diff --git a/docker/requirements-full.txt b/docker/requirements-full.txt
index 68e332d..5340f2c 100644
--- a/docker/requirements-full.txt
+++ b/docker/requirements-full.txt
@@ -10,6 +10,7 @@ numpy==2.5.4
 openpyxl==3.1.5
 pandas==3.1.0
 Pillow==13.0.0
+plot-serializer==0.2.3
 PuLP==4.2.0
 Pygments==2.21.0
 pytest==9.2.0
diff --git a/docker/run.sh b/docker/run.sh
index 2fbc49a..9d7aa4a 100644
--- a/docker/run.sh
+++ b/docker/run.sh
@@ -10,14 +10,15 @@ if [ ! -d /grade ]; then
 fi
 
 # the parent directory containing everything about this grading job
-export JOB_DIR='/grade'
+JOB_DIR='/grade'
 # the job subdirectories
 STUDENT_DIR=$JOB_DIR'/student'
 TEST_DIR=$JOB_DIR'/tests'
 OUT_DIR=$JOB_DIR'/results'
+DATA_DIR=$JOB_DIR'/data'
 
 # where we will copy everything
-export MERGE_DIR=$JOB_DIR'/run'
+MERGE_DIR=$JOB_DIR'/run'
 
 # now set up the stuff so that our run.sh can work
 mkdir $MERGE_DIR
@@ -33,6 +34,21 @@ mv "$STUDENT_DIR" "$MERGE_DIR/test_student/student"
 mv $TEST_DIR/* $MERGE_DIR/test_student/
 mv $MERGE_DIR/test_student/test_student.py $MERGE_DIR/test_student.py
 
+# PrairieLearn writes data.json to the data directory, and pytest is told to read it from there.
+# It can hold correct_answers, so it's for the grader only. It must be owned by root with
+# owner-only modes: student code keeps root's supplementary groups, and its uid may match the owner
+# of files mounted from the host.
+protect_data_json() {
+    chown -R root:root "$DATA_DIR" || return 1
+    chmod 700 "$DATA_DIR" && chmod 600 "$DATA_DIR/data.json"
+}
+
+if [ -f "$DATA_DIR/data.json" ] && ! protect_data_json; then
+    echo '{"succeeded": false, "score": 0.0, "message": "The autograder could not protect its grading data. Please contact course staff and have them check the logs for this submission."}' > "$OUT_DIR/results.json"
+    echo "ERROR: could not protect $DATA_DIR/data.json"
+    exit 1
+fi
+
 # Do not allow ag user to modify, rename, or delete any existing files
 #chmod -R 755 "$MERGE_DIR"
 #chmod 1777 "$MERGE_DIR"
@@ -45,7 +61,9 @@ echo "[run] starting autograder"
 
 # run the autograder as a limited user called ag
 # Use --worker-username to drop privileges to the 'ag' user in the student code sandbox
-uv run pytest -p prairielearn-grader --color=no --log-cli-level=DEBUG --worker-username ag "$MERGE_DIR"
+# Pass plugin options with '=': pytest's first pass over the arguments, before plugin options are
+# registered, would treat a separate value as a test path.
+uv run pytest -p prairielearn-grader --color=no --log-cli-level=DEBUG --worker-username=ag --data-json-path="$DATA_DIR/data.json" "$MERGE_DIR"
 
 # TODO change the default output name
 mv "$MERGE_DIR/autograder_results.json" "$OUT_DIR/results.json"
diff --git a/quick_start.md b/quick_start.md
index d045a30..d4e6d29 100644
--- a/quick_start.md
+++ b/quick_start.md
@@ -264,7 +264,7 @@ result = sandbox.query_function("slow_computation", data, query_timeout=2.0)
 
 **Note**: Student code must define the queried symbols, and return values must be JSON-serializable.
 Supported types include: `int`, `float`, `str`, `list`, `dict`, `bool`, `None`, numpy arrays, pandas DataFrames,
-and matplotlib figures.
+and plots built with `MatplotlibSerializer` (see [Testing Matplotlib Plots](#testing-matplotlib-plots)).
 
