diff --git a/scripts/mneme_compile_memory.py b/scripts/mneme_compile_memory.py
index aed9e9c..121475c 100755
--- a/scripts/mneme_compile_memory.py
+++ b/scripts/mneme_compile_memory.py
@@ -58,7 +58,10 @@
             r"\bbounty\b", r"\bissue\b", r"\bprefers\b", r"\bavoid\b", r"\bno public\b",
             r"\bproof summaries\b", r"\bdecision support\b",
         ],
-        "headingHints": [r"systems", r"infrastructure", r"key infrastructure", r"access", r"memory stack"],
+        "headingHints": [
+            r"systems", r"infrastructure", r"key infrastructure", r"access", r"memory stack",
+            r"server access", r"services on the box",
+        ],
     },
     "decisions": {
         "include": [
@@ -120,7 +123,7 @@
 }
 HEADING_BUCKET_HINTS = {
     "projects": [r"active projects", r"durable project facts", r"bdeep", r"yibin", r"aqua", r"mneme"],
-    "systems": [r"key infrastructure", r"systems", r"access", r"memory stack"],
+    "systems": [r"key infrastructure", r"systems", r"access", r"memory stack", r"server access", r"services on the box"],
     "decisions": [r"durable decisions", r"preferences", r"methods", r"hard constraints"],
     "incidents": [r"incidents", r"warnings"],
     "people": [r"identity", r"user", r"people", r"profile"],
@@ -138,7 +141,7 @@
     "conversation summary", "source files", "sources", "tools", "methods", "pending",
     "what it does", "what it is", "what it is not", "goal", "status",
     "current status", "stable assumptions", "recommended next step", "practical interpretation",
-    "docs", "automation", "runtime orchestration", "continuation guide",
+    "docs", "automation", "runtime orchestration", "continuation guide", "build",
 }
 GENERIC_SECTION_PATTERNS = [
     re.compile(r"\bactive projects\b", re.I),
@@ -322,6 +325,8 @@ def is_generic_or_noise_title(text: str) -> bool:
         return True
     if norm in GENERIC_TITLES:
         return True
+    if any(p.search(norm) for p in GENERIC_SECTION_PATTERNS):
+        return True
     if re.fullmatch(r"20\d{2} \d{2} \d{2}", norm):
         return True
     return False
@@ -337,8 +342,12 @@ def is_low_value_item(item: SourceLine, category: str | None = None) -> bool:
         return True
     if is_heading_only_text(text):
         return True
-    if item.kind == "note_section" and not body_lines(text):
-        return True
+    if item.kind == "note_section":
+        section_title = extract_section_title(item)
+        if section_title and is_generic_or_noise_title(section_title):
+            return True
+        if not body_lines(text):
+            return True
     if is_bulky_section_dump(item):
         return True
     body = first_body_line(text)
@@ -355,7 +364,7 @@ def heading_bucket(item: SourceLine) -> tuple[str | None, int]:
     score = 0
     for category, pats in HEADING_BUCKET_HINTS.items():
         s = sum(1 for pat in pats if re.search(pat, heading_text))
-        if s > score:
+        if s > score or (s == score and s > 0 and best and CATEGORY_PRIORITY[category] > CATEGORY_PRIORITY[best]):
             best, score = category, s
     return best, score
 
diff --git a/tests/test_smoke.py b/tests/test_smoke.py
index 89c5f2b..cc48c8a 100644
--- a/tests/test_smoke.py
+++ b/tests/test_smoke.py
@@ -28,22 +28,7 @@ def make_workspace(self) -> Path:
         self.addCleanup(lambda: shutil.rmtree(tmp, ignore_errors=True))
         return ws
 
