Submitting your work¶
One repository, pushed before the deadline.
Submission is deliberately simple. There is no upload form and no archive to send: your team's GitHub repository is the submission, and its history is part of what gets graded.
The four rules¶
The test that decides the Execution points¶
A stranger clones your repository, follows the README, and everything runs.
From the raw data to your final number, with no undocumented steps and nothing that exists only on your laptop.
Before you submit, do it for real: clone your own repository into a new folder, follow your README as written, and run everything. Most lost Execution points are found in those ten minutes.
What the repository contains¶
The full list is in each project brief. Every project expects at least:
README.md |
What this is, how to run it, and where you used AI |
PROBLEM.md |
The question, the target and its decisions, the metric, the baseline |
MODEL_CARD.md |
What the model does, how well, where it fails |
data/ |
The script that downloads the data. Never the data itself |
src/, notebooks/ |
The code that rebuilds everything, and the story of how you got there |
The repo conventions page has the layout and a .gitignore to start from.
Deadlines and lateness¶
Deadlines are in each project brief, announced when the project opens. If something real happens, ask for an extension before the deadline rather than explaining afterwards.
Before you push for the last time¶
- A fresh clone runs from the README alone
- No data files, no secrets, no absolute paths from your machine
- Seeds set, so the numbers come out the same way twice
-
PROBLEM.mdandMODEL_CARD.mdare written and current - The README declares any use of AI
- Your peer review of another team is delivered
- Every member has commits, and can explain any line of the submission
See also: all projects · rubrics · repo conventions