Participant guide

Create an account with your email address (or GitHub / Google where enabled) and confirm it via the link we send you. You can browse and join competitions straight away, but you need a verified email before your first submission.

Fill in your profile — your display name and affiliation appear on leaderboards and in the discussion. Your email address is never shown publicly.

Then pick a competition from the list. Each one has the same tabs: Overview (the task and how it's scored), Data, Leaderboard, Discussion and Rules. Once you've joined, Team and My submissions appear too.

Press Join competition and accept the rules. Joining includes one clause that every competition on HULA shares: the hosts can see your email address and may contact you about the competition — for example about verification, prizes or a workshop. This is what lets hosts actually reach their participants, and it's why we ask explicitly.

• Some competitions are invite only: joining is limited to listed email addresses or institutional domains.
• Many competitions have an entry deadline — the last moment to join. All deadlines are shown in UTC.
• Hosts can update the rules. When that happens, you'll be asked to accept the new version on the Rules tab before you continue submitting.
• You can leave a competition from the Rules tab. Your past submissions stay on record with your team.

Need to ask the organisers something privately? Use Contact the hosts on the Overview tab; they'll reply to your email directly.

Everyone competes as a team. When you join you get a solo team named after you; the leaderboard shows one entry per team. On the Team tab the team leader can rename the team and invite other participants by username or email.

Accepting an invitation merges the two teams: members and submissions move into the inviting team, which keeps its name and leader. Merges are permanent and must respect three rules, checked when the invitation is sent and again when it's accepted:

1. Before the team merger deadline (if the competition sets one) and before the competition ends.
2. The merged team can't exceed the maximum team size.
3. Combined, the teams can't have made more submissions (per track) than a single team could have made so far — the daily limit times the number of days since the start. This keeps merging from being a way around the limits.

You can leave a team at any time before the end; you'll continue in a new solo team, and the submissions you made stay with the old team.

Most competitions ask for a prediction file: a CSV with an id column and your predictions, in exactly the format of the sample submission on the Data tab. Large files are fine — upload .csv, gzipped .csv.gz, or a .zip containing a single CSV.

• Submissions are scored automatically, usually within seconds to a few minutes. If a file can't be scored you'll see a precise reason (missing or duplicate ids, unexpected columns, non-numeric values…) on My submissions.
• Each team has a daily submission limit per track, reset at 00:00 UTC, and sometimes a total cap. The submit page shows what's left.
• Competitions with several tracks (for example a single-subject and a cross-subject task) have a separate leaderboard and limit per track. If a competition has only one track you won't see the concept at all.
• Hosts may ask a few questions with each submission, such as whether you used external data. Answer honestly — it helps everyone interpret the results.

The test set is split. Your score on the public split is shown on the public leaderboard while the competition runs. The private split decides the final ranking — it's held back so the winners are the models that generalise, not the ones tuned to the public board.

Before the final selection deadline, choose up to the allowed number of submissions per track on My submissions. Only those count for the private leaderboard. If you don't choose any, your best public submissions are picked automatically, and the leaderboard marks them as auto-picked.

Rankings are deterministic: teams are ordered by the primary metric, ties are broken by the competition's tie-breaker metrics in order, and if still tied the earlier submission wins. Many competitions report several metrics; the Evaluation section explains which one ranks.

If the hosts allow it, you can keep submitting after the competition ends. Late submissions are scored so you can keep experimenting and compare against the final results, but they are never ranked and don't affect prizes or standings.

After the deadline, hosts may ask top teams to verify their results. You'll be contacted by email, and a Verification tab explains what to provide — typically code and checkpoints, uploaded as files or a link — and which submission it should reproduce. Hosts then mark the team as verified (or ask for clarification); some competitions show this as a badge on the leaderboard.

Some competitions are code competitions: instead of predictions you upload a code bundle, which we run in an isolated GPU sandbox against hidden test inputs. The predictions it writes are scored like any other submission, and the test answers never leave the platform. Unlike most platforms, network access is on by default, so you can install packages and download weights at run time.

Read the code submission guide

Everything you can do on the submit page you can script: create a personal API token in Settings and use it as a bearer token to submit files, poll their status and fetch leaderboards (where they're visible to you) as JSON or CSV. Treat tokens like passwords; revoke them any time.

Read the API reference