in batches, it might be due to other system-level factors or configurations outside of Prefect's direct control.
For more details on configuring slot decay, you can refer to the
Prefect documentation on global concurrency limits.