diff --git a/examples/0192-types-size-of-qualified-alias-mod.sx b/examples/0192-types-size-of-qualified-alias-mod.sx
new file mode 100644
index 00000000..e5b5437f
--- /dev/null
+++ b/examples/0192-types-size-of-qualified-alias-mod.sx
@@ -0,0 +1,3 @@
+// Companion module for 0192 — exports `Selection`, a struct whose size is NOT 8
+// bytes (so a wrong fallback like `.i64` would be detectable).
+Selection :: struct { a: i32; b: i32; c: i32; }   // 12 bytes
diff --git a/examples/0192-types-size-of-qualified-alias.sx b/examples/0192-types-size-of-qualified-alias.sx
new file mode 100644
index 00000000..8a631836
--- /dev/null
+++ b/examples/0192-types-size-of-qualified-alias.sx
@@ -0,0 +1,18 @@
+// `size_of(alias.Type)` on a module-alias-qualified type must resolve the type,
+// exactly as a declaration `: alias.Type` does. In argument position the
+// qualified name parses as a field-access expression; resolveTypeArg now
+// reconstructs the dotted name and resolves it through the alias map instead of
+// returning `.unresolved`.
+//
+// Regression (issue 0147).
+#import "modules/std.sx";
+sel :: #import "0192-types-size-of-qualified-alias-mod.sx";
+
+main :: () -> i32 {
+    // Qualified through the module alias, in a type-arg slot.
+    print("size={}\n", size_of(sel.Selection));   // 12
+    // Same name also resolves in a declaration (already worked).
+    s : sel.Selection = .{ a = 1, b = 2, c = 3 };
+    print("sum={}\n", s.a + s.b + s.c);            // 6
+    return 0;
+}
diff --git a/examples/expected/0192-types-size-of-qualified-alias.exit b/examples/expected/0192-types-size-of-qualified-alias.exit
new file mode 100644
index 00000000..573541ac
--- /dev/null
+++ b/examples/expected/0192-types-size-of-qualified-alias.exit
@@ -0,0 +1 @@
+0
diff --git a/examples/expected/0192-types-size-of-qualified-alias.stderr b/examples/expected/0192-types-size-of-qualified-alias.stderr
new file mode 100644
index 00000000..8b137891
--- /dev/null
+++ b/examples/expected/0192-types-size-of-qualified-alias.stderr
@@ -0,0 +1 @@
+
diff --git a/examples/expected/0192-types-size-of-qualified-alias.stdout b/examples/expected/0192-types-size-of-qualified-alias.stdout
new file mode 100644
index 00000000..4ce09bd7
--- /dev/null
+++ b/examples/expected/0192-types-size-of-qualified-alias.stdout
@@ -0,0 +1,2 @@
+size=12
+sum=6
diff --git a/issues/0147-size-of-qualified-aliased-type-unresolved.md b/issues/0147-size-of-qualified-aliased-type-unresolved.md
new file mode 100644
index 00000000..cae3a732
--- /dev/null
+++ b/issues/0147-size-of-qualified-aliased-type-unresolved.md
@@ -0,0 +1,68 @@
+# 0147 — `size_of(alias.Type)` on a module-aliased type fails: "expects a type, got 'unresolved'"
+
+> **RESOLVED.** In a reflection-builtin argument slot, a module-alias-qualified
+> type (`sel.Selection`) parses as a `.field_access` expression (not the dotted
+> `.type_expr` a declaration annotation produces), and neither `isStaticTypeArg`
+> nor `resolveTypeArg` (src/ir/lower/generic.zig) had a `.field_access` arm — so
+> the reflection guard rejected it as a non-type. Both now handle the qualified
+> form: `isStaticTypeArg` recognizes `alias.Type` when `alias` is a namespace
+> alias whose target authors a type named `Type` (pure decl scan), and
+> `resolveTypeArg` resolves it via `namespaceAliasTarget` + `resolveNominalLeaf`
+> in the target module's context (the same mechanism `lowerFieldAccess` uses for
+> `alias.Type` in value position). Regression test:
+> `examples/0192-types-size-of-qualified-alias.sx`.
+
+## Summary
+`size_of(T)` does not resolve a type referenced through a module ALIAS
+(`alias :: #import "..."`), even though that same alias resolves fine everywhere
+else (declarations, casts, struct-literal construction). The compiler reports the
+qualified type as `unresolved` only inside `size_of`.
+
+## Repro
+```sx
+sel :: #import "doc/selection.sx";   // selection.sx exports `Selection`
+
+box :: () -> *sel.Selection {
+    // both of these fail:
+    p : *sel.Selection = xx context.allocator.alloc_bytes(size_of(sel.Selection));
+    memset(xx p, 0, size_of(sel.Selection));
+    p
+}
+```
+
+Error:
+```
+error: size_of expects a type, got 'unresolved'
+   |
+   |  ... alloc_bytes(size_of(sel.Selection));
+   |                          ^^^^^^^^^^^^^
+```
+
+Note the SAME `sel.Selection` resolves correctly in the variable declaration
+`p : *sel.Selection` and in calls like `sel.selection_create(...)` — only the
+`size_of(...)` argument position treats the qualified name as unresolved.
