diff --git a/.github/workflows/pr.yml b/.github/workflows/pr.yml
index 5871409..e88d479 100644
--- a/.github/workflows/pr.yml
+++ b/.github/workflows/pr.yml
@@ -21,6 +21,7 @@ jobs:
       is-bot: ${{ steps.check-author.outputs.is_bot }}
       is-cfc: ${{ steps.check-cfc.outputs.is_cfc }}
       is-page: ${{ steps.check-page.outputs.is_page }}
+      is-skip: ${{ steps.check-skip.outputs.is_skip }}
     steps:
       - name: Checkout
         uses: actions/checkout@v3
@@ -57,13 +58,23 @@ jobs:
             fi
           done
 
+      - name: Check if action is label but it is unrelated
+        id: check-skip
+        run: |
+          if [[ '${{ github.event.action }}' == 'labeled' ]]; then
+            if [[ '${{ github.event.label.name }}' != 'conformance-file' ]]; then
+              echo "is_skip=true" >> $GITHUB_OUTPUT
+            fi
+          fi
+
   cfc:
     needs: changes
     if: |
       github.event_name == 'pull_request' &&
       contains(github.event.pull_request.labels.*.name, 'conformance-file') &&
       needs.changes.outputs.is-cfc == 'true' &&
-      needs.changes.outputs.is-bot != 'true'
+      needs.changes.outputs.is-bot != 'true' &&
+      needs.changes.outputs.is-skip != 'true'
     uses: ./.github/workflows/cfc.yml
     secrets: inherit
 
@@ -72,7 +83,8 @@ jobs:
     if: |
       always() &&
       (needs.cfc.result == 'success' || needs.cfc.result == 'skipped') &&
-      needs.cfc.outputs.created != 'true'
+      needs.cfc.outputs.created != 'true' &&
+      needs.changes.outputs.is-skip != 'true'
     uses: ./.github/workflows/test.yml
     with:
       test-page: ${{ github.event_name != 'pull_request' || needs.changes.outputs.is-page == 'true' }}
diff --git a/data/schemas/gpac-extension.schema.json b/data/schemas/gpac-extension.schema.json
index 2158354..75a307c 100644
--- a/data/schemas/gpac-extension.schema.json
+++ b/data/schemas/gpac-extension.schema.json
@@ -2,6 +2,20 @@
   "$schema": "https://json-schema.org/draft/2020-12/schema",
   "title": "GPAC MP4Box extension declaration schema",
   "$defs": {
+    "box_descendant": {
+      "allOf": [
+        {
+          "type": "object",
+          "properties": {
+            "@Type": {
+              "type": "string"
+            }
+          },
+          "required": ["@Type"]
+        },
+        { "$ref": "#/$defs/box" }
+      ]
+    },
     "box": {
       "type": "object",
       "properties": {
@@ -30,19 +44,23 @@
           "description": "Combined flags in base 10"
         },
         "@Specification": { "type": "string" },
-        "@Container": { "type": "string" },
-        "@data": {
-          "type": "string",
-          "description": "dump of the box in hexadecimal format",
-          "pattern": "^0x[0-9a-fA-F]+$"
-        },
-        "children": {
-          "type": "array",
-          "items": { "$ref": "#/$defs/box" }
+        "@Container": { "type": "string" }
+      },
+      "patternProperties": {
+        "^(?!@).+": {
+          "anyOf": [
+            {
+              "type": "array",
+              "items": {
+                "$ref": "#/$defs/box_descendant"
+              }
+            },
+            { "$ref": "#/$defs/box_descendant" }
+          ]
         }
       },
       "additionalProperties": true,
-      "required": ["@Type", "@data"]
+      "required": ["@Type"]
     }
   },
   "type": "object",
diff --git a/src/construct/coverage.py b/src/construct/coverage.py
index 330ca09..25a7cf5 100644
--- a/src/construct/coverage.py
+++ b/src/construct/coverage.py
@@ -5,6 +5,26 @@
 from common import *
 
 
+def get_all_fourccs_inside(root):
+    fourccs = set([root["@Type"]])
+
+    def crawl(hierarchy):
+        for key, value in hierarchy.items():
+            if isinstance(value, dict):
+                if "@Type" not in value:
+                    continue
+                fourcc = value["@Type"]
+
+                fourccs.add(fourcc)
+                crawl(value)
+            elif isinstance(value, list):
+                for item in value:
+                    crawl({key: item})
+
+    crawl(root)
+    return fourccs
+
+
 def main():
     with open("output/dictionary.json", "r", encoding="utf-8") as f:
         dictionary = json.load(f)
@@ -80,11 +100,13 @@ def main():
     for extension in extensions:
         with open(extension, "r", encoding="utf-8") as f:
             ext_data = json.load(f)
-        fourcc_in_extensions.update([e["box"]["@Type"] for e in ext_data["extensions"]])
         missing_extensions.update(
             [f"{e['location']}.{e['box']['@Type']}" for e in ext_data["extensions"]]
         )
 
