diff --git a/changes/unreleased/typed-by-sweep.fixed.md b/changes/unreleased/typed-by-sweep.fixed.md
new file mode 100644
index 0000000000..bacbc4271f
--- /dev/null
+++ b/changes/unreleased/typed-by-sweep.fixed.md
@@ -0,0 +1 @@
+- The nonstandard-notation pass no longer walks every expression of a document looking for a `typed by` on a body-expression parameter (#1478) when the document spells none: the parser counts the spellings and the pass sweeps only while one is unreported. Loading a generated 1 000-element model allocates 58.5 MB instead of 61.2 MB, recovering most of what #1478 added over 0.11.0's 58.2 MB on the same toolchain; the remaining 0.3 MB is other work.
diff --git a/internal/check/passes/nonstandard_notation.go b/internal/check/passes/nonstandard_notation.go
index b0b816b45a..ca5925b759 100644
--- a/internal/check/passes/nonstandard_notation.go
+++ b/internal/check/passes/nonstandard_notation.go
@@ -63,7 +63,12 @@ func (NonstandardNotationPass) Run(ctx *Context, name string, root *ast.RootName
 	w.bodies = append(w.bodies, bodyFrame{members: root.Members, namespaceish: true})
 	w.walk(root.Members)
 	w.bodies = w.bodies[:0]
-	w.typedByInExpressions(root)
+	// The member walk reaches every `typed by` but a body expression's; the
+	// document counts its spellings, so the expression sweep runs only when
+	// one is left to find.
+	if w.typedByReported < root.TypedBy {
+		w.typedByInExpressions(root)
+	}
 	// A comment where no member may start is the parser's own warning, which
 	// strict conformance escalates; the escalated finding replaces the warning.
 	if ctx.Options.Conformance.IsStrict() {
@@ -100,6 +105,8 @@ type notationWalker struct {
 	// inKerMLDeclaration counts the enclosing declarations already reported as
 	// KerML notation; their members move with them, so are not reported again.
 	inKerMLDeclaration int
+	// typedByReported counts the `typed by` findings, against RootNamespace.TypedBy.
+	typedByReported int
 	// keywordName holds the offsets where the parser recovered a keyword written as
 	// a name, the only spans keywordAsName escalates.
 	keywordName map[int]bool
@@ -765,6 +772,7 @@ func (w *notationWalker) typedByRelationships(rels []*ast.Relationship) {
 // typedBy reports a typing spelled `typed by` in a SysML file (SysML.xtext:395 admits
 // `:` and `defined by`; KerML.xtext:600 `typed by`); the fix rewrites the keyword alone.
 func (w *notationWalker) typedBy(span source.Span) {
+	w.typedByReported++
 	w.diags = append(w.diags, diag.Diagnostic{
 		Severity: w.severity,
 		Span:     span,
diff --git a/internal/syntax/ast/astcodec/nodes.go b/internal/syntax/ast/astcodec/nodes.go
index 953d94cb12..7e21d77bca 100644
--- a/internal/syntax/ast/astcodec/nodes.go
+++ b/internal/syntax/ast/astcodec/nodes.go
@@ -1000,6 +1000,7 @@ func (e *Encoder) encodeFields(node ast.Node) {
 		e.base(&n.NodeBase)
 		e.nodes(n.Members)
 		e.operatorExprs(n.UndefinedOperators)
+		e.w.Int(int64(n.TypedBy))
 	case *ast.SelectExpr:
 		e.base(&n.NodeBase)
 		e.node(n.Operand)
@@ -1514,6 +1515,7 @@ func (d *Decoder) decodeFields(node ast.Node) {
 		d.base(&n.NodeBase)
 		n.Members = d.nodes()
 		n.UndefinedOperators = d.operatorExprs()
+		n.TypedBy = int(d.r.Int())
 	case *ast.SelectExpr:
 		d.base(&n.NodeBase)
 		n.Operand = d.node()
diff --git a/internal/syntax/ast/namespace.go b/internal/syntax/ast/namespace.go
index 02e4ca08ea..c88608ed01 100644
--- a/internal/syntax/ast/namespace.go
+++ b/internal/syntax/ast/namespace.go
@@ -266,6 +266,9 @@ type RootNamespace struct {
 	// UndefinedOperators are the `~` operator expressions of the document in
 	// source order: KerML leaves DataFunctions::'~' undefined, so a tool warns at each.
 	UndefinedOperators []*OperatorExpr
+	// TypedBy counts the typings the document spells `typed by` (KerML.xtext:600),
+	// on declarations, prefix metadata and body-expression parameters alike.
+	TypedBy int
 }
 
 // PrefixMetadata records a `# QualifiedName` metadata annotation reference.
diff --git a/internal/syntax/parser/defusage.go b/internal/syntax/parser/defusage.go
index 0135bec8d6..a4aaa4ef86 100644
--- a/internal/syntax/parser/defusage.go
+++ b/internal/syntax/parser/defusage.go
@@ -4260,6 +4260,7 @@ func (p *Parser) typedBySpan(clauseTok lexer.Token) source.Span {
 	if clauseTok.Kind != lexer.Keyword || clauseTok.KeywordID != "typed" {
 		return source.Span{}
 	}
+	p.typedByCount++
 	return p.spanFrom(clauseTok.Span.Offset)
 }
 
diff --git a/internal/syntax/parser/parser.go b/internal/syntax/parser/parser.go
index 8d12b1fcbb..c27eba6947 100644
--- a/internal/syntax/parser/parser.go
+++ b/internal/syntax/parser/parser.go
@@ -43,6 +43,8 @@ type Parser struct {
 	// undefinedOps are the `~` operator expressions the parse built, in source
 	// order; ParseFile hands them to the root.
 	undefinedOps []*ast.OperatorExpr
+	// typedByCount counts the typings spelled `typed by`, for RootNamespace.TypedBy.
+	typedByCount int
 
