diff --git a/4-manufacturing/lessons/2-check-fruit-from-device/README.md b/4-manufacturing/lessons/2-check-fruit-from-device/README.md
index b8228d95..46767307 100644
--- a/4-manufacturing/lessons/2-check-fruit-from-device/README.md
+++ b/4-manufacturing/lessons/2-check-fruit-from-device/README.md
@@ -137,12 +137,22 @@ To improve the model, you can retrain it using the images captured from the IoT
 
 ## 🚀 Challenge
 
+How much does image resolution or lighting affect the prediction?
+
+Try changing the resolution of the images in your device code and see if it makes a difference to the quality of the images. Also try changing lighting.
+
+If you were to create a production device to sell to farms or factories, how would you ensure it gives consistent results all the time?
+
 ## Post-lecture quiz
 
 [Post-lecture quiz](https://brave-island-0b7c7f50f.azurestaticapps.net/quiz/32)
 
 ## Review & Self Study
 
+You trained your custom vision model using the portal. This relies on having images available - and in the real world you may not be able to get training data that matches what the camera on your device captures. You can work round this by training directly from your device using the training API, to train a model using images captured from your IoT device.
+
+* Read up on the training API in the [Using the Custom Vision SDK quickstart](https://docs.microsoft.com/azure/cognitive-services/custom-vision-service/quickstarts/image-classification?tabs=visual-studio&pivots=programming-language-python&WT.mc_id=academic-17441-jabenn)
+
 ## Assignment
 
-[](assignment.md)
+[Respond to classification results](assignment.md)
diff --git a/4-manufacturing/lessons/2-check-fruit-from-device/assignment.md b/4-manufacturing/lessons/2-check-fruit-from-device/assignment.md
index da157d5c..560bc554 100644
--- a/4-manufacturing/lessons/2-check-fruit-from-device/assignment.md
+++ b/4-manufacturing/lessons/2-check-fruit-from-device/assignment.md
@@ -1,9 +1,13 @@
-#
+# Respond to classification results
 
 ## Instructions
 
+Your device has classified images, and has the values for the predictions. Your device could use this information to do something - it could sent it to IoT Hub for processing by other systems, or it could control an actuator such as an LED to light up when the fruit is unripe.
+
+Add code to your device to respond in a way of your choosing - either send data to an IoT Hub, control an actuator, or combine the two and send data to an IoT Hub with some serverless code that determines if the fruit is ripe or not and sends back a command to control an actuator.
+
 ## Rubric
 
 | Criteria | Exemplary | Adequate | Needs Improvement |
 | -------- | --------- | -------- | ----------------- |
-| |  |  |  |
+| Respond to predictions | Was able to implement a response to predictions that works consistently with predictions of the same value. | Was able to implement a response that is not dependant on the predictions, such as just sending raw data to an IoT Hub | Was unable to program the device to respond to the predictions |