diff --git a/src/ai_client.py b/src/ai_client.py
index 6fd05ad5..4e89ded5 100644
--- a/src/ai_client.py
+++ b/src/ai_client.py
@@ -1635,6 +1635,35 @@ def _delete_gemini_cache_result() -> Result[None]:
    errors=[ErrorInfo(kind=ErrorKind.INTERNAL, message=f"failed to delete gemini cache: {e}", source="ai_client._delete_gemini_cache_result", original=e)],
   )
 
+_GEMINI_CACHE_TOKEN_THRESHOLD: int = 2048
+
+def _should_cache_gemini_result(sys_instr: str) -> Result[bool]:
+ """Decide whether the current Gemini context warrants caching.
+
+ Returns Result(data=True) if token count >= 2048, Result(data=False)
+ if below threshold (with a [CACHING SKIPPED] comms note), or
+ Result(data=False, errors=[ErrorInfo]) on SDK failure.
+
+ The caller (_send_gemini) ignores errors and treats failure as
+ 'do not cache' (safe default: cache create is expensive; skipping
+ on count failure is a soft fallback to inline system_instruction).
+ """
+ if _gemini_client is None:
+  return Result(data=False)
+ try:
+  count_resp = _gemini_client.models.count_tokens(model=_model, contents=[sys_instr])
+  total      = count_resp.total_tokens
+  if total and total >= _GEMINI_CACHE_TOKEN_THRESHOLD:
+   return Result(data=True)
+  _append_comms("OUT", "request", {"message": f"[CACHING SKIPPED] Context too small ({total} tokens < {_GEMINI_CACHE_TOKEN_THRESHOLD})"})
+  return Result(data=False)
+ except Exception as e:
+  _append_comms("OUT", "request", {"message": f"[COUNT FAILED] {e}"})
+  return Result(
+   data=False,
+   errors=[ErrorInfo(kind=ErrorKind.INTERNAL, message=f"failed to count gemini tokens: {e}", source="ai_client._should_cache_gemini_result", original=e)],
+  )
+
 def _extract_gemini_thoughts(resp: Any) -> str:
  """
   Extracts concatenated thinking text from a Gemini response object's parts.
@@ -1721,16 +1750,7 @@ def _send_gemini(md_content: str, user_message: str, base_dir: str,
     safety_settings    = [types.SafetySetting(category=types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT, threshold=types.HarmBlockThreshold.BLOCK_ONLY_HIGH)]
    )
    
-   should_cache = False
-   try:
-    if _gemini_client:
-     count_resp = _gemini_client.models.count_tokens(model=_model, contents=[sys_instr])
-     if count_resp.total_tokens and count_resp.total_tokens >= 2048:
-      should_cache = True
-     else:
-      _append_comms("OUT", "request", {"message": f"[CACHING SKIPPED] Context too small ({count_resp.total_tokens} tokens < 2048)"})
-   except Exception as e:
-    _append_comms("OUT", "request", {"message": f"[COUNT FAILED] {e}"})
+   should_cache = _should_cache_gemini_result(sys_instr).data
    if should_cache and _gemini_client:
     try:
      _gemini_cache = _gemini_client.caches.create(
diff --git a/tests/tier2/phase10_site4_test.py b/tests/tier2/phase10_site4_test.py
new file mode 100644
index 00000000..7d152a2b
--- /dev/null
+++ b/tests/tier2/phase10_site4_test.py
@@ -0,0 +1,18 @@
+"""Phase 10 site 4: _should_cache_gemini_result helper."""
+import sys
+sys.path.insert(0, ".")
+
+
+def test_phase10_site4_should_cache_gemini_result_exists():
+    import src.ai_client
+    assert hasattr(src.ai_client, "_should_cache_gemini_result"), \
+        "_should_cache_gemini_result helper missing"
+
+
+def test_phase10_site4_should_cache_gemini_result_returns_result():
+    import src.ai_client
+    import inspect
+    fn = src.ai_client._should_cache_gemini_result
+    sig = inspect.signature(fn)
+    assert "Result" in str(sig.return_annotation), \
+        f"_should_cache_gemini_result return must be Result, got {sig.return_annotation}"
\ No newline at end of file