Competition-based labs
Read the competition’s own rules
The competition page is the source for its permitted data and models, team size, compute and submission limits, evaluation protocol and other conditions. The course lab page links to the competition and states any released checkpoint or reporting requirements. Read both before starting; ask the instructor if their instructions appear inconsistent.
Do not assume that a model, external dataset, pretrained weight or collaboration arrangement allowed in one competition is allowed in another. Follow any restrictions on sharing competition files or uploading them to AI services.
Explain the evidence
Leaderboard position alone is not evidence of understanding. Keep a record of the baseline, validation design, controlled changes, main results, error analysis and limitations. Distinguish validation evidence from the final test/leaderboard result. Explain which sources and assistance were used and be ready to reproduce the reported work.
The general course policy applies to attribution, honest reporting and evaluation integrity. Grading weights remain in the syllabus; each released assignment supplies its particular requirements.
TODO(instructor): Release remaining competition-specific instructions, including platform/data choices, team size, compute/submission limits, external-data rules, final-report requirements and private-test handling. Released competition pages remain the reference for already published conditions.