diff --git a/klausur-service/backend/grid_editor_api.py b/klausur-service/backend/grid_editor_api.py
index e510aaf..4aaed6f 100644
--- a/klausur-service/backend/grid_editor_api.py
+++ b/klausur-service/backend/grid_editor_api.py
@@ -1324,64 +1324,9 @@ async def _build_grid_core(session_id: str, session: dict) -> dict:
             if orig:
                 cell["col_type"] = orig
 
-        # 5d. Remove IPA continuation rows — rows where the printed
-        # phonetic transcription wraps to a line below the headword.
-        # These rows have text only in the English column (+ margin
-        # noise) and fix_cell_phonetics did NOT insert IPA brackets
-        # (because there's no real English word to look up).
-        ipa_cont_rows: set = set()
-        for z in zones_data:
-            for row in z.get("rows", []):
-                ri = row["index"]
-                row_cells = [
-                    c for c in z.get("cells", [])
-                    if c.get("row_index") == ri
-                ]
-                en_cells = [
-                    c for c in row_cells
-                    if c.get("col_type") == en_col_type
-                ]
-                # Other cells with ≥3 chars (ignore margin noise)
-                other_cells = [
-                    c for c in row_cells
-                    if c.get("col_type") != en_col_type
-                    and len((c.get("text") or "").strip()) >= 3
-                ]
-                if en_cells and not other_cells:
-                    en_text = en_cells[0].get("text", "")
-                    # Strip any IPA brackets that fix_cell_phonetics
-                    # may have added for short dictionary matches
-                    # (e.g. "si" → "[si]") to check underlying text.
-                    text_bare = re.sub(r'\[[^\]]*\]', '', en_text).strip()
-                    # Garbled IPA typically contains ':' (length mark)
-                    # or starts with ' (stress mark), and has no word
-                    # with ≥3 letters that could be a real headword.
-                    has_headword = any(
-                        len(re.sub(r'[^a-zA-Z]', '', w)) >= 3
-                        for w in text_bare.split()
-                    ) if text_bare else False
-                    looks_phonetic = (
-                        ':' in text_bare
-                        or text_bare.startswith("'")
-                        or text_bare.startswith("\u2019")
-                        or not has_headword
-                    )
-                    if looks_phonetic:
-                        ipa_cont_rows.add(ri)
-        if ipa_cont_rows:
-            for z in zones_data:
-                z["rows"] = [
-                    r for r in z.get("rows", [])
-                    if r["index"] not in ipa_cont_rows
-                ]
-                z["cells"] = [
-                    c for c in z.get("cells", [])
-                    if c.get("row_index") not in ipa_cont_rows
-                ]
-            logger.info(
-                "removed %d IPA continuation rows: %s",
-                len(ipa_cont_rows), sorted(ipa_cont_rows),
-            )
+        # 5d. IPA continuation rows are preserved — they contain the
+        # printed phonetic transcription that wraps to a line below the
+        # headword.  The user can manually delete them if not needed.
 
     duration = time.time() - t0