The effort was opened to ask whether the catalogue's missing structure is a
graph. It is not missing. 126 manifests, 103 edges, no cycles, nothing dangling,
deepest chain of five — and a resolver in the SDK that topologically sorts them,
already called by the tool loader at startup, the installer when syncing modules
onto a node, and the delivery coordinator when expanding what a change affects.
It already does something this effort assumed would need designing: a
requirement on another module's provision is treated as an implicit edge to the
module that provides it. So "ordering by the graph", which ADR 0043 makes the
control plane's job, is a thing to call rather than a thing to build.
The one place the graph is wrong, it is wrong about the substrate. A module
needing a database declares `provider: postgres` inside `provisions:` — which is
what a module OFFERS — so the resolver, which reads `dependencies:` and
`requires:`, never sees it. Three edges are invisible this way, and they are the
mesh's own database, the mesh's own broker, and the work engine's database.
The consequence is measurable: computing what a working mesh needs from the
declared graph gives registry -> sdk -> mesh -> meshware. Four modules, four
levels, no database. Arithmetically correct and obviously wrong, for exactly one
reason — a field that means "depends on" is not read as one. That is
04-ISSUES/003 in a new form: not a key nothing reads, but a key read as
something other than what it means.
Two latent defects, both contrary to ADR 0008 and both in the component ADR 0043
makes responsible for ordering a host will apply without question: a cycle warns
and falls back to input order, and a dependency that does not exist warns and
continues. Neither has fired, because the catalogue currently has no cycles and
nothing dangling, which is why nobody has noticed.
And placement is decided in the catalogue: a provision pins itself to a named
node in the manifest. Which node runs what is an inventory decision — tier 2 by
the skeleton's own test — so a second node cannot provide the mesh's database
without editing the module that consumes it.
What the graph would DELETE is currently nothing. What it would add is three
declarations that no manifest uses today: excludes, a required node capability,
and an interface with adapters. Whether they would be used is not measured, and
zero usage is equally consistent with nobody needing them and nobody being able
to express them.
