A run on a person's provider waited in its process, and a job that had not
finished in two hours was cancelled at the provider, losing its place in the
queue. If the process stopped first, the job went on and its id was gone, so its
counts could not be read. A device's queue can take days.
A job is now written down before it is sent, with everything sealing it will
need, and the provider's id as soon as the provider gives one. Nothing that
cannot be recorded is sent. Any process can come back to the record, and its
counts are read against the program that was sent, whatever the device's
calibration has become. Running out of patience leaves the job where it is. A
request that did not complete may have reached the provider, so it is never sent
again: that could be a second bill.
The command line gains run --no-wait and providers jobs, collect and cancel. The
local MCP server's run_on_provider answers within a call, and gains
get_provider_job, cancel_provider_job and list_provider_jobs.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
5307518b27