diff --git a/.github/workflows/slow-tests.yml b/.github/workflows/slow-tests.yml
new file mode 100644
index 00000000..77fd3b3c
--- /dev/null
+++ b/.github/workflows/slow-tests.yml
@@ -0,0 +1,26 @@
+name: Run slow tests after PR merge
+
+on:
+  pull_request:
+    types: [closed]
+    branches: [main]
+
+jobs:
+  slow-tests:
+    if: github.event.pull_request.merged == true
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v3
+
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+
+      - name: Install dependencies
+        run: pip install -e .[test]
+
+      - name: Run slow tests
+        run: pytest -m slow
diff --git a/.github/workflows/run-scripts.yml b/.github/workflows/test-scripts.yml
similarity index 89%
rename from .github/workflows/run-scripts.yml
rename to .github/workflows/test-scripts.yml
index f51e2c8d..64b713b5 100644
--- a/.github/workflows/run-scripts.yml
+++ b/.github/workflows/test-scripts.yml
@@ -18,10 +18,10 @@ jobs:
           - scripts/analyze_element_errors.py
     steps:
       - name: Check out repository
-        uses: actions/checkout@v2
+        uses: actions/checkout@v3
 
       - name: Set up Python
-        uses: actions/setup-python@v2
+        uses: actions/setup-python@v4
         with:
           python-version: 3.9
 
diff --git a/matbench_discovery/data.py b/matbench_discovery/data.py
index 4d1afc7a..afabe7c4 100644
--- a/matbench_discovery/data.py
+++ b/matbench_discovery/data.py
@@ -28,6 +28,18 @@
 )
 
 
+def as_dict_handler(obj: Any) -> dict[str, Any] | None:
+    """Pass this to json.dump(default=) or as pandas.to_json(default_handler=) to
+    serialize Python classes with as_dict(). Warning: Objects without a as_dict() method
+    are replaced with None in the serialized data.
+    """
+    try:
+        return obj.as_dict()  # all MSONable objects implement as_dict()
+    except AttributeError:
+        return None  # replace unhandled objects with None in serialized data
+        # removes e.g. non-serializable AseAtoms from M3GNet relaxation trajectories
+
+
 def load_train_test(
     data_names: str | Sequence[str],
     version: str = figshare_versions[-1],
@@ -62,8 +74,7 @@ def load_train_test(
         ValueError: On bad version number or bad data names.
 
     Returns:
-        pd.DataFrame: Single dataframe or dictionary of dfs if
-        multiple data were requested.
+        pd.DataFrame: Single dataframe or dictionary of dfs if multiple data requested.
     """
     if version not in figshare_versions:
         raise ValueError(f"Unexpected {version=}. Must be one of {figshare_versions}.")
@@ -85,16 +96,24 @@ def load_train_test(
         reader = pd.read_csv if file.endswith(csv_ext) else pd.read_json
 
         cache_path = f"{cache_dir}/{file}"
-        if os.path.isfile(cache_path):
+        if os.path.isfile(cache_path):  # load from disk cache
             print(f"Loading {key!r} from cached file at {cache_path!r}")
             df = reader(cache_path, **kwargs)
-        else:
+        else:  # download from Figshare URL
+            # manually set compression since pandas can't infer from URL
+            if file.endswith(".gz"):
+                kwargs.setdefault("compression", "gzip")
+            elif file.endswith(".bz2"):
+                kwargs.setdefault("compression", "bz2")
             url = file_urls[key]
             print(f"Downloading {key!r} from {url}")
             try:
-                df = reader(url)
+                df = reader(url, **kwargs)
             except urllib.error.HTTPError as exc:
                 raise ValueError(f"Bad {url=}") from exc
+            except Exception:
+                print(f"\n\nvariable dump:\n{file=},\n{url=},\n{reader=},\n{kwargs=}")
+                raise
             if cache_dir and not os.path.isfile(cache_path):
                 os.makedirs(os.path.dirname(cache_path), exist_ok=True)
                 if ".csv" in file:
@@ -143,7 +162,7 @@ def glob_to_df(
     Returns:
         pd.DataFrame: Combined dataframe.
     """
-    reader = reader or pd.read_csv if ".csv" in pattern else pd.read_json
+    reader = reader or pd.read_csv if ".csv" in pattern.lower() else pd.read_json
 
