diff --git a/.jules/bolt.md b/.jules/bolt.md
index 2389b4f..8956cf3 100644
--- a/.jules/bolt.md
+++ b/.jules/bolt.md
@@ -40,3 +40,7 @@
 ## 2026-06-30 - Regex over Generator `any` string loops
 **Learning:** Using `any(...)` with a generator comprehension in string evaluation paths allocates a new generator and adds Python-level loop overhead for every character.
 **Action:** Replace `any()` generators with a pre-compiled regex (`re.compile().search()`) to evaluate string patterns in C, achieving a ~7x speedup for text-heavy operations.
+
+## 2024-07-01 - Avoid List Evaluation on Glob Matches
+**Learning:** Calling `list()` on a generator like `path.glob("*/ocr")` iterates over all matches and allocates a list, even if we only need the first element. In our profiling of file discovery routines, eagerly resolving a glob generator caused unnecessary directory traversals and list allocations.
+**Action:** Use `next(path.glob(...))` with a `try/except StopIteration` block when looking for the first matching directory to avoid evaluating the entire glob iterator.
diff --git a/src/newsdom_api/mineru_runner.py b/src/newsdom_api/mineru_runner.py
index 2ad987b..74e862e 100644
--- a/src/newsdom_api/mineru_runner.py
+++ b/src/newsdom_api/mineru_runner.py
@@ -78,10 +78,11 @@ def _resolve_mineru_bin() -> str:
 def _find_output_dir(base_output_dir: Path) -> Path:
     """Locate the OCR output directory created by MinerU."""
 
-    candidates = list(base_output_dir.glob("*/ocr"))
-    if not candidates:
+    try:
+        # ⚡ Bolt: Use next() to avoid evaluating the entire glob iterator into a list
+        return next(base_output_dir.glob("*/ocr"))
+    except StopIteration:
         raise FileNotFoundError("MinerU OCR output directory was not produced")
-    return candidates[0]
 
 
 def _execute_mineru(cmd: list[str]) -> subprocess.CompletedProcess[str]: