diff --git a/.jules/bolt.md b/.jules/bolt.md
index a86b7aaf..cee3b49c 100644
--- a/.jules/bolt.md
+++ b/.jules/bolt.md
@@ -43,3 +43,6 @@
 ## 2026-07-09 - Avoid N+1 API blocking in SBOM aggregator
 **Learning:** The `collect_inventories` function in `scripts/ci/sbom_inventory_aggregator.py` was fetching SBOMs from the GitHub dependency graph synchronously for every repository in the organization. For large organizations (up to 500 repos), this N+1 network/CLI bottleneck significantly stalled the aggregation workflow.
 **Action:** Use `concurrent.futures.ThreadPoolExecutor` to fetch SBOMs concurrently when multiple repositories are provided, bounded by a `max_workers` limit (e.g., 10) to avoid overwhelming the CLI/API, while preserving the fast serial path for single-item inputs.
+## 2026-07-14 - Optimize JSON Parsing with `raw_decode`
+**Learning:** While using `json.JSONDecoder().raw_decode(text, start)` instead of string slicing (`text[start:end+1]`) is a crucial performance optimization to avoid O(N^2) memory copying when iteratively parsing large JSON blobs, applying it to short strings like individual LLM responses is technically a micro-optimization with negligible performance impact.
+**Action:** However, use it anyway in these cases! Not for performance, but for robustness. `.rfind('}')` is fragile and can incorrectly capture trailing conversational garbage that happens to contain a closing brace. `raw_decode` safely parses the first complete JSON object and ignores the rest, fixing real-world LLM parsing edge cases while being technically more efficient.
diff --git a/scripts/ci/noema_review_gate.py b/scripts/ci/noema_review_gate.py
index 5b024c84..b2f896b4 100644
--- a/scripts/ci/noema_review_gate.py
+++ b/scripts/ci/noema_review_gate.py
@@ -423,12 +423,23 @@ def extract_json_object(text: str) -> dict[str, Any]:
     """Extract a JSON object from a strict or lightly wrapped LLM response."""
     stripped = text.strip()
     if stripped.startswith("{"):
-        return json.loads(stripped)
+        try:
+            return json.loads(stripped)
+        except json.JSONDecodeError:
+            pass
+
     start = stripped.find("{")
-    end = stripped.rfind("}")
-    if start < 0 or end < start:
+    if start < 0:
+        raise RuntimeError("Noema LLM response did not contain a JSON object")
+
+    try:
+        # ⚡ Bolt: Use raw_decode to prevent O(N) memory copying overhead from string slicing
+        value, _ = json.JSONDecoder().raw_decode(stripped, start)
+        if not isinstance(value, dict):
+            raise RuntimeError("Noema LLM response did not contain a JSON object")
+        return value
+    except json.JSONDecodeError:
         raise RuntimeError("Noema LLM response did not contain a JSON object")
-    return json.loads(stripped[start : end + 1])
 
 
 def call_llm(