diff --git a/CHANGELOG.md b/CHANGELOG.md
index 8e876ca..560bb23 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,3 +1,10 @@
+## Unreleased
+
+- Add statically-typed `scale()` to `Matrix64`, `Matrix32`, `Vector64`, and
+  `Vector32`, and explicit axis broadcasting (`addVector`, `subtractVector`,
+  `multiplyVector`, `divideVector`) on the matrix types to give a fully typed
+  multiply path and remove the square-matrix broadcast ambiguity.
+
 ## 2.0.0
 
 - Redesign the package around high performance and script-friendly ergonomics,
diff --git a/lib/src/matrix.dart b/lib/src/matrix.dart
index 8e7883c..c475e5d 100644
--- a/lib/src/matrix.dart
+++ b/lib/src/matrix.dart
@@ -653,6 +653,29 @@ class Matrix64 {
 
   Matrix64 hadamard(Object other) => _binary(other, ElementwiseOp.multiply);
 
+  /// Scalar multiply that preserves the static [Matrix64] type.
+  ///
+  /// `operator *` returns `dynamic` because it also dispatches to
+  /// matrix/vector products; use [scale] when the result must stay
+  /// statically typed for chaining.
+  Matrix64 scale(num factor) => _binary(factor, ElementwiseOp.multiply);
+
+  /// Add [vector] to every line along [axis] with an explicit
+  /// orientation. `Axis.columns` expects `vector.length == columns`,
+  /// `Axis.rows` expects `vector.length == rows` — removing the
+  /// square-matrix ambiguity of the broadcasting operators.
+  Matrix64 addVector(Vector64 vector, Axis axis) =>
+      _broadcastAxis(vector, axis, ElementwiseOp.add);
+
+  Matrix64 subtractVector(Vector64 vector, Axis axis) =>
+      _broadcastAxis(vector, axis, ElementwiseOp.subtract);
+
+  Matrix64 multiplyVector(Vector64 vector, Axis axis) =>
+      _broadcastAxis(vector, axis, ElementwiseOp.multiply);
+
+  Matrix64 divideVector(Vector64 vector, Axis axis) =>
+      _broadcastAxis(vector, axis, ElementwiseOp.divide);
+
   Matrix64 matmul(Matrix64 other) {
     if (columns != other.rows) {
       throw ArgumentError(
@@ -941,6 +964,30 @@ class Matrix64 {
     );
   }
 
+  Matrix64 _broadcastAxis(Vector64 vector, Axis axis, ElementwiseOp op) {
+    final expected = axis == Axis.columns ? columns : rows;
+    if (vector.length != expected) {
+      throw ArgumentError(
+        'Vector64 length ${vector.length} must equal '
+        '${axis == Axis.columns ? 'columns $columns' : 'rows $rows'} '
+        'to broadcast along $axis.',
+      );
+    }
+    final vectorData = vector.unsafeValuesView;
+    final result = Float64List(count);
+    for (var row = 0; row < rows; row++) {
+      final offset = row * columns;
+      for (var column = 0; column < columns; column++) {
+        result[offset + column] = _apply(
+          _data[offset + column],
+          vectorData[axis == Axis.columns ? column : row],
+          op,
+        );
+      }
+    }
+    return Matrix64.storage(rows, columns, result, copy: false);
+  }
+
   Matrix64 _broadcastVector(Vector64 vector, ElementwiseOp op) {
     final vectorData = vector.values;
     final result = Float64List(count);
diff --git a/lib/src/matrix32.dart b/lib/src/matrix32.dart
index 9184db5..bb484ed 100644
--- a/lib/src/matrix32.dart
+++ b/lib/src/matrix32.dart
@@ -396,6 +396,29 @@ class Matrix32 {
 
   Matrix32 hadamard(Object other) => _binary(other, ElementwiseOp.multiply);
 
+  /// Scalar multiply that preserves the static [Matrix32] type.
+  ///
+  /// `operator *` returns `dynamic` because it also dispatches to
+  /// matrix/vector products; use [scale] when the result must stay
+  /// statically typed for chaining.
+  Matrix32 scale(num factor) => _binary(factor, ElementwiseOp.multiply);
+
+  /// Add [vector] to every line along [axis] with an explicit
+  /// orientation. `Axis.columns` expects `vector.length == columns`,
+  /// `Axis.rows` expects `vector.length == rows` — removing the
+  /// square-matrix ambiguity of the broadcasting operators.
+  Matrix32 addVector(Vector32 vector, Axis axis) =>
+      _broadcastAxis32(vector, axis, ElementwiseOp.add);
+
+  Matrix32 subtractVector(Vector32 vector, Axis axis) =>
+      _broadcastAxis32(vector, axis, ElementwiseOp.subtract);
+
+  Matrix32 multiplyVector(Vector32 vector, Axis axis) =>
+      _broadcastAxis32(vector, axis, ElementwiseOp.multiply);
+
+  Matrix32 divideVector(Vector32 vector, Axis axis) =>
+      _broadcastAxis32(vector, axis, ElementwiseOp.divide);
+
   Matrix32 vstack(Matrix32 other) =>
       Matrix32.fromMatrix(toFloat64().vstack(other.toFloat64()));
 