     # prefix pattern with ROOT if not absolute path
     files = glob(pattern)
@@ -194,9 +213,9 @@ def __getattribute__(self, key: str) -> str:
         val = super().__getattribute__(key)
         if key in self and not os.path.isfile(val):
             msg = f"Warning: {val!r} associated with {key=} does not exist."
-            try:
-                self._on_not_found(key, msg)  # type: ignore[misc]
-            except TypeError:
+            if self._on_not_found:
+                self._on_not_found(key, msg)
+            else:
                 print(msg, file=sys.stderr)
         return val
 
@@ -214,7 +233,7 @@ def _on_not_found(self, key: str, msg: str) -> None:  # type: ignore[override]
 
         # default to 'y' if not in interactive session, and user can't answer
         answer = "" if sys.stdin.isatty() else "y"
-        while answer not in ("y", "n", "\x1b", ""):
+        while answer not in ("y", "n"):
             answer = input(f"{msg} [y/n] ").lower().strip()
         if answer == "y":
             load_train_test(key)  # download and cache data file
@@ -239,17 +258,5 @@ def _on_not_found(self, key: str, msg: str) -> None:  # type: ignore[override]
 DATA_FILES = DataFiles()
 
 
-def as_dict_handler(obj: Any) -> dict[str, Any] | None:
-    """Pass this to json.dump(default=) or as pandas.to_json(default_handler=) to
-    serialize Python classes with as_dict(). Warning: Objects without a as_dict() method
-    are replaced with None in the serialized data.
-    """
-    try:
-        return obj.as_dict()  # all MSONable objects implement as_dict()
-    except AttributeError:
-        return None  # replace unhandled objects with None in serialized data
-        # removes e.g. non-serializable AseAtoms from M3GNet relaxation trajectories
-
-
 df_wbm = load_train_test("wbm_summary")
 df_wbm["material_id"] = df_wbm.index
diff --git a/tests/test_data.py b/tests/test_data.py
index 4fb9ce1d..410ce36b 100644
--- a/tests/test_data.py
+++ b/tests/test_data.py
@@ -107,8 +107,9 @@ def test_load_train_test_raises(tmp_path: Path) -> None:
 
     assert (
         str(exc_info.value)
-        == f"Unexpected version='invalid-version'. Must be one of {figshare_versions}."
+        == f"Unexpected {version=}. Must be one of {figshare_versions}."
     )
+    assert os.listdir(tmp_path) == [], "cache_dir should be empty"
 
 
 def test_load_train_test_doc_str() -> None:
@@ -123,60 +124,58 @@ def test_load_train_test_doc_str() -> None:
     assert os.path.isdir(f"{ROOT}/site/src/routes/{route}")
 
