Your team's expedition data has come back incomplete, messy, and mixed in with material that doesn't belong. Across three fast-paced stages, you'll dig through what's been recovered, figure out what's useful, sort and organize it, and get it ready so it can be used to build something later. You don't need any deep machine learning knowledge for this — just good observation and logical thinking.

• Stages run in order and cannot be skipped, repeated, or reset once closed.
• All software, code, and data needed will be provided on-site. No outside tools, code, datasets, or AI assistