diff --git a/examples/contracts/log_indexer.py b/examples/contracts/log_indexer.py
index 15fdbd17d..0ad420694 100644
--- a/examples/contracts/log_indexer.py
+++ b/examples/contracts/log_indexer.py
@@ -66,27 +66,29 @@ def get_closest_vector(self, text: str) -> dict | None:
     def add_log(self, log: str, log_id: int) -> None:
         key = log_id
         if key in self.log_vector_ids:
-            self.vector_store.get_by_id(self.log_vector_ids[key]).value = StoreValue(
-                text=log, log_id=key
-            )
-            return
+            # VecDBElement.key (the stored vector used by knn()) is
+            # read-only - only .value can be updated in place. Re-embedding
+            # requires removing the stale entry and inserting a fresh one,
+            # otherwise a changed log stays ranked by its old embedding.
+            self.vector_store.get_by_id(self.log_vector_ids[key]).remove()
 
         emb = self.get_embedding(log)
         vector_id = self.vector_store.insert(emb, StoreValue(text=log, log_id=key))
         self.log_vector_ids[key] = vector_id
+        self.removed_log_ids[key] = False
 
     @gl.public.write
     def update_log(self, log_id: int, log: str) -> None:
         key = log_id
         if key in self.log_vector_ids:
-            self.vector_store.get_by_id(self.log_vector_ids[key]).value = StoreValue(
-                text=log, log_id=key
-            )
-            return
+            # See add_log: .key is read-only, must remove+reinsert to
+            # actually re-embed rather than leaving the stale vector.
+            self.vector_store.get_by_id(self.log_vector_ids[key]).remove()
 
         emb = self.get_embedding(log)
         vector_id = self.vector_store.insert(emb, StoreValue(text=log, log_id=key))
         self.log_vector_ids[key] = vector_id
+        self.removed_log_ids[key] = False
 
     @gl.public.write
     def remove_log(self, id: int) -> None:
diff --git a/tests/integration/icontracts/tests/test_log_indexer.py b/tests/integration/icontracts/tests/test_log_indexer.py
index 8db4802c1..3bf38dc67 100644
--- a/tests/integration/icontracts/tests/test_log_indexer.py
+++ b/tests/integration/icontracts/tests/test_log_indexer.py
@@ -108,3 +108,65 @@ def test_log_indexer(setup_validators):
     assert float(closest_vector_log_2["similarity"]) > 0.99
     assert closest_vector_log_2["id"] == 3
     assert closest_vector_log_2["text"] == "This is the third log"
+
+    # ########################################
+    # ### Removed log_id becomes visible again
+    # ###         after being re-added
+    # ########################################
+    transaction_response_remove_log_2 = contract.remove_log(args=[3]).transact()
+    assert tx_execution_succeeded(transaction_response_remove_log_2)
+
+    # tombstoned: must not be found anymore
+    closest_vector_after_remove = contract.get_closest_vector(
+        args=["This is the third log"]
+    ).call()
+    assert (
+        closest_vector_after_remove is None
+        or closest_vector_after_remove["id"] != 3
+    )
+
+    # re-add the same log_id via add_log (takes the "already indexed"
+    # in-place-update branch, since log_id 3 is still in log_vector_ids)
+    transaction_response_readd_log_2 = contract.add_log(
+        args=["This is the third log, again", 3]
+    ).transact()
+    assert tx_execution_succeeded(transaction_response_readd_log_2)
+
+    # must be visible again — the tombstone from remove_log must have
+    # been cleared, not left permanently set
+    closest_vector_after_readd = contract.get_closest_vector(
+        args=["This is the third log, again"]
+    ).call()
+    assert closest_vector_after_readd is not None
+    assert closest_vector_after_readd["id"] == 3
+    assert closest_vector_after_readd["text"] == "This is the third log, again"
+
+    # ########################################
+    # ### Updating an existing log_id must
+    # ###   re-embed, not just re-label
+    # ########################################
+    # log_id 3 currently holds "This is the third log, again" — an
+    # unrelated topic. If add_log's "already indexed" branch only updated
+    # .value (the old bug: VecDBElement.key, the stored vector, is
+    # read-only and can't be changed in place) instead of removing and
+    # re-inserting with a fresh embedding, this log would still be found
+    # by queries about its OLD topic and missed by queries about its new
+    # one, despite .text correctly reporting the new content.
+    transaction_response_retopic_log_2 = contract.add_log(
+        args=["Quantum computers use superconducting qubits", 3]
+    ).transact()
+    assert tx_execution_succeeded(transaction_response_retopic_log_2)
+
+    # found by a query about the NEW topic — this alone proves re-embedding
+    # happened: the OLD embedding ("This is the third log, again") has
+    # nothing in common with quantum computing, so this could only match
+    # if a fresh embedding was actually computed and stored.
+    closest_vector_new_topic = contract.get_closest_vector(
+        args=["How do quantum computers work"]
+    ).call()
+    assert closest_vector_new_topic is not None
+    assert closest_vector_new_topic["id"] == 3
+    assert (
+        closest_vector_new_topic["text"]
+        == "Quantum computers use superconducting qubits"
+    )