+
+## Expected
+`size_of(sel.Selection)` resolves the aliased type and yields its size, exactly
+as `size_of(Selection)` does for an unqualified/flat-imported type.
+
+## Workaround (clean)
+Introduce an unqualified local type alias and feed THAT to `size_of`:
+```sx
+sel :: #import "doc/selection.sx";
+Selection :: sel.Selection;   // unqualified alias
+
+box :: () -> *Selection {
+    p : *Selection = xx context.allocator.alloc_bytes(size_of(Selection));
+    memset(xx p, 0, size_of(Selection));
+    p
+}
+```
+`size_of(Selection)` (the unqualified alias) resolves fine. Used in
+photo `tests/toolbar.sx`'s `box_sel` (the selection model is imported qualified
+as `sel` there to avoid a `Point` collision with `modules/ui/types.sx`).
+
+## Impact
+Minor. Only bites when a type must be reached through a module alias AND its size
+is needed (heap-boxing a zeroed value of that type). The unqualified-alias
+workaround is a one-liner and reads clearly.
diff --git a/src/ir/lower/generic.zig b/src/ir/lower/generic.zig
index 12d8b545..f86e53f1 100644
--- a/src/ir/lower/generic.zig
+++ b/src/ir/lower/generic.zig
@@ -243,6 +243,35 @@ pub fn isStaticTypeArg(self: *Lowering, node: *const Node) bool {
             }
             return true;
         },
+        .field_access => |fa| {
+            // A module-alias-qualified type name (`sel.Selection`) is a static
+            // type iff `fa.object` is a namespace ALIAS (not a runtime scope var)
+            // whose target module authors a TYPE named `fa.field` (issue 0147).
+            // Pure predicate: scan the target's own decls — no type resolution
+            // side effects (the actual TypeId is produced later by
+            // `resolveTypeArg`'s matching `.field_access` arm).
+            if (fa.object.data != .identifier) return false;
+            const oname = fa.object.data.identifier.name;
+            if (self.scope) |scope| {
+                if (scope.lookup(oname) != null) return false;
+            }
+            const target = self.namespaceAliasTarget(oname, node.span) orelse return false;
+            for (target.own_decls) |decl| {
+                const dn = decl.data.declName() orelse continue;
+                if (!std.mem.eql(u8, dn, fa.field)) continue;
+                return switch (decl.data) {
+                    .struct_decl, .enum_decl, .union_decl, .error_set_decl => true,
+                    // A const-wrapped type definition or a type alias
+                    // (`Foo :: Bar;` / `Foo :: ns.Bar;`).
+                    .const_decl => |cd| switch (cd.value.data) {
+                        .struct_decl, .enum_decl, .union_decl, .error_set_decl, .identifier, .field_access => true,
+                        else => false,
+                    },
+                    else => false,
+                };
+            }
+            return false;
+        },
         .pack_index_type_expr,
         .pointer_type_expr,
         .many_pointer_type_expr,
@@ -438,6 +467,36 @@ pub fn resolveTypeArg(self: *Lowering, node: *const Node) TypeId {
         .optional_type_expr,
         .function_type_expr,
         => return self.resolveTypeWithBindings(node),
+        // A module-alias-qualified type name in a type-arg slot
+        // (`size_of(sel.Selection)`) parses as a field-access EXPRESSION — unlike
+        // the dotted `.type_expr` a declaration annotation produces — so without
+        // this arm it fell through to `else` and resolved to `.unresolved`
+        // (issue 0147). Reconstruct the qualified `obj.field` name and resolve it
+        // through the same alias map a declaration uses. Look it up EXPLICITLY
+        // (findByName + alias map) rather than via `resolveNamed`, whose
+        // empty-struct-stub fallback would silently fabricate a 0-sized type for
+        // an unregistered name (the silent-default trap) — a failed lookup must
+        // surface as a diagnostic + `.unresolved`.
+        .field_access => |fa| {
+            // Resolve the member as a TYPE in the alias's TARGET module context —
+            // the same mechanism `lowerFieldAccess` uses for `alias.Type` in value
+            // position (src/ir/lower/expr.zig): the alias edge authorizes the reach,
+            // so set the current source to the target module and resolve the bare
+            // member name through the source-aware nominal leaf.
+            if (fa.object.data == .identifier) {
+                if (self.namespaceAliasTarget(fa.object.data.identifier.name, node.span)) |target| {
+                    const saved_src = self.current_source_file;
+                    self.setCurrentSourceFile(target.target_module_path);
+                    const ty = self.resolveNominalLeaf(fa.field, false, node.span);
+                    self.setCurrentSourceFile(saved_src);
+                    if (ty != .unresolved) return ty;
+                }
+            }
+            if (self.diagnostics) |diags| {
+                diags.addFmt(.err, node.span, "unresolved qualified type in type-argument position", .{});
+            }
+            return .unresolved;
+        },
         else => return .unresolved,
     }
 }