+        for e in ext_data["extensions"]:
+            fourcc_in_extensions.update(get_all_fourccs_inside(e["box"]))
+
     # Remove all known and unknown paths from missing extensions (to reduce redundancy)
     missing_extensions.difference_update(set(files["not_found"]))
     missing_extensions.difference_update(set(files["path_file_map"].keys()))
diff --git a/src/construct/files.py b/src/construct/files.py
index 0ce1bdb..983097d 100644
--- a/src/construct/files.py
+++ b/src/construct/files.py
@@ -98,15 +98,30 @@ def add(path, variant):
         paths[path].add(variant)
 
     def crawl(root, path):
-        if mp4ra_check and path[-1] not in get_mp4ra_boxes():
-            return
-        add_variant(add, root, path)
-        if "children" in root:
-            for child in root["children"]:
-                crawl(child, path + [root["@Type"]])
+        for key, value in root.items():
+            if isinstance(value, dict):
+                if "@Type" not in value:
+                    continue
+                fourcc = value["@Type"]
+
+                if mp4ra_check and fourcc not in get_mp4ra_boxes():
+                    continue
+
+                add_variant(add, value, path)
+                crawl(value, path + [fourcc])
+            elif isinstance(value, list):
+                for item in value:
+                    crawl({key: item}, path)
 
     for ext in extension["extensions"]:
-        crawl(ext["box"], ext["location"].split("."))
+        root = ext["box"]
+        path = ext["location"].split(".")
+
+        if mp4ra_check and path[-1] not in get_mp4ra_boxes():
+            continue
+
+        add_variant(add, root, path)
+        crawl(root, path + [root["@Type"]])
 
     return paths
 
diff --git a/src/feature_extractor/file_features.py b/src/feature_extractor/file_features.py
index 4b06d86..3e520fe 100644
--- a/src/feature_extractor/file_features.py
+++ b/src/feature_extractor/file_features.py
@@ -722,7 +722,13 @@ def _extract_file_features(args, exit_on_error=True):
         with open(gpac_path, "r", encoding="utf-8") as f:
             gpac_dict_gt = json.load(f)
 
-        if gpac_dict_gt["IsoMediaFile"] != gpac_dict["IsoMediaFile"]:
+        # Check if this was manual dump
+        manual_dump = gpac_dict_gt.get("manualDump", False)
+
+        if (
+            gpac_dict_gt["IsoMediaFile"] != gpac_dict["IsoMediaFile"]
+            and not manual_dump
+        ):
             print(
                 f'WARNING: GPAC file for "{input_path}" already exists but the contents have been changed. Forcing overwrite!'
             )
@@ -733,10 +739,19 @@ def _extract_file_features(args, exit_on_error=True):
     else:
         dump_to_json(gpac_path, gpac_dict)
 
+    # If it was a manual dump, use that as the GPAC file
+    if not args.overwrite and manual_dump:
+        print(f'WARNING: "{gpac_path}" was a manual dump. Using that as the GPAC file.')
+        gpac_dict = gpac_dict_gt
+
     # Create GPAC extension
     mp4ra_check = "under_consideration" not in os.path.dirname(metadata_path)
     unknown_boxes = traverse_gpac_dict(gpac_dict, mp4ra_check)
     if len(unknown_boxes) == 0:
+        # Remove GPAC extension if there are no unknown boxes anymore
+        if os.path.exists(gpac_extension_path):
+            os.remove(gpac_extension_path)
+
         if exit_on_error:
             sys.exit(0)
         else:
diff --git a/src/tests/test_files.py b/src/tests/test_files.py
index d7e265d..ec6e12b 100644
--- a/src/tests/test_files.py
+++ b/src/tests/test_files.py
@@ -314,10 +314,8 @@ def test_gpac_ext_consistency(check):
             gpac_ext_dict = json.load(f)
 
         # Test if boxes are the same
-        gt_locations = [(ub["location"], ub["box"]["@data"]) for ub in unknown_boxes]
-        ref_locations = [
-            (ub["location"], ub["box"]["@data"]) for ub in gpac_ext_dict["extensions"]
-        ]
+        gt_locations = [ub["location"] for ub in unknown_boxes]
+        ref_locations = [ub["location"] for ub in gpac_ext_dict["extensions"]]
 
         # Reference must match exactly
         with check: