We hope you’re managing to work smoothly with the video data. We’ve made a few important updates to help you get the most out of this challenge:

Prediction window updated

This challenge is tough — we’ve adjusted the prediction window to 5 minutes into the future. You will now have the following structure: Training data → Test input → 2-minute embargo (operational lag) → 5-minute test output.

Inference data clarified

The inference dataset is now explicitly labelled as test_input, so you know exactly what to use for inference.

Clarifying the “No Backpropagation” rule

As stated in the challenge design, backpropagation should not be used in training or inference in a way that uses future information to predict the past. Your solution should operate in real time, meaning:

• Each minute must be predicted sequentially.
• You cannot use data from minute N+1 to predict minute N.
• You cannot use a training segment that follows a test period to inform or adjust your model during inference.

Backpropagation within a training loop (i.e. updating model weights during normal training) is, of course, allowed — just keep the real-time deployment context in mind.

This means you should not use future data to predict the pass as this would not be possible in the real world.

Data sharing and licensing update

Videos may now be uploaded to Kaggle, and the dataset license has been updated to CC BY 4.0.

These updates are designed to keep the competition fair, transparent, and fun for everyone.

If you’ve already started working, please re-download the updated files on Zindi (note the video data on the bucket has NOT changed), review the latest challenge description before resubmitting your solution.

Thank you for your patience and all the great feedback so far — and good luck!