The platform's routes go first, so nothing has to be policed

[czpython]

The guard #131 added to keep an extension out of the subject namespace read every declared route's path and compared it to the segment. Both halves of that were wrong.

parent.include_router(child) leaves a lazily-flattened entry behind that carries no path at all, so an extension composing its own routers stopped loading:

AttributeError: '_IncludedRouter' object has no attribute 'path'

And a path string was never enough to decide the question. @router.get("/{resource}") names no segment and still answers /api/x/widget; a mount("/", …) has no path to read. Inspection cannot tell whether a route will match a segment — only routing can.

So the platform stops asking and simply goes first:

platform = [cls._get_transcript_routes()]
if cls.subject_type:
    platform.append(cls._get_subject_routes(cls.subject_type))
return [*platform, *declared]

Both platform routers are confined to their own segment, so matching them first takes nothing from an extension anywhere else and leaves it no way to take a read druks serves — not with a catch-all, not by composing routers, and not by overriding get_routers to put its own first. The guard, its exception and the two tests that policed spelling are gone; what replaces them is a request proving a {anything:path} route still loses /widget and still wins everything else.

One collision inspection did have to keep: transcripts is the segment every extension's agent-call reads live on, so a subject of that name is refused where the extension is declared, not where it would have been confusing.

The convention this documented was also written down wrong. night_watch and pull_request are snake_case and both are URL segments — what the spelling tells you is who owns the segment, not whether it is a URL. And usage still declared the tag the platform now stamps, so every one of its operations carried it twice.