A node type's `ports` block is inert — decide enforce / delete / generate

[MSD21091969]

Measured, and pinned in mtdc-lab. The ports block on a node type gates nothing in this repo.

The mechanism

internal/operad/validate.go ValidateLINK checks: WF known → WF allows LINK → WF15 contract → linkPairDeclared (the (src_port, tgt_port) tuple against the WF's primary + additional pairs) → port colour compatibility. Pair-level src/tgt types are checked in the state-aware path.

Nothing reads a node type's ports. internal/operad/loader.go:181 parses the block into a PortSpec and stores it at registry.go:34; every non-test read of .Ports / PortSpec in internal/ is the loader and the struct definition. Written, never read.

What that permits, measured against ontology 4.0.7

	count
declared ports named by no WF pair — un-LINKable	44 of 82 (53.7%)
(WF-pair, type) obligations naming a port the type does not declare	123, across 42 of 56 types

Concrete instance: ontology 4.0.6 added manifold to the WF01 owns/owned-by tgt_types, while manifold declares only {"self": ["identity"]}.

Both directions are large because the two namespaces are never reconciled by anything executable.

Evidence

• mtdc-lab/moos-generated/tests/ports_are_inert.rs — 3 tests, pinned at 4.0.7.
• mtdc-lab/findings/20260809-t281-round7-port-census.md — the 44.
• Independent corroboration: mtdc-lab's genlib::emit_tables emits id, qualifier_segments, urn_nid, urn_pattern per node type and no ports — the generator already concluded they are not part of verification.

What is owed here

A Go test pinning the oracle side. mtdc-lab can only pin that its mirror admits a LINK onto an undeclared port; it cannot execute this repo's validator. Until a test here asserts that ValidateLINK admits such an envelope, the claim rests on the source read above. That test is small and belongs in internal/operad.

The ruling, which is not this issue's to make

Three options — enforce (fail-closed, breaks 123 obligations, needs the ontology repaired first), delete, or generate ports as a projection of the pair table with a drift gate. Costed at mtdc-lab/proposals/ffs0-ports-as-projection.md. Reserved to Sam.

Note that this repo's own CLAUDE.md already says "Relations are truth. Properties never duplicate topology" — which argues one way without settling it.