diff --git a/CHANGELOG.md b/CHANGELOG.md
index 49662fe..a5d047b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -14,6 +14,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **Parameterized LIMIT clauses**: `get_reads`, `get_actions`, and `get_activity` now use bind parameters (`%s`) for LIMIT values instead of f-string interpolation, eliminating a fragile SQL construction pattern (MEDIUM #3)
 
 ### Fixed
+- **Ollama response validation**: `OllamaEmbedding.embed()` now validates that the number of returned embeddings matches the number of input texts, raising `ValueError` on partial responses instead of silently dropping entries via `zip(strict=False)` (MEDIUM #21)
 - **Alert expiry filter**: `get_active_alerts` and `get_all_active_alerts` now filter out expired alerts (`expires > NOW()`), matching the behavior of `get_active_suppressions` (MEDIUM #18)
 - **Intention lifecycle**: `generate_briefing` now transitions fired intentions from "pending" to "fired" state, preventing them from firing on every subsequent briefing read
 - **Custom prompt sync uses DEFAULT_OWNER**: `_sync_custom_prompts` now queries `DEFAULT_OWNER` instead of the request-scoped `_owner_id()`, preventing User A's prompt sync from leaking into User B's prompt registry in multi-tenant deployments (MEDIUM #14)
diff --git a/src/mcp_awareness/embeddings.py b/src/mcp_awareness/embeddings.py
index a14ee85..8032bee 100644
--- a/src/mcp_awareness/embeddings.py
+++ b/src/mcp_awareness/embeddings.py
@@ -118,6 +118,15 @@ def embed(self, texts: list[str]) -> list[list[float]]:
         with urllib.request.urlopen(req, timeout=self._timeout) as resp:
             data = json.loads(resp.read())
         result: list[list[float]] = data["embeddings"]
+        if len(result) != len(texts):
+            logger.warning(
+                "Ollama returned %d embeddings for %d texts — partial response",
+                len(result),
+                len(texts),
+            )
+            raise ValueError(
+                f"Ollama returned {len(result)} embeddings for {len(texts)} input texts"
+            )
         return result
 
     def is_available(self) -> bool:
diff --git a/tests/test_embeddings.py b/tests/test_embeddings.py
index 090e626..72fed23 100644
--- a/tests/test_embeddings.py
+++ b/tests/test_embeddings.py
@@ -18,7 +18,9 @@
 
 from __future__ import annotations
 
+import json
 import os
+from unittest.mock import MagicMock, patch
 
 import pytest
 
@@ -203,6 +205,35 @@ def test_unreachable_is_not_available(self):
         p = OllamaEmbedding(base_url="http://localhost:19999")
         assert p.is_available() is False
 
+    def test_partial_response_raises(self):
+        """embed() raises ValueError when Ollama returns fewer embeddings than inputs."""
+        p = OllamaEmbedding(base_url="http://localhost:19999")
+        # Simulate Ollama returning only 1 embedding for 3 input texts
+        partial_body = json.dumps({"embeddings": [[0.1] * 768]}).encode()
+        mock_resp = MagicMock()
+        mock_resp.read.return_value = partial_body
+        mock_resp.__enter__ = lambda s: s
+        mock_resp.__exit__ = MagicMock(return_value=False)
+
+        with (
+            patch("urllib.request.urlopen", return_value=mock_resp),
+            pytest.raises(ValueError, match=r"1 embeddings for 3 input texts"),
+        ):
+            p.embed(["text one", "text two", "text three"])
+
+    def test_correct_response_count_succeeds(self):
+        """embed() returns normally when embedding count matches input count."""
+        p = OllamaEmbedding(base_url="http://localhost:19999")
+        good_body = json.dumps({"embeddings": [[0.1] * 768, [0.2] * 768]}).encode()
+        mock_resp = MagicMock()
+        mock_resp.read.return_value = good_body
+        mock_resp.__enter__ = lambda s: s
+        mock_resp.__exit__ = MagicMock(return_value=False)
+
+        with patch("urllib.request.urlopen", return_value=mock_resp):
+            result = p.embed(["text one", "text two"])
+        assert len(result) == 2
+
 
 # ---------------------------------------------------------------------------
 # create_provider