 
-# TODO skip this test if offline
-# @pytest.mark.skipif(online, reason="requires internet connection")
-@pytest.mark.parametrize("version", [figshare_versions[-1]])
-def test_load_train_test_no_mock_mp_refs(
-    version: str, capsys: CaptureFixture[str], tmp_path: Path
-) -> None:
-    # this function runs the download from Figshare for real hence takes some time and
-    # requires being online
-    file_key = "mp_elemental_ref_entries"
-    df = load_train_test(file_key, version=version, cache_dir=tmp_path)
-    assert df.shape == (5, 89)
-
-    stdout, stderr = capsys.readouterr()
-    assert stderr == ""
-    rel_path = getattr(type(DATA_FILES), file_key)
-    cache_path = f"{tmp_path}/{rel_path}"
-    assert (
-        f"Downloading {file_key!r} from {figshare_urls[file_key]}\nCached "
-        f"{file_key!r} to {cache_path!r}" in stdout
-    )
-
-    # test that df loaded from cache is the same as initial df
-    pd.testing.assert_frame_equal(
-        df, load_train_test(file_key, version=version, cache_dir=tmp_path)
-    )
-    stdout, stderr = capsys.readouterr()
-    assert stderr == ""
-    assert stdout == f"Loading {file_key!r} from cached file at {cache_path!r}\n"
+wbm_summary_expected_cols = {
+    "bandgap_pbe",
+    "e_form_per_atom_mp2020_corrected",
+    "e_form_per_atom_uncorrected",
+    "e_form_per_atom_wbm",
+    "e_above_hull_wbm",
+    "formula",
+    "n_sites",
+    "uncorrected_energy",
+    "uncorrected_energy_from_cse",
+    "volume",
+    "wyckoff_spglib",
+}
 
 
 # TODO skip this test if offline
-@pytest.mark.slow
-@pytest.mark.parametrize("version", [figshare_versions[-1]])
-def test_load_train_test_no_mock_wbm_summary(
-    version: str, capsys: CaptureFixture[str], tmp_path: Path
+# @pytest.mark.skipif(online, reason="requires internet connection")
+@pytest.mark.parametrize(
+    "file_key, version, expected_shape, expected_cols",
+    [
+        ("mp_elemental_ref_entries", figshare_versions[-1], (5, 89), set()),
+        pytest.param(
+            "wbm_summary",
+            figshare_versions[-1],
+            (256963, 15),
+            wbm_summary_expected_cols,
+            marks=pytest.mark.slow,  # run pytest -m 'slow' to select this marker
+        ),
+        pytest.param(
+            # large file but needed to test loading compressed JSON from URL
+            "mp_computed_structure_entries",
+            figshare_versions[-1],
+            (154718, 1),
+            {"entry"},
+            marks=pytest.mark.slow,
+        ),
+    ],
+)
+def test_load_train_test_no_mock(
+    file_key: str,
+    version: str,
+    expected_shape: tuple[int, int],
+    expected_cols: set[str],
+    capsys: CaptureFixture[str],
+    tmp_path: Path,
 ) -> None:
-    # this function runs the download from Figshare for real hence takes some time and
+    assert os.listdir(tmp_path) == [], "cache_dir should be empty"
+    # This function runs the download from Figshare for real hence takes some time and
     # requires being online
-    file_key = "wbm_summary"
     df = load_train_test(file_key, version=version, cache_dir=tmp_path)
-    assert df.shape == (256963, 15)
-    expected_cols = {
-        "bandgap_pbe",
-        "e_form_per_atom_mp2020_corrected",
-        "e_form_per_atom_uncorrected",
-        "e_form_per_atom_wbm",
-        "e_above_hull_wbm",
-        "formula",
-        "n_sites",
-        "uncorrected_energy",
-        "uncorrected_energy_from_cse",
-        "volume",
-        "wyckoff_spglib",
-    }
+    assert len(os.listdir(tmp_path)) == 1, "cache_dir should have one file"
+    assert df.shape == expected_shape
     assert (
         set(df) >= expected_cols
     ), f"Loaded df missing columns {expected_cols - set(df)}"
@@ -184,7 +183,7 @@ def test_load_train_test_no_mock_wbm_summary(
     stdout, stderr = capsys.readouterr()
     assert stderr == ""
     rel_path = getattr(type(DATA_FILES), file_key)
-    cache_path = f"{tmp_path}/{figshare_versions[-1]}/{rel_path}"
+    cache_path = f"{tmp_path}/{rel_path}"
     assert (
         f"Downloading {file_key!r} from {figshare_urls[file_key]}\nCached "
         f"{file_key!r} to {cache_path!r}" in stdout