diff --git a/src/KavitaPersonUpdater.py b/src/KavitaPersonUpdater.py
index 1a65692..1eb7b79 100644
--- a/src/KavitaPersonUpdater.py
+++ b/src/KavitaPersonUpdater.py
@@ -30,7 +30,7 @@ import requests
 from KavitaClient import KavitaClient
 from MALResolver import MALResolver
 from AniListResolver import AniListResolver
-from TextUtils import best_similarity, paragraphs_to_html
+from TextUtils import best_similarity, paragraphs_to_html, person_name_with_id
 
 
 class KavitaPersonUpdater:
@@ -152,11 +152,28 @@ class KavitaPersonUpdater:
             if not name and not raw_name:
                 continue
 
-            # Search by the cleaned (XML-safe) name first; if Kavita stores
-            # the legacy comma form, retry with the raw MAL name.
-            matches = self._find_kavita_person(name) if name else []
-            if not matches and raw_name and raw_name != name:
-                matches = self._find_kavita_person(raw_name)
+            if kind == "character":
+                # Characters are stored under their disambiguated name
+                # ("Rem (MAL 118737)") — see person_name_with_id.  The
+                # series metadata write creates the person under exactly
+                # this name, so only that form is searched.
+                search_names = [person_name_with_id(
+                    name, mal_id=entry.get("mal_id"),
+                    al_id=entry.get("al_id"))]
+            else:
+                # Staff: cleaned (XML-safe) name first; if Kavita stores
+                # the legacy comma form, retry with the raw MAL name.
+                search_names = [name]
+                if raw_name and raw_name != name:
+                    search_names.append(raw_name)
+
+            matches: list[dict] = []
+            for search_name in search_names:
+                if not search_name:
+                    continue
+                matches = self._find_kavita_person(search_name)
+                if matches:
+                    break
 
             if not matches:
                 result["not_found"] += 1
@@ -234,6 +251,20 @@ class KavitaPersonUpdater:
 
         current_mal_id: int = person.get("malId") or 0
         current_al_id:  int = person.get("aniListId") or 0
+
+        # Collision guard: the Kavita person is already linked to a
+        # *different* tracker entity — same display name, different
+        # character/person.  Never overwrite; first writer wins.
+        if ((mal_id and current_mal_id and current_mal_id != mal_id)
+                or (al_id and current_al_id and current_al_id != al_id)):
+            if errors is not None:
+                errors.append(
+                    f"conflict: '{person_name}' (#{person_id}) is linked to "
+                    f"malId={current_mal_id or '-'}/aniListId={current_al_id or '-'} "
+                    f"but this entry has malId={mal_id or '-'}/aniListId={al_id or '-'} "
+                    f"— skipped")
+            return False
+
         needs_mal_id = bool(mal_id and current_mal_id != mal_id)
         needs_al_id  = bool(al_id  and current_al_id  != al_id)
 
diff --git a/src/LightNovelMetadataBuilder.py b/src/LightNovelMetadataBuilder.py
index 3a89e02..871c30a 100644
--- a/src/LightNovelMetadataBuilder.py
+++ b/src/LightNovelMetadataBuilder.py
@@ -26,7 +26,7 @@ from MangaBakaRateLimit import apply_to_session as _apply_mangabaka_rate_limit
 from MALResolver import MALResolver
 from AniListResolver import AniListResolver
 from MatchesCache import MatchesCache
-from TextUtils import paragraphs_to_html
+from TextUtils import paragraphs_to_html, person_name_with_id
 
 
 # MangaBaka series type for the search endpoint.
@@ -311,9 +311,15 @@ class LightNovelMetadataBuilder:
         if not staff_detailed and al_id:
             staff_detailed = self._al.get_staff_detailed(al_id)
 
-        # Character / writer name lists for SeriesMetadata
-        character_names = [c["name"] for c in characters_detailed
-                           if c.get("name")]
+        # Character names for SeriesMetadata, disambiguated with the
+        # tracker character id ("Rem (MAL 118737)") because Kavita person
+        # records are global and keyed by name only.
+        character_names = [
+            person_name_with_id(c["name"],
+                                mal_id=c.get("mal_id"),
+                                al_id=c.get("al_id"))
+            for c in characters_detailed if c.get("name")
+        ]
         # Writers come from MangaBaka first (authoritative for novels)
         writers = list(md.get("authors") or [])
         # Illustrators / artists -> CoverArtists (Kavita has no dedicated
diff --git a/src/TextUtils.py b/src/TextUtils.py
index 1924b71..5b343cf 100644
--- a/src/TextUtils.py
+++ b/src/TextUtils.py
@@ -43,3 +43,30 @@ def best_similarity(query: str, candidates: Iterable[str]) -> float:
                 None, q, str(candidate).lower()).ratio()
             best = max(best, ratio)
     return best
+
+
+def person_name_with_id(name: str, *,
+                        mal_id: "int | None" = None,
+                        al_id:  "int | None" = None) -> str:
+    """
+    Disambiguates a character name with its tracker id: "Rem (MAL 118737)".
+
+    Kavita Person records are global and keyed by name only, so two
+    different characters who share a name would collapse into one record.
+    Suffixing the tracker *character* id keeps them apart while still
+    sharing the record across the manga and light-novel version of the
+    same series (MAL/AniList character ids are per character, not per
+    medium).  MAL is preferred; AniList ids get an "AL" marker so the two
+    id spaces cannot collide.  Without any id the name is returned as-is.
+
+    The format must stay in sync with the manga project so both tools
+    address the same Kavita person records.
+    """
+    name = (name or "").strip()
+    if not name:
+        return name
+    if mal_id:
+        return f"{name} (MAL {mal_id})"
+    if al_id:
+        return f"{name} (AL {al_id})"
+    return name