-    def test_ingest_fixture_workspace(self) -> None:
-        ws = self.make_workspace()
-        raw_out = ws / 'raw-out'
-        data = run_json([
-            sys.executable,
-            str(SCRIPTS / 'mneme_ingest_memory.py'),
-            '--root', str(ws),
-            '--out', str(raw_out),
-        ])
-        self.assertGreaterEqual(data['sourceCount'], 2)
-        self.assertGreater(data['itemCount'], 0)
-        self.assertTrue((raw_out / 'sources.jsonl').exists())
-        self.assertTrue((raw_out / 'items.jsonl').exists())
-
-    def test_compile_fixture_raw_into_outputs(self) -> None:
-        ws = self.make_workspace()
+    def compile_workspace(self, ws: Path) -> Path:
         raw_out = ws / 'raw-out'
         compiled_out = ws / 'compiled-out'
         run_json([
@@ -61,14 +46,70 @@ def test_compile_fixture_raw_into_outputs(self) -> None:
         ], capture_output=True, text=True)
         if cp.returncode != 0:
             raise AssertionError(f"Compile failed ({cp.returncode})\nSTDOUT:\n{cp.stdout}\nSTDERR:\n{cp.stderr}")
+        return compiled_out
+
+    def test_ingest_fixture_workspace(self) -> None:
+        ws = self.make_workspace()
+        raw_out = ws / 'raw-out'
+        data = run_json([
+            sys.executable,
+            str(SCRIPTS / 'mneme_ingest_memory.py'),
+            '--root', str(ws),
+            '--out', str(raw_out),
+        ])
+        self.assertGreaterEqual(data['sourceCount'], 2)
+        self.assertGreater(data['itemCount'], 0)
+        self.assertTrue((raw_out / 'sources.jsonl').exists())
+        self.assertTrue((raw_out / 'items.jsonl').exists())
+
+    def test_compile_fixture_raw_into_outputs(self) -> None:
+        ws = self.make_workspace()
+        compiled_out = self.compile_workspace(ws)
+        projects_text = (compiled_out / 'projects.md').read_text()
+        people_text = (compiled_out / 'people.md').read_text()
         self.assertTrue((compiled_out / 'projects.md').exists())
         self.assertTrue((compiled_out / 'people.md').exists())
         self.assertTrue((compiled_out / 'timeline.md').exists())
-        self.assertIn('Project Alpha', (compiled_out / 'projects.md').read_text())
-        self.assertIn('Compiled Memory — People', (compiled_out / 'people.md').read_text())
+        self.assertIn('Project Alpha', projects_text)
+        self.assertNotIn('## Active Projects', projects_text)
+        self.assertIn('Compiled Memory — People', people_text)
+        self.assertIn('What to call them', people_text)
         self.assertTrue((compiled_out / 'documents.jsonl').exists())
         self.assertTrue((compiled_out / 'entries.jsonl').exists())
 
+    def test_compile_demotes_generic_sections_and_prefers_system_subheadings(self) -> None:
+        ws = self.make_workspace()
+        extra_note = ws / 'memory' / '2026-04-01-routing.md'
+        extra_note.write_text(
+            '# 2026-04-01\n\n'
+            '## Aqua-CQ Project\n\n'
+            '### Server Access\n'
+            '- Hostname: iv-ydyut13e9ss6ipm2he1t\n'
+            '- Ubuntu 24.04.4, 8GB RAM, 40GB disk\n'
+            '- Project at /opt/aqua-cq (backend only), frontend deployed to /var/www/html/cq/\n\n'
+            '### Three services on the box\n'
+            '- :8003 — aqua-qdh (backend-dev, /opt/aqua-qdh)\n\n'
+            '## Deploy process (correct)\n\n'
+            '```bash\n'
+            '# Build\n'
+            'export PATH="$HOME/.openclaw/tools/node-v22.22.0/bin:$PATH"\n'
+            'cd /opt/aqua-cq && npm run build\n'
+            '```\n',
+            encoding='utf-8',
+        )
+
+        compiled_out = self.compile_workspace(ws)
+        projects_text = (compiled_out / 'projects.md').read_text()
+        systems_text = (compiled_out / 'systems.md').read_text()
+
+        self.assertNotIn('## Build', projects_text)
+        self.assertNotIn('## Hostname', projects_text)
+        self.assertNotIn('## Ubuntu 24.04.4, 8GB RAM, 40GB disk', projects_text)
+        self.assertNotIn('## :8003 — aqua-qdh (backend-dev, /opt/aqua-qdh)', projects_text)
+        self.assertIn('## Server Access — Hostname', systems_text)
+        self.assertIn('## Ubuntu 24.04.4, 8GB RAM, 40GB disk', systems_text)
+        self.assertIn('## :8003 — aqua-qdh (backend-dev, /opt/aqua-qdh)', systems_text)
+
     def test_runtime_prepare_people_category(self) -> None:
         ws = self.make_workspace()
         raw_out = ws / 'runtime-raw'