Sume jobs_wait limits: 20 ids and 55 seconds per call

A storyboard of 30 clips means 30 job ids, and one jobs_wait call will not take them all. Sume's work-bounds notes for hosted MCP state the limits: a wait accepts at most 20 ids, each at most 256 characters, and at most 55 seconds. Client headers cannot raise that to 600 seconds. Result batches also take 20 ids.

Group, do not shrink the wave

The notes say to divide observation into groups, and that this does not require smaller generation waves. Submit all 30 clips; the queue and concurrency of your plan decide when each starts (Generation admission lists them). Then wait on ids 1 to 20 and 21 to 30 in separate calls.

What a timed-out wait means

When a wait reaches its slice it returns timed_out: true with the ids you asked for, and the jobs keep running. Call again. When the server is busy, jobs_wait still answers with the same shape, and other tools can return wait_busy with retryable: true and retry_after_seconds: 1. Do not resubmit paid work because a wait ended.

Each principal can hold eight waits and the process 20, so four sibling batches of waits fit in one turn. Read results with jobs_result in groups of 20, and see the MCP tools and gates page for the inventory.