ProtectedcheckingProtectedcurrentProtectedfailedProtectedloadingProtectednewestProtectedpreflightProtectedseenAsked, like the nightly build. Updating Flarum is the change whose blast radius is the whole forum, and it was the one button here that started on a single press.
Core AND every extension that needs to move, resolved together in one request. Updating core alone would leave Composer refusing, because the extensions still on disk pin the old version — and updating them one at a time afterwards means a forum that is broken in between.
Whether the pre-flight has answered, and nothing it found blocks the update.
Not just "nothing is blocked". Until the check comes back nothing is blocked because nothing has been checked, so the button was live from the moment the panel drew, and pressing it then sent core alone: no extension that needed to move with it, and no warning about the ones that could not. A pre-flight that failed to load counts as not answered. A batch of extensions that could not be checked does not: those rows say so, and an unknown is not a refusal.
One batch at a time, so no single request can cause a burst of outbound calls.
Flarum itself.
Updating core is the one operation whose blast radius is the whole forum, so the question "what would this do to my extensions" is answered before anything is pressed, per extension, by name. Composer's answer to the same question is a refusal at the end of a long wait, written in package constraints, from which somebody has to work out which of their thirty extensions is the problem.
The pre-flight arrives in two parts and this is not an optimisation. The cheap half — does the version you ALREADY have declare support — is in composer.lock and is instant for any number of extensions. The rest costs one HTTP call each: 58 of them took 28 seconds in a single request when this was first built, which is the very failure the extension exists to prevent.