@@ -719,6 +742,30 @@ class Matrix32 {
     );
   }
 
+  Matrix32 _broadcastAxis32(Vector32 vector, Axis axis, ElementwiseOp op) {
+    final expected = axis == Axis.columns ? columns : rows;
+    if (vector.length != expected) {
+      throw ArgumentError(
+        'Vector32 length ${vector.length} must equal '
+        '${axis == Axis.columns ? 'columns $columns' : 'rows $rows'} '
+        'to broadcast along $axis.',
+      );
+    }
+    final vectorData = vector.unsafeValuesView;
+    final result = Float32List(count);
+    for (var row = 0; row < rows; row++) {
+      final offset = row * columns;
+      for (var column = 0; column < columns; column++) {
+        result[offset + column] = _apply(
+          _data[offset + column],
+          vectorData[axis == Axis.columns ? column : row],
+          op,
+        );
+      }
+    }
+    return Matrix32.storage(rows, columns, result, copy: false);
+  }
+
   Matrix32 _broadcastVector32(Vector32 vector, ElementwiseOp op) {
     final vectorData = vector.unsafeValuesView;
     final result = Float32List(count);
diff --git a/lib/src/vector.dart b/lib/src/vector.dart
index 03c8238..aae7c0a 100644
--- a/lib/src/vector.dart
+++ b/lib/src/vector.dart
@@ -108,6 +108,14 @@ class Vector64 extends IterableBase<double> {
 
   dynamic operator /(Object other) => _binary(other, ElementwiseOp.divide);
 
+  /// Scalar multiply that preserves the static [Vector64] type.
+  ///
+  /// `operator *` returns `dynamic` because it also dispatches to the
+  /// vector/matrix product; use [scale] when the result must stay
+  /// statically typed for chaining.
+  Vector64 scale(num factor) =>
+      Vector64.storage(Kernels.scale(_data, factor.toDouble()), copy: false);
+
   double dot(Vector64 other) {
     _checkLength(other);
     return Kernels.dot(_data, 0, other._data, 0, length);
diff --git a/lib/src/vector32.dart b/lib/src/vector32.dart
index 939db28..e0df39c 100644
--- a/lib/src/vector32.dart
+++ b/lib/src/vector32.dart
@@ -80,6 +80,14 @@ class Vector32 extends IterableBase<double> {
 
   dynamic operator /(Object other) => _binary(other, ElementwiseOp.divide);
 
+  /// Scalar multiply that preserves the static [Vector32] type.
+  ///
+  /// `operator *` returns `dynamic` because it also dispatches to the
+  /// vector/matrix product; use [scale] when the result must stay
+  /// statically typed for chaining.
+  Vector32 scale(num factor) =>
+      Vector32.storage(Kernels.scale32(_data, factor.toDouble()), copy: false);
+
   double dot(Vector32 other) {
     _checkLength(other);
     return Kernels.dot32(_data, 0, other._data, 0, length);
diff --git a/test/matrix_api_test.dart b/test/matrix_api_test.dart
index bdb8649..5adbcf2 100644
--- a/test/matrix_api_test.dart
+++ b/test/matrix_api_test.dart
@@ -122,4 +122,67 @@ void main() {
       throwsArgumentError,
     );
   });
+
+  test('typed arithmetic methods preserve static types', () {
+    final Matrix64 a = mat([
+      [1, 2, 3],
+      [4, 5, 6],
+    ]);
+
+    final Matrix64 scaled = a.scale(2);
+    expect(scaled.transpose.toRows(), [
+      [2, 8],
+      [4, 10],
+      [6, 12],
+    ]);
+
+    final Vector64 x = vec([1, 2, 3]);
+    final Vector64 doubled = x.scale(2);
+    expect(doubled.toList(), [2, 4, 6]);
+
+    final Matrix32 a32 = mat32([
+      [1, 2],
+      [3, 4],
+    ]);
+    expect(a32.scale(3).transpose.toRows(), [
+      [3, 9],
+      [6, 12],
+    ]);
+    expect(vec32([1, 2]).scale(4).toList(), [4, 8]);
+  });
+
+  test('explicit axis broadcasting removes square-matrix ambiguity', () {
+    final a = mat([
+      [1, 2],
+      [3, 4],
+    ]);
+    final v = vec([10, 20]);
+
+    expect(a.addVector(v, Axis.columns).toRows(), [
+      [11, 22],
+      [13, 24],
+    ]);
+    expect(a.addVector(v, Axis.rows).toRows(), [
+      [11, 12],
+      [23, 24],
+    ]);
+    expect(a.subtractVector(v, Axis.columns).toRows(), [
+      [-9, -18],
+      [-7, -16],
+    ]);
+
+    final a32 = mat32([
+      [1, 2],
+      [3, 4],
+    ]);
+    expect(a32.multiplyVector(vec32([2, 3]), Axis.rows).toRows(), [
+      [2, 4],
+      [9, 12],
+    ]);
+
+    expect(
+      () => a.addVector(vec([1, 2, 3]), Axis.columns),
+      throwsArgumentError,
+    );
+  });
 }