find_missing_admin_chunks() only parsed index.js, but a lazily
imported page chunk pulls in its own deps (shared-api, PageHeader,
sortable-options…). A deploy that drops one of those still leaves the
entry internally consistent, so the existing notice never fires and the
failure only shows up when an admin opens that menu — every schemaPage
keeps working, which makes the breakage look arbitrary.
The same scan never looked at CSS at all: 18 stylesheets are reachable
from the graph and none of them were checked, so a missing page
stylesheet ships an unstyled screen with no warning.
Walk the module graph transitively from index.js, queueing each
discovered dependency, and match .js and .css specifiers. Deleting
PageHeader now also reports every page that depends on it.
This is a diagnostic hardening only — it makes an incomplete deploy loud
instead of guessing at the cause of a broken page.
