Published is not the same as answerable

A crawl or a connected system can carry a schedule of daily or weekly. A check runs every hour and queues any scheduled source whose last run is older than that interval, so a daily source becomes due at some point in the day after it last ran rather than at a fixed time you can plan around.

The consequence is the part people miss. Publish a page an hour after the crawl ran and the wait is close to a full interval, not close to nothing. On a weekly schedule that is most of a week. The average wait on a daily source is around half a day and the worst case is a full one, and neither of those is a fault.

Uploaded files have no schedule at all. A file source is indexed when it is uploaded and again when somebody presses Resync. Editing the document on your own machine changes nothing here, because nothing is watching your machine.

A crawl starts from an address and follows links, to a depth, staying on the same site by default, and it can use the site's own map of itself. A page that nothing links to and that is not in that map is never fetched, and waiting longer does not help. The lag in that case is not long, it is infinite, and it shows up as a topic that keeps refusing.

There is a size ceiling too. A crawl takes up to the pages your plan allows, minus what is already indexed, and stops there. Where there is no room left the source fails with a message naming the limit rather than pretending, so it is visible when it happens, but a page can be missing because there was no space rather than because there was no time.

Pages are fetched, text is extracted, split into passages and stored so they can be matched against questions. While that happens the source shows as indexing with a progress bar, and it finishes as ready, partial or failed, with the reason attached when something went wrong.

Content that has not changed is recognised and skipped, which is what makes a daily re-crawl of a mostly static site cheap. It also explains a confusing case: a page you edited trivially may produce no new material, and a page you thought you changed but did not will look exactly like a page the crawler ignored.

Answers are cached for at most a day, and only answers that cited something are cached at all. So a question somebody already asked can come back with the previous answer even after new material exists, for up to that long.

This matters most when you test your own change. Ask the question in the exact words somebody asked it yesterday and you may be reading yesterday's answer. Ask it in your own words and you are more likely to get a fresh one. Either way the citation underneath is what tells you which material actually answered.

Start the clock when the sync started. The run takes minutes and the figure is excellent, and the hours or days of waiting for the run to be due are outside the measurement entirely. This is the honest looking version and it describes the half nobody was asking about.

Set everything to daily and call the lag a day. Pages nothing links to are never fetched at any schedule, and they contribute an unbounded lag that no average will ever contain, because they are not in the sample.

The accidental version is a test that lies to you. Press Resync, ask the question in the same words as before, get the old cached answer, conclude the sync did not work, and press it again. Nothing is wrong except the test. Read the citation rather than the answer, since the citation names which document was actually used.