- Shipped
- September 28, 2026 at 6:48 AM UTC
- Author
- Kamo
- Commit
- 7a4f1b9
An AI workstation's wake took 9 min 19 s against ComputeService's 75 s estimate (SP11 Task 30). Every computer_* call on a waking machine answered WORKSTATION_WAKING with retryAfterSeconds 15, so an AI waiting out that wake could spend up to about thirty of a run's sixty tool calls asking. The pace now follows how long the machine has been getting ready: half of that, from 15 s up to the 90 s the runtime waits at most. The start is the Tool Plane's own wake, or the first time any pod saw the machine waking (Redis, remembered 20 minutes and forgotten once it answers), and after two minutes the notice says how long it has been and that a wake can take up to about ten minutes. The first answer after a wake is ComputeService's estimate, never longer than 90 s. Without Redis the pace is 15 s as before.