diff --git a/.github/workflows/prerelease.yml b/.github/workflows/prerelease.yml
index afcdf874..e486db0e 100644
--- a/.github/workflows/prerelease.yml
+++ b/.github/workflows/prerelease.yml
@@ -261,8 +261,7 @@ jobs:
       - name: Build Python
         run: |
           python -m pip install --upgrade pip
-          pip install pytest pytest-repeat numpy
-          python -m pip install .
+          python -m pip install . --group tests
         env:
           CXX: clang++
           CC: clang++ # Override the default compiler
diff --git a/pyproject.toml b/pyproject.toml
index 7ef6a497..0b0205ef 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -20,8 +20,10 @@ requires = [
 addopts = "-ra --showlocals --strict-markers --strict-config -s -x -p no:warnings"
 filterwarnings = ["error"]
 minversion = "6.0"
-testpaths = ["python/scripts"]
 xfail_strict = true
+testpaths = [
+    "python/tests",
+]
 
 # Avoid running tests, as everything is happening in a super slow container
 # We have already run all the relavent Python tests in `prerelease.yml`
@@ -94,6 +96,7 @@ lint = [
 tests = [
   "numpy>=1.21",
   "pytest>=9.0.2",
+  "pytest-cov>=7.1.0",
   "pytest-repeat>=0.9.4",
 ]
 
diff --git a/python/scripts/test_distances.py b/python/tests/test_distances.py
similarity index 100%
rename from python/scripts/test_distances.py
rename to python/tests/test_distances.py
diff --git a/python/scripts/test_index.py b/python/tests/test_index.py
similarity index 87%
rename from python/scripts/test_index.py
rename to python/tests/test_index.py
index 4a339914..c8654a53 100644
--- a/python/scripts/test_index.py
+++ b/python/tests/test_index.py
@@ -510,6 +510,94 @@ def test_index_quantized_add_search(quantization, ndim):
         assert matches[i].keys[0] == i, f"Expected self-match for vector {i} with {quantization}"
 
 
+def test_index_ip_metric():
+    """Inner product metric should be usable and produce valid searches."""
+    ndim = 32
+    count = 20
+    index = Index(ndim=ndim, metric=MetricKind.IP)
+    keys = np.arange(count)
+    vectors = random_vectors(count=count, ndim=ndim, metric=MetricKind.IP)
+    index.add(keys, vectors)
+
+    matches = index.search(vectors[0], 5)
+    assert isinstance(matches, Matches)
+    assert len(matches) == 5
+
+
+def test_index_specs():
+    """specs property should return a dict with expected keys."""
+    ndim = 16
+    index = Index(ndim=ndim, metric=MetricKind.L2sq, dtype=ScalarKind.F32)
+    s = index.specs
+    assert isinstance(s, dict)
+    for key in ("ndim", "multi", "connectivity", "expansion_add", "expansion_search", "dtype"):
+        assert key in s
+    assert s["ndim"] == ndim
+
+
+@pytest.mark.parametrize("ndim", [16, 64])
+@pytest.mark.parametrize("batch_size", [10, 50])
+def test_index_exact_search(ndim, batch_size):
+    """Exact search must return the query vector itself as the top match."""
+    index = Index(ndim=ndim, metric=MetricKind.L2sq)
+    keys = np.arange(batch_size)
+    vectors = random_vectors(count=batch_size, ndim=ndim)
+    index.add(keys, vectors)
+
+    if batch_size == 1:
+        matches = index.search(vectors, 1, exact=True)
+        assert int(matches.keys[0]) == keys[0]
+    else:
+        matches: BatchMatches = index.search(vectors, 1, exact=True)
+        top_keys = [int(m.keys[0]) for m in matches]
+        assert top_keys == list(keys)
+
+
+@pytest.mark.parametrize("batch_size", [6, 20])
+def test_index_pairwise_distance_array(batch_size):
+    """pairwise_distance with equal-length key arrays returns element-wise distances."""
+    ndim = 16
+    index = Index(ndim=ndim, metric=MetricKind.L2sq)
+    keys = np.arange(batch_size)
+    vectors = random_vectors(count=batch_size, ndim=ndim)
+    index.add(keys, vectors)
+
+    half = batch_size // 2
+    left_keys = keys[:half]
+    right_keys = keys[half : 2 * half]
+    distances = index.pairwise_distance(left_keys, right_keys)
+    assert distances.shape == (half,)
+    assert np.all(distances >= 0)
+
+
+def test_index_pairwise_distance_scalar():
+    """pairwise_distance with scalar keys returns a scalar distance."""
+    ndim = 16
+    index = Index(ndim=ndim, metric=MetricKind.L2sq)
+    keys = np.arange(4)
+    vectors = random_vectors(count=4, ndim=ndim)
+    index.add(keys, vectors)
+
+    dist = index.pairwise_distance(0, 1)
+    assert isinstance(dist, float)
+    assert dist >= 0
+    # Distance from a vector to itself should be ~0
+    assert index.pairwise_distance(0, 0) == pytest.approx(0.0, abs=1e-3)
+
+
+def test_index_memory_usage_grows():
+    """Memory usage should increase as vectors are added."""
+    ndim = 32
+    index = Index(ndim=ndim)
+    mem_empty = index.memory_usage
+
+    keys = np.arange(100)
+    vectors = random_vectors(count=100, ndim=ndim)
+    index.add(keys, vectors)
+
+    assert index.memory_usage > mem_empty
+
+
 def test_index_copied_memory_usage():
     """Test that copy=False results in lower memory usage than copy=True."""
     reset_randomness()
diff --git a/python/scripts/test_jit.py b/python/tests/test_jit.py
similarity index 100%
rename from python/scripts/test_jit.py
rename to python/tests/test_jit.py
diff --git a/python/scripts/test_sparse.py b/python/tests/test_sparse.py
similarity index 100%
rename from python/scripts/test_sparse.py
rename to python/tests/test_sparse.py
diff --git a/python/scripts/test_sqlite.py b/python/tests/test_sqlite.py
similarity index 100%
rename from python/scripts/test_sqlite.py
rename to python/tests/test_sqlite.py
diff --git a/python/scripts/test_tooling.py b/python/tests/test_tooling.py
similarity index 100%
rename from python/scripts/test_tooling.py
rename to python/tests/test_tooling.py