Competition rules and terms

1. What a competition is

A Square 1 AI competition is a contest of skill. Entrants submit predictions against withheld data and, where the competition says so, a written solution report. Standings are decided by published criteria (section 5) with no element of chance. There is no fee to enter and nothing to buy.

2. Who may enter

Anyone aged 18 or over with a Square 1 AI student account, one entry per person per competition. Square 1 AI staff, contractors, judges and members of their households may take part but are not eligible for medals or prizes. Where the law of your country prohibits the competition or the prize, or where sanctions prevent us from paying you, you may enter for the record but are not eligible for a prize. Team entries are not offered in Season 1.

3. The data

Every dataset we distribute is either a public-domain government work or released under a licence that permits the use we make of it, including commercial use. Each competition's data card names the source, the licence, the retrieval date and checksum, the split rule, the changes we made and the fields we withheld. You receive the data under the source licence and these terms. You may not obtain the withheld test targets from any source, distribute the derived files, or use external data the data card does not permit. The withheld split is never distributed; retrieving it by any means is disqualification.

4. How you submit

Predictions are CSV files scored on the public split within the daily limit shown on the competition page; you nominate up to the stated number for final scoring, and if you nominate none your best public score's file is used. Where a competition requires a solution report, it is written on the competition page under the five headings shown there, is editable until close, and is graded as it stands at close. Reports must be your own account of your own work; a link to the code that regenerates your predictions is required for medal and prize eligibility.

5. How you are judged

At close your nominated submissions are scored on the withheld private split; your rank there becomes metric points. Where a report is required, three independent, blind judges grade it against the published rubric — problem framing and impact, data understanding and validation, method and rigour, error analysis and limitations, reproducibility and clarity — each criterion on four levels; the level is the median of the three, and any spread of more than one level is decided by a human. Final points combine the two in the proportion printed on the competition page (Season 1: 70% metric, 30% report). Medals require beating the published baseline and filing a report. The rubric, the judge prompts, the model identifiers and settings, the calibration statistics and the public/private rank correlation are published with the results.

The top ten finishers and a random 5% are audited: their code is run from the training data and must regenerate their predictions within the stated tolerance; their report is read by a human; their conduct is checked against these terms. An entry that cannot be reproduced, or that breached these terms, is removed from the standings and the reason recorded. Results are published once, after the audit, and are final subject to section 8.

6. Prizes

Where a competition offers prizes, they go to the eligible entrants ranked first, second and third on the final standings after the audit. Prize amounts are stated on the season page from the announced reveal date; until then a competition page says that prizes exist and when the amounts will be announced. Prizes are paid by bank transfer or PayPal in Australian dollars (or the equivalent in US dollars at the day's rate) within 30 days of results, after identity verification. Taxes, bank charges and any reporting are the winner's responsibility. Prizes cannot be substituted or transferred. If a winner is ineligible, cannot be verified or cannot be paid, the prize passes to the next eligible rank.

7. Conduct

The following are breaches and result in disqualification, recorded on the entry: obtaining or using the withheld test targets; sharing submission files with another entrant (identical files are detected); entering under more than one account; including text in a report or submission intended to influence the automated grader; misrepresenting the declared AI-use mode; misrepresenting authorship of the work. You may discuss approaches and share code publicly during a competition; you may not share a submission file.

8. Appeals

You may appeal your result within seven days of publication by writing to tech@square1ai.com with your handle and the competition. A human re-grades from scratch without seeing the original scores. The number of appeals and their outcomes are published. An upheld appeal re-issues standings, medals and, where applicable, prizes. Decisions after appeal are final.

9. Your work and our use of it

You keep all intellectual property in your code and your report. By entering you grant Square 1 AI a worldwide, non-exclusive, royalty-free licence to publish your report, your result and your leaderboard entry under your handle, and to use anonymised submissions and reports to calibrate our graders. Your code is seen only for the audit unless you publish it yourself. Your handle is public; your name is not, unless you choose to publish it on your profile.

10. Changes, retraction and liability

We may correct a competition page, extend a deadline, or pause grading, and will say so on the page. If a dataset's provenance fails after results, the results stand but the competition is marked retracted and removed from season standings, and the reason is published. The competition is provided as-is; to the extent the law allows, Square 1 AI's liability to an entrant is limited to the prize the entrant was entitled to.

11. Governing law

These terms are governed by the laws of New South Wales, Australia. Nothing in them removes rights you have under the Australian Consumer Law or the consumer law of your own country that cannot be excluded.