Record value types as out of scope (ADR-0007)

docs/06-schema-reference.md

@@ -248,7 +248,8 @@ is *not* here is as useful as knowing what is.
 - **Value objects.** There is no value-object type. Model a value-shaped concept
   as an owned entity or as attributes on its owner. The parking-garage example
   models `Ticket` this way and names the tension. First-class structured value
-  types are being explored
+  types were considered and set aside: named, typed fields define record
+  structure — an implementation detail this format abstracts away
   ([issue #11](https://github.com/stacklok/modelith/issues/11)).
 - **Domain events.** There is no event construct. A state change is expressed as
   an `action` plus the invariants it `preserves`, and enums **name** states

project-docs/adr/0005-value-types.md

@@ -1,3 +1,7 @@
+---
+Status: superseded by ADR-0007
+---
+
 # Value types for structured non-entity values
 
 Add a top-level `types:` section for named value types: structured values with

project-docs/adr/0006-action-reports-not-invariants.md

@@ -33,7 +33,7 @@ delete." Its correct home is the action's own description, which is what
 - **Typed `inputs` + `reports`** (the issue's original) — rejected. Typed
   entries turn an action from a *description* into a *signature*, the clearest
   step toward an interface definition language. modelith is a communication
-  artifact, not an IDL; rules and shapes stay prose (see ADR-0005).
+  artifact, not an IDL; rules and shapes stay prose (see ADR-0007).
 - **Defer, keep prose in invariants** — rejected. The conditional-consequence
   argument above says the invariant is the wrong home even today; `reports:`
   is the small, honest fix.

project-docs/adr/0007-value-types-out-of-scope.md

@@ -0,0 +1,51 @@
+# Structured value types are out of scope
+
+Do not add a `types:` section or any structured value-type construct. A value
+type with named, typed fields defines record structure — implementation
+semantics modelith deliberately does not capture. Supersedes ADR-0005.
+
+## Context
+
+ADR-0005 decided to add a top-level `types:` section: named value types with
+`fields` (name + type) and prose `constraints`, resolvable from an attribute
+`type`, with `repeated: true` for collections. Nothing was built. On revisiting
+the shape before implementation, the construct read as too close to an
+interface definition language: a list of named, typed fields is a record
+declaration — "define the columns" — which is exactly the implementation
+detail the format is meant to abstract away.
+
+## Decision
+
+Structured value types are out of scope. modelith captures the *general shape of
+a system* — the concepts, their relationships, the rules that must hold — not
+the internal field layout of a value. A value-shaped concept is modeled the way
+the schema reference already prescribes: as an owned entity, or as attributes on
+its owner, with any internal rules stated as prose (a glossary entry or an
+invariant).
+
+## Why
+
+- **It crosses the IDL line.** The format's identity is a communication artifact,
+  not a schema language. `{amount: decimal, currency: Currency}` is a data
+  structure definition; that belongs in code or an IDL, not here (the same line
+  ADR-0006 held for action reports).
+- **The general shape is already expressible.** Owned entities and attributes
+  cover value-shaped concepts well enough to communicate intent. Where identity
+  is genuinely undeserved, the tension is named in prose (the parking-garage
+  `Ticket`), which is honest about the trade-off rather than papering over it.
+- **"For now," not "never."** If a real model demonstrates that prose plus an
+  owned entity genuinely cannot carry a needed distinction, revisit — but the
+  default stance is omission, and the bar to reopen is a concrete modeling
+  failure, not expressiveness for its own sake.
+
+## Consequences
+
+- The schema reference keeps value objects in "what this format deliberately
+  leaves out," reframed from "being explored" to a settled omission.
+- Issue #11 is closed as out of scope rather than implemented.
+- Issue #4 sub-item 4 (the `Ticket` value-object teaching moment) no longer
+  waits on value types; the current owned-entity-with-named-tension treatment
+  is the answer.
+- Nothing to pin with a test: the decision is the absence of a construct, and
+  the schema already has no `types:` section. The human-facing guard is the
+  "deliberately leaves out" section of the schema reference.