When an assistant is asked to wait

Your site shares a machine with other organizations’ sites. An assistant works far faster than a person and does not get tired, so without limits one enthusiastic job could make everybody’s lessons slow to load. These are the limits, and the thing to know about all of them is that they pace work rather than refuse it.

A large change is queued

Ask for something big — rewriting forty lessons, translating a whole course — and the reply will tell you it has been queued, and give you a job number. Nothing has been written at that point. The work is then applied steadily over the next few minutes, a little at a time.

Your assistant can ask about that job number and will be told whether it is waiting, running, finished, or failed. If it finished, it gets the full report of what changed.

If an assistant tells you work was queued and then appears to do nothing, ask it to check the job rather than to try again. Submitting the same work twice does not make it happen sooner, and it does spend your allowance twice.

The hourly allowances

Two things are counted, per hour: changes made and rows read. Requests are not counted, because one request that rewrites a thousand lessons is expensive and a hundred questions about a course outline are not — counting requests would have it exactly backwards.

Level	Changes an hour	What it is for
One connection	2,000	One assistant that has lost its footing
Your whole site	5,000	Several connections that each look reasonable

Two thousand changes an hour is a bigger course than anybody writes in a day. If you are meeting it, something has gone round in a circle.

The site allowance is the one worth understanding. It is shared by every key on your site — so creating more keys does not give you more capacity, and splitting a job between them does nothing. It is worth about two and a half connections’ worth, so running two assistants at once is a normal thing that works, and running ten is not.

There is a third limit above both, covering every site on the platform together. You will almost certainly never meet it. If you do, the message says plainly that it is nothing to do with your site and nothing you change will clear it sooner.

What the messages mean

Every refusal says which limit was reached, roughly when it refills, and — the useful part — what will not help:

• “This key has used its hourly allowance.” Wait. Do not move the work to another key; they share the site’s allowance.
• “Every key on this site has together used the site’s hourly allowance.” Wait. A new key will not raise it.
• “The platform has reached its hourly limit across all sites.” Wait. Not yours, not fixable from your end.
• “Grapevine is catching up on queued work and is not accepting more just now.” There is a backlog. It clears on its own.

These are written for the assistant to read, not just for you, and a well-behaved one will wait and then carry on by itself.

Why you may notice a job is slow and should not worry

Queued work is deliberately gentle. It does a piece, measures how long that took, and rests for proportionally as long before the next piece. So when the server is busy, it slows down on its own — without anything having to decide what “busy” means. A translation that would take forty seconds flat out may take several minutes instead, and in exchange nobody taking a course notices anything at all.

Large course copies are queued the same way. If you copy a course with hundreds of lessons, expect a job number rather than an immediate copy.