An update announced on one day may not appear on a given device for weeks. The delay is a deliberate release method rather than a queue.
Testing cannot cover real conditions
Internal testing runs on a controlled set of devices, configurations and networks, all reasonably current and reasonably clean.
The installed base is nothing like that. It includes older hardware, unusual regional carrier settings, storage that is nearly full and years of accumulated apps.
Faults that appear only in those combinations cannot be found in a laboratory, so they are found in a small, monitored share of real installations instead.
Rollouts expand in stages
A release typically goes to internal users, then to volunteers, then to a small percentage of the public, widening only if the measurements stay clean.
Each stage is held for a period long enough for problems to surface, since some faults only appear after a device has run the new version for days.
If a stage looks wrong, the expansion pauses. This is why an update can appear on a colleague's device and never arrive on yours that month.
The signals watched are indirect
Engineers monitor crash rates, battery drain, application start times and how many users roll back or disable a feature.
These are compared against the previous version on similar hardware, because an absolute number means little without the baseline it replaced.
Support contacts and store reviews feed in too, and a rise in a particular complaint can stop a rollout before the technical measurements move.
Server-side switches decouple shipping from launching
Much of what is new arrives in the installed code but stays inactive behind a remote flag.
That separation lets a feature be enabled for a fraction of users and switched off instantly without publishing a new build or waiting for stores to review it.
It also means two people running the same version number can see different interfaces, which is often mistaken for a partial download.
Device makers add another layer
On phones, an operating system release passes to manufacturers and sometimes to carriers, each of which adapts and tests it for their own models.
Older models are validated last, or not at all, because the testing cost is fixed while the remaining user base shrinks each year.
The practical result is that the gap between announcement and arrival depends less on the software than on how many parties must approve it for your particular device.