 ## Partial Credit and Feedback
 
@@ -603,6 +603,8 @@ def generate(data):
 
 These parameters are automatically passed to the autograder via `/grade/data/data.json` and enforced during student code execution.
 
+In the grader images, `run.sh` passes this path to pytest with `--data-json-path` and makes the file readable only by root, so student code can't open it. The file can contain `correct_answers`. Without that option, for example when running locally, the fixture reads `data.json` next to your test file. The sandbox still receives `params` through `__data_params`, so don't rely on values in `params` staying hidden from student code.
+
 #### Testing Import/Builtin Restrictions Locally
 
 To test your restrictions in the test scenarios:
@@ -950,9 +952,27 @@ since students can learn from reading full tracebacks.
 
 ### Testing Matplotlib Plots
 
-The grader supports automatic serialization and deserialization of matplotlib figures:
+To grade a plot, student code builds it with `MatplotlibSerializer` from the `plot-serializer` package and returns the serializer. The test receives a matplotlib `Figure`.
+
+`plot-serializer` is included in the `:latest` image but not in `:lite`. Returning a plain matplotlib `Figure` from student code isn't supported.
+
+```python
+# Student code
+from plot_serializer.matplotlib.serializer import MatplotlibSerializer
+
+
+def create_plot(data):
+    serializer = MatplotlibSerializer()
+    fig, ax = serializer.subplots()
+    ax.plot(data)
+    ax.set_xlabel("Time")
+    ax.set_ylabel("Value")
+    ax.set_title("Data Visualization")
+    return serializer
+```
 
 ```python
+# Test code
 import matplotlib
 matplotlib.use("Agg")  # Non-GUI backend for server environments
 
@@ -965,7 +985,7 @@ from pytest_prairielearn_grader.fixture import StudentFixture
 @pytest.mark.grading_data(name="Plot Test", points=5)
 def test_student_plot(sandbox: StudentFixture) -> None:
     """Test that student creates a correct plot."""
-    # Student function returns a matplotlib figure
+    # Student function returns a MatplotlibSerializer, which arrives here as a Figure
     plot = sandbox.query_function("create_plot", data)
 
     assert isinstance(plot, Figure)
@@ -984,7 +1004,7 @@ def test_student_plot(sandbox: StudentFixture) -> None:
 
 The grader automatically serializes/deserializes:
 
-- Matplotlib figures
+- Plots built with `MatplotlibSerializer` (received as a `Figure`)
 - NumPy arrays
 - Pandas DataFrames
 - Standard Python types (int, float, str, list, dict, bool, None)
@@ -1019,7 +1039,7 @@ Two image variants are published. Pick the smallest one that includes what your
 
 | Tag       | Size (approx.) | Includes                                                                                                                                                                                 | Best for                                                                                        |
 | --------- | -------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------- |
-| `:latest` | ~400–500 MB    | Full scientific stack: `numpy`, `pandas`, `scipy`, `matplotlib` + `matplotcheck2`, `seaborn`, `scikit-learn`, `sympy`, `networkx`, `nltk`, `nbformat`, and more | Any course that needs scientific computing, data science, plotting, or Jupyter notebook grading |
+| `:latest` | ~400–500 MB    | Full scientific stack: `numpy`, `pandas`, `scipy`, `matplotlib` + `matplotcheck2` + `plot-serializer`, `seaborn`, `scikit-learn`, `sympy`, `networkx`, `nltk`, `nbformat`, and more | Any course that needs scientific computing, data science, plotting, or Jupyter notebook grading |
 | `:lite`   | ~60–80 MB      | Grader core only: `pytest-prairielearn-grader`, `Pygments`, `beautifulsoup4`, `defusedxml`, `requests`                                                                                   | Introductory CS courses where student code uses only the Python standard library                |
 
 To use the lite image, change the `image` field:
diff --git a/src/pytest_prairielearn_grader/plugin.py b/src/pytest_prairielearn_grader/plugin.py
index 00b15c5..81b96da 100644
--- a/src/pytest_prairielearn_grader/plugin.py
+++ b/src/pytest_prairielearn_grader/plugin.py
@@ -67,19 +67,30 @@ def get_datadir(test_module: ModuleType) -> Path | None:
     return data_dir
 
 
+def _find_data_json(request: pytest.FixtureRequest) -> Path | None:
+    """Prefer the file named by --data-json-path, and otherwise use data.json next to the test file."""
+    data_json_path = request.config.getoption("--data-json-path")
+    if data_json_path is not None and Path(data_json_path).is_file():
+        return Path(data_json_path)
+
+    datadir = get_datadir(request.module)
+    return None if datadir is None else datadir / "data.json"
+
+
 @pytest.fixture(scope="module")
 def data_json(request: pytest.FixtureRequest) -> dict[str, Any] | None:
     try:
-        datadir = get_datadir(request.module)
-        assert datadir is not None
-        data_file = datadir / "data.json"
+        data_file = _find_data_json(request)
+        if data_file is None:
+            logger.debug("No location to load data.json from")
+            return None
         logger.debug(f"Loading data.json from {data_file}")
         return json.loads(data_file.read_text(encoding="utf-8"))
-    except Exception as e:
-        logger.debug(f"No data.json found or failed to load: {e}")
-
-    # If the data file is not found or cannot be read, return None
-    return None
+    except FileNotFoundError:
+        logger.debug("No data.json found")
+        return None
+    except (OSError, ValueError) as e:
+        raise RuntimeError(f"Could not load data.json: {e}") from e
 
     # if datadir is None or not datadir.is_dir():
     #     raise ValueError(f"Data directory '{datadir}' not found or is not a directory.")
@@ -677,6 +688,12 @@ def pytest_addoption(parser: pytest.Parser) -> None:
     """
     group = parser.getgroup("pytest_pl_autograder", "Prairielearn Python Autograder Options")
 
+    group.addoption(
+        "--data-json-path",
+        action="store",
+        default=None,
+        help="Path to the data.json file for this grading job. If it doesn't exist, data.json next to the test file is used.",
+    )
     group.addoption(
         "--worker-username",
         action="store",
diff --git a/tests/scenario_root/test_invalid_data_json/__init__.py b/tests/scenario_root/test_invalid_data_json/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/scenario_root/test_invalid_data_json/data.json b/tests/scenario_root/test_invalid_data_json/data.json
new file mode 100644
index 0000000..42e2ff7
--- /dev/null
+++ b/tests/scenario_root/test_invalid_data_json/data.json
@@ -0,0 +1 @@
+{"params": 
diff --git a/tests/scenario_root/test_invalid_data_json/expected_outcome.json b/tests/scenario_root/test_invalid_data_json/expected_outcome.json
new file mode 100644
index 0000000..2762715
--- /dev/null
+++ b/tests/scenario_root/test_invalid_data_json/expected_outcome.json
@@ -0,0 +1,15 @@
+{
+  "expected_data_object": {
+    "score": 0.0,
+    "tests": [
+      {
+        "max_points": 1,
+        "points_frac": 0.0,
+        "points": 0.0,
+        "test_id": "test_invalid_data_json.py::test_query[student_code]",
+        "outcome": "error",
+        "message": "Could not load data.json: Expecting value"
+      }
+    ]
+  }
+}
diff --git a/tests/scenario_root/test_invalid_data_json/scenario.py b/tests/scenario_root/test_invalid_data_json/scenario.py
new file mode 100644
index 0000000..1950ffa
--- /dev/null
+++ b/tests/scenario_root/test_invalid_data_json/scenario.py
@@ -0,0 +1,5 @@
+from pytest_prairielearn_grader.fixture import StudentFixture
+
+
+def test_query(sandbox: StudentFixture) -> None:
+    assert sandbox.query("ANSWER") == 42
diff --git a/tests/scenario_root/test_invalid_data_json/student_code.py b/tests/scenario_root/test_invalid_data_json/student_code.py
new file mode 100644
index 0000000..b0f8b2d
--- /dev/null
+++ b/tests/scenario_root/test_invalid_data_json/student_code.py
@@ -0,0 +1 @@
+ANSWER = 42