 	// calcBodyDepth counts the calculation bodies being parsed, so a `return`
 	// reached in a statement position inside one is read as the result
@@ -172,6 +174,7 @@ type parseCheckpoint struct {
 	diagnosticLen  int
 	warningLen     int
 	undefinedOpLen int
+	typedByCount   int
 	memberStartLen int
 	commentBodyLen int
 	attachedLen    int
@@ -599,6 +602,7 @@ func (p *Parser) ParseFile() *ast.RootNamespace {
 	}
 	root.NodeSpan = p.spanFrom(start)
 	root.UndefinedOperators = p.undefinedOps
+	root.TypedBy = p.typedByCount
 	p.reportMisplacedComments()
 	return root
 }
@@ -661,6 +665,7 @@ func (p *Parser) checkpoint() parseCheckpoint {
 		diagnosticLen:  len(p.Diagnostics),
 		warningLen:     len(p.Warnings),
 		undefinedOpLen: len(p.undefinedOps),
+		typedByCount:   p.typedByCount,
 		memberStartLen: len(p.memberStarts),
 		commentBodyLen: len(p.commentBodies),
 		attachedLen:    len(p.attachedStarts),
@@ -682,6 +687,7 @@ func (p *Parser) restore(cp parseCheckpoint) {
 	p.Diagnostics = p.Diagnostics[:cp.diagnosticLen]
 	p.Warnings = p.Warnings[:cp.warningLen]
 	p.undefinedOps = p.undefinedOps[:cp.undefinedOpLen]
+	p.typedByCount = cp.typedByCount
 	p.memberStarts = p.memberStarts[:cp.memberStartLen]
 	p.commentBodies = p.commentBodies[:cp.commentBodyLen]
 	p.attachedStarts = p.attachedStarts[:cp.attachedLen]
diff --git a/internal/syntax/parser/typed_by_count_test.go b/internal/syntax/parser/typed_by_count_test.go
new file mode 100644
index 0000000000..09bd020009
--- /dev/null
+++ b/internal/syntax/parser/typed_by_count_test.go
@@ -0,0 +1,36 @@
+package parser
+
+import (
+	"testing"
+
+	"github.com/Open-MBEE/OpenSysML/internal/syntax/source"
+)
+
+// The document counts its `typed by` spellings (RootNamespace.TypedBy), on
+// declarations, prefix metadata and body-expression parameters alike, so a
+// pass that reports the spelling knows when it has found them all.
+func TestRootNamespaceCountsTypedBy(t *testing.T) {
+	tests := []struct {
+		name string
+		src  string
+		want int
+	}{
+		{"none", "package P { part def A; part a : A; }", 0},
+		{"typed_as_name", "package P { attribute typed; }", 0},
+		{"declaration", "package P { part def A; part a typed by A; }", 1},
+		{"two_declarations", "package P { part def A; part a typed by A; part b typed by A, A; }", 2},
+		{"prefix_metadata", "package P { metadata def M; @ m typed by M; }", 1},
+		{"body_parameter", "package P { part def A; calc c { in x; return y = x.{ in v typed by A; v }; } }", 1},
+		{"end_without_cross_feature", "package P { part def A; connection def C { end x typed by A; end y typed by A; } }", 2},
+		{"kerml_feature", "package P { classifier A; feature f typed by A; }", 1},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			p := New(source.New(tt.name+".sysml", []byte(tt.src)))
+			root := p.ParseFile()
+			if root.TypedBy != tt.want {
+				t.Errorf("RootNamespace.TypedBy = %d, want %d", root.TypedBy, tt.want)
+			}
+		})
+	}
+}
diff --git a/internal/workspace/libs/snapshot.go b/internal/workspace/libs/snapshot.go
index 122f5eef87..8da75479d6 100644
--- a/internal/workspace/libs/snapshot.go
+++ b/internal/workspace/libs/snapshot.go
@@ -28,7 +28,7 @@ var stdlibSnapshot []byte
 // whenever the stream layout does, so a stale blob is refused, not misread.
 const (
 	snapshotMagic         = "OpenSysML library snapshot\n"
-	snapshotFormatVersion = 26
+	snapshotFormatVersion = 27
 )
 
 // snapshotCRC checksums the index stream, so a damaged byte that still parses
diff --git a/internal/workspace/libs/snapshot_test.go b/internal/workspace/libs/snapshot_test.go
index d5c72de10a..2d40b80f76 100644
--- a/internal/workspace/libs/snapshot_test.go
+++ b/internal/workspace/libs/snapshot_test.go
@@ -79,7 +79,7 @@ func TestSnapshotIndexMatchesFreshLoad(t *testing.T) {
 
 // snapshotKindDigest pins the SymbolKind and PseudostateKind numbering the snapshot
 // stream persists under snapshotFormatVersion; a kind added or moved renumbers the ones after it.
-const snapshotKindDigest = "19f8c6091c4acf8b"
+const snapshotKindDigest = "e7b3ced9fc89d152"
 
 func TestSnapshotFormatVersionPinsSymbolKinds(t *testing.T) {
 	var b strings.Builder
diff --git a/internal/workspace/libs/stdlib.snapshot b/internal/workspace/libs/stdlib.snapshot
index 868bcaa306..bfccadac37 100644
Binary files a/internal/workspace/libs/stdlib.snapshot and b/internal/workspace/libs/stdlib.snapshot differ