diff --git a/lovstudio-any2pdf/scripts/md2pdf.py b/lovstudio-any2pdf/scripts/md2pdf.py
index a593154..a62eadf 100644
--- a/lovstudio-any2pdf/scripts/md2pdf.py
+++ b/lovstudio-any2pdf/scripts/md2pdf.py
@@ -1085,10 +1085,12 @@ def parse_md(self, md):
                 toc.append(('chapter', title, cm.key))
                 i += 1; continue
 
-            # H3 = Section
-            if stripped.startswith('### '):
+            # H3–H6 = Section. Deeper levels fall back to the H3 style so their
+            # content is still rendered (ReportLab has no built-in H4–H6 styles here).
+            m_h = re.match(r'^(#{3,6})\s+(.+)$', stripped)
+            if m_h:
                 story.append(Spacer(1, 3*mm))
-                story.append(Paragraph(md_inline(stripped[4:].strip(), ah), ST['h3']))
+                story.append(Paragraph(md_inline(m_h.group(2).strip(), ah), ST['h3']))
                 story.append(Spacer(1, 1*mm))
                 i += 1; continue
 
@@ -1134,6 +1136,10 @@ def parse_md(self, md):
                     else:
                         merged += ' ' + pl
                 story.append(Paragraph(md_inline(merged, ah), ST['body']))
+            else:
+                # Defensive: current line matched no handler above (e.g. a stray
+                # marker we don't parse). Advance to prevent an infinite loop.
+                i += 1
             continue
 
         return story, toc
diff --git a/tests/07-deep-headings.md b/tests/07-deep-headings.md
new file mode 100644
index 0000000..c3c02f2
--- /dev/null
+++ b/tests/07-deep-headings.md
@@ -0,0 +1,44 @@
+---
+title: "Deep Headings Regression Test"
+subtitle: "H4/H5/H6 must not hang the parser"
+author: "any2pdf"
+version: "test"
+---
+
+# Part
+
+Body paragraph before the nested headings.
+
+## Chapter
+
+### Section (H3 — already supported)
+
+Body text under H3.
+
+#### Subsection (H4 — previously hung the parser)
+
+Body text under H4. The parse_md loop used to never advance `i` when it saw a
+deeper heading like this, because H4–H6 had no handler and the paragraph
+collector broke on lines starting with `#`. The result was 100% CPU forever.
+
+##### Deeper subsection (H5)
+
+Body text under H5 with some `inline code` and **bold**.
+
+###### Deepest (H6)
+
+Body text under H6.
+
+#### Another H4 right after a table
+
+| Col A | Col B | Col C |
+|-------|-------|-------|
+| a1    | b1    | c1    |
+| a2    | b2    | c2    |
+
+Body paragraph after the table. This used to never render because the H4 above
+would spin forever.
+
+#### H4 with & ampersand and `${VAR}/path.md`
+
+Inline code inside deep heading context should not break anything.