A result you can cite.Season 1: four problems for the planet.
Predict data you never see, get scored the moment you submit, climb a public leaderboard, then find out where you really stand on the private split at close. Results are published once and are final: a rank, a percentile, a medal if you earned one, and whether you beat the baseline.
4 contests open · next close 29 Nov 2026 · free with a student account · the brief and the leaderboard need no account
Four problems for the planet.Four verticals, one grading standard.
Four real problems, four verticals, one grading standard. Cash prizes for the top three in each. Each one is predictions plus a solution report, marked the same way.
- Climate tech · Data scienceMAE, lower wins Estimate the sun Given the weather that happened, how much solar power came onto Great Britain's grid in each half-hour? The grid balances on this number. Cash prizes for the top three. 700.405baseline27 days to go
- Climate health · Data scienceMAE, lower wins Breathe Predict the daily PM2.5 concentration at 250 monitoring sites across the United States from the weather, the season and the site. Cash prizes for the top three. 3.401baseline27 days to go
- ESG · Machine learningRMSLE, lower wins The emissions gap Most power plants never report their emissions. Estimate a plant's annual CO2e from what is public about it, fuel, capacity, country, location, so the gap can be priced. Cash prizes for the top three. 1.173baseline27 days to go
- Sports · Data scienceMAE, lower wins Call the game Predict the home team's winning margin in an NBA game from the two teams' form and ratings before tip-off. Cash prizes for the top three. 10.619baseline27 days to go
70% the private-split rank, 30% a solution report graded against a published rubric by three blind judges; the top ten audited. Rules and terms.
Eight weeks open.One results day.
Results are published once, and they are final. The last 48 hours before close are a quiet period: the public board freezes for viewers.
- Mon5 OctSeason opens. Data and briefs go live.Open now
- Mon2 NovMid-point. The prize pool is announced.
- Sun29 NovEntries close at 23:59 UTC. Reports and selections are frozen.
- Mon7 DecResults: private standings, medals and the marking pack.
- Mon14 DecAppeals close, seven days after results.
Two contests to practise on.Predictions only, no report.
Same engine, same public and private split, scored the moment you submit. All 100 points come from the metric.
- Playground · Data scienceRMSE, lower wins Predict monthly rent A playground regression contest on a synthetic rental dataset. Beat the mean, then beat each other. 748.564baselineClosed 12 Oct 2026
- Playground · Machine learningAUC, higher wins Subscription churn risk Predict which subscribers cancel in the next 30 days. 0.763baselineClosed 12 Oct 2026
The same four steps.Every time.
The public leaderboard is feedback while you work; the private split is the result.
Get the data
The training set, a sample submission and a note on every column. The brief is open to anyone; the download needs a student account.
Submit predictions
A CSV of your predictions for the test rows, scored the moment it lands. Five a day, resetting at midnight UTC.
Climb the public board
While it runs, 30% of the test rows rank you in public. You choose the two submissions that count at close; the rest are practice.
The private split decides
At close the other 70% is scored once. Rank, percentile, medal and whether you beat the baseline are published together, final, and written to your record.
100 points, two parts.Your rank first, then your report.
The Square1 Ai grading standard, the same for all four Season 1 problems and printed in full on every competition page, so you can mark your own work before you file it.
70 points from your rank on the private split. Your rank there becomes your points: first place earns the full 70, last place earns 0, and everyone between is spaced evenly by rank. A score that does not beat the baseline earns 0, whatever its rank.
30 points from your report. Five criteria, equal weight, 6 points each, each marked at one of four levels.
points = 70 × (1 − (rank − 1) ÷ (entrants − 1))For example, 3rd of 21: 70 × (1 − 2 ÷ 20) = 63 points. Ranks, not raw scores, so one runaway entry cannot squash the rest of the field.
Playground contests have no report: all 100 points come from the metric, on the same rank rule.
| Criterion | 0Absent0 pts | 1Developing2 pts | 2Solid4 pts | 3Exemplary6 pts |
|---|---|---|---|---|
| Problem framing and impactWho uses this estimate, what decision changes if it is better, and what an error costs them. | No statement of who uses the estimate or why. | Names the use but not the decision it changes. | States the decision, the user and what a given error costs them. | As 2, and quantifies the value of the improvement over the baseline in the user’s terms. |
| Data understanding and validationA validation design that mirrors the test split, with leakage risks named and handled. | No validation design, or validates on the public leaderboard. | Holds out data but the split can leak: random rows in a time series, or the same entity on both sides. | A split that mirrors the test split; leakage risks named and handled. | As 2, with evidence the local validation tracks the public score, and a data finding that changed the approach. |
| Method and rigourChoices justified, compared against both pinned baselines, failures reported honestly. | No description, or a method with no justification. | A reasonable method described, no comparison. | Choices justified, compared against both pinned baselines, at least one thing that failed reported honestly. | As 2, with an ablation or sensitivity check that shows which parts matter. |
| Error analysis and limitationsWhere the model is wrong and why, and what would go wrong if someone deployed it. | No error analysis at all. | Says errors exist, no pattern. | Shows where the model fails (segments, conditions) and gives a plausible reason. | As 2, and states the deployment risks: who is harmed if it is over-trusted, what would have to be monitored. |
| Reproducibility and clarityCode, environment and steps a reader could run; prose that is clear and within length. | No code, or steps that cannot be followed. | Code linked but steps incomplete or environment unstated. | Code, environment and steps that a reader could run; prose is clear and within length. | As 2, with fixed seeds, a one-command run, and results that match the report. |
Word count, formatting, confidence and tone earn nothing. A level is earned by meeting its descriptor in full.
Blind first.
Links, emails, handles and your display name are removed from the report before any judge reads it. The code link goes to the audit, never to a judge.
Three judges, the middle mark.
Three judges mark blind and independently, each citing the descriptor it matched. Each criterion takes the middle mark of the three.
A person when they disagree.
If two judges are more than one level apart on a criterion, a person marks it instead. Text aimed at the grader scores zero.
The top ten are audited.
We re-run your code from train.csv and the result has to match. Judge prompts, model ids and calibration figures are published with the results.
One result,published once.
Scored against rows you never saw, ranked among everyone who entered, and kept. Nobody can grade it kindly.
- Top 10%
- Gold
- Next 20%
- Silver
- Next 30%
- Bronze
Shares of the field, cumulative, each at least one place. No medals in a field under five. Every medal needs a score that beats the baseline, and in Season 1 a filed report.
- A percentile
Where you finished in the field, from the private split. Ties share the better one.
- The baseline
Every contest ships one, named, so a score means something on day one. Beating it is the first line of your result.
- A medal, if earned
By the bands above, from the private rank.
- On your record
Published once and never edited, then shown on your public record, where an employer can open it.
| # | Entrant | Public | Score |
|---|---|---|---|
| 1 | Entrant A | #3 | 0.812 |
| 2 | Entrant B | #1 | 0.809 |
| 3 | Entrant C | #2 | 0.801 |
| 4 | Entrant D | #4 | 0.788 |
| 5 | Entrant E | #7 | 0.781 |
Private split · AUC, higher wins · 0 entrants · published once, final
Free to enter.Your account is your entry.
Read any brief and watch any public leaderboard without signing in. Downloading the data and entering need a free student account, so the result has a name to go on.
- 1Register
A free student account. It is what your score and your report are filed under.
- 2Join a contest
Pick a leaderboard handle, fixed once you join, and accept the rules. Entry is solo for now: one entry per person.
- 3Get the data
train.csv, test.csv and sample_submission.csv, with the source, licence and split rule on the data card.
- 4Upload predictions
A CSV with the id column and your prediction, one row per test id. Five scored uploads a day, resetting at 00:00 UTC. A file that fails the checks is rejected with the reason and does not use one.
- 5Pick your finals, file your report
Select up to two submissions for final scoring; pick none and your best public score is used. Season 1 contests also take a solution report: 600 to 1,500 words under five headings, a code link and your declared AI use.
Fair data, fair play.
The short version. Each competition page prints its own rules, and the full terms cover eligibility, judging, prizes and appeals.
Read the full terms- One entry, soloOne entry per person; your handle is fixed once you join. One account per person.
- Only the data we shipNo external data beyond what the data card lists, and never the withheld targets from any source. Looking them up is disqualification.
- Share ideas, not filesSharing code or ideas publicly is fine; sharing a submission file is not. Identical files are detected.
- Write for peopleText in your report aimed at the grader scores zero.
- The quiet periodThe final 48 hours: the public board freezes for viewers while you still see your own scores.
- Final, onceResults are published once, after the audit, and are final subject to appeal. Seven days to appeal.
The brief is open. The data needs an account.
Season 1 closes on Sunday 29 November at 23:59 UTC. Entry is free.