diff --git a/test-basic-module-cpp/src/test_basic_module_cpp_impl.cpp b/test-basic-module-cpp/src/test_basic_module_cpp_impl.cpp
index abf8fee..585a898 100644
--- a/test-basic-module-cpp/src/test_basic_module_cpp_impl.cpp
+++ b/test-basic-module-cpp/src/test_basic_module_cpp_impl.cpp
@@ -21,12 +21,35 @@ bool TestBasicModuleCppImpl::returnTrue()  { return true;  }
 bool TestBasicModuleCppImpl::returnFalse() { return false; }
 bool TestBasicModuleCppImpl::isPositive(int64_t value) { return value > 0; }
 
+namespace {
+
+// A character is a Unicode CODE POINT, so count the bytes that START one: in
+// UTF-8 every continuation byte matches 10xxxxxx. Identical to the helper in
+// test_basic_module — these two modules are deliberate mirrors and must agree
+// on non-ASCII, which they did not: this one answered in BYTES while its twin
+// answered in characters.
+inline bool isContinuationByte(char c)
+{
+    return (static_cast<unsigned char>(c) & 0xC0) == 0x80;
+}
+
+int64_t characterCount(const std::string& s)
+{
+    int64_t n = 0;
+    for (char c : s) {
+        if (!isContinuationByte(c)) ++n;
+    }
+    return n;
+}
+
+} // namespace
+
 // ── int64_t ──────────────────────────────────────────────────────────────
 
 int64_t TestBasicModuleCppImpl::returnInt() { return 42; }
 int64_t TestBasicModuleCppImpl::addInts(int64_t a, int64_t b) { return a + b; }
 int64_t TestBasicModuleCppImpl::stringLength(const std::string& s) {
-    return static_cast<int64_t>(s.size());
+    return characterCount(s);
 }
 
 // ── uint64_t ─────────────────────────────────────────────────────────────
@@ -81,7 +104,7 @@ StdLogosResult TestBasicModuleCppImpl::validateInput(const std::string& input) {
     }
     nlohmann::json data;
     data["input"]  = input;
-    data["length"] = static_cast<int64_t>(input.size());
+    data["length"] = characterCount(input);
     return {true, data, ""};
 }
 
diff --git a/tests/run_tests.sh b/tests/run_tests.sh
index 2423f5e..e1eeb53 100755
--- a/tests/run_tests.sh
+++ b/tests/run_tests.sh
@@ -508,6 +508,14 @@ test_basic_cpp "addInts(3, 4)"      "Result: 7"   "test_basic_module_cpp.addInts
 test_basic_cpp "addInts(0, 0)"      "Result: 0"   "test_basic_module_cpp.addInts(0, 0)"
 test_basic_cpp "addInts(-5, 10)"    "Result: 5"   "test_basic_module_cpp.addInts(-5, 10)"
 test_basic_cpp "stringLength(hello)" "Result: 5"  "test_basic_module_cpp.stringLength(hello)"
+# The two modules are deliberate MIRRORS, so they must agree on non-ASCII. They
+# did not: this one answered in BYTES (6) while test_basic_module answered in
+# CHARACTERS (5). The blind spot was that this group only ever asserted ASCII,
+# where bytes and characters coincide — so the disagreement was invisible.
+test_basic_cpp "stringLength(héllo) [characters, mirrors test_basic_module]" "Result: 5" \
+    "test_basic_module_cpp.stringLength(héllo)"
+test_basic_cpp "stringLength(😀) [non-BMP character counts 1]" "Result: 1" \
+    "test_basic_module_cpp.stringLength(😀)"
 skip_test      "stringLength()"     "logoscore cannot call 1-arg method with 0 args"
 
 # ── Return type: uint64_t (unique to the C++ surface) ───────────────────────