diff --git a/TODO b/TODO
index 82f37e562a901d56d9c3e7d64438177aa4a77d3d..a4d0e208bf7677baf2e111f039d059ecb9f1a339 100644
--- a/TODO
+++ b/TODO
@@ -13,6 +13,11 @@
 Ideas
 =====
 
+- Auto-expire
+
+- Do print() and plot() from testing code count?
+  Make a setting.
+
 - Session reset UI
 
 - Allow just viewing flow without any possibility for submission
diff --git a/cfrunpy/cfrunpy_backend.py b/cfrunpy/cfrunpy_backend.py
index aab583eb3e72a06ec94af7a30de061e042db953d..ddcbbb429d8e344cfa89f30c122e68c57c0e14fe 100644
--- a/cfrunpy/cfrunpy_backend.py
+++ b/cfrunpy/cfrunpy_backend.py
@@ -152,6 +152,55 @@ class Feedback:
         self.set_points(points)
         raise GradingComplete()
 
+    def _check_numpy_array_base(self, name, ref, data):
+        import numpy as np
+        assert isinstance(ref, np.ndarray)
+
+        if not isinstance(data, np.ndarray):
+            self.finish(0, "'%s' is not a numpy array" % name)
+
+        if ref.shape != data.shape:
+            self.finish(
+                    0, "'%s' does not have correct shape--"
+                    "got: '%s', expected: '%s'" % (
+                        name, data.shape, ref.shape))
+
+        if ref.dtype.kind != data.dtype.kind:
+            self.finish(
+                    0, "'%s' does not have correct data type--"
+                    "got: '%s', expected: '%s'" % (
+                        name, data.dtype.kind, ref.dtype.kind))
+
+    def check_numpy_array_allclose(self, name, ref, data, accuracy_critical=True,
+            rtol=1e-05, atol=1e-08):
+        import numpy as np
+
+        self._check_numpy_array_base(name, ref, data)
+        good = np.allclose(ref, data, rtol=rtol, atol=atol)
+
+        if not good:
+            self.add_feedback("'%s' is inaccurate" % name)
+
+        if accuracy_critical and not good:
+            self.set_points(0)
+            raise GradingComplete()
+
+        return good
+
+    def check_list(self, name, ref, data, entry_type=None):
+        assert isinstance(ref, list)
+        if not isinstance(data, list):
+            self.finish(0, "'%s' is not a list" % name)
+
+        if len(ref) != len(data):
+            self.finish(0, "'%s' has the wrong length--expected %d, got %d"
+              % (name, len(ref), len(list)))
+
+        if entry_type is not None:
+            for i, entry in enumerate(data):
+                if not isinstance(entry, entry_type):
+                    self.finish(0, "'%s[i]' has the wrong type" % (name, i))
+
 
 def run_code(result, run_req):
     # {{{ compile code
diff --git a/course/page/code.py b/course/page/code.py
index d13681abbb5949d33e0f220cc8840a4c2ef6fb59..9e7e8e31e1b4c489aa5a4bd677df07f10ceb8651 100644
--- a/course/page/code.py
+++ b/course/page/code.py
@@ -351,7 +351,7 @@ class PythonCodeQuestion(PageBaseWithTitle, PageBaseWithValue):
     * ``GradingComplete``: An exception class that can be raised to indicated
       that the grading code has concluded.
 
-    * ``feedback``: A class instance with three methods::
+    * ``feedback``: A class instance with the following interface::
 
           feedback.set_points(0.5) # 0<=points<=1 (usually)
           feedback.add_feedback("This was wrong")
@@ -359,6 +359,11 @@ class PythonCodeQuestion(PageBaseWithTitle, PageBaseWithValue):
           # combines the above two and raises GradingComplete
           feedback.finish(0, "This was wrong")
 
+          feedback.check_numpy_array_allclose(name, ref, data,
+              accuracy_critical=True, rtol=1e-5, atol=1e-8)
+
+          feedback.check_list(name, ref, data, entry_type=None)
+
     * ``data_files``: A dictionary mapping file names from :attr:`data_files`
       to :class:`bytes` instances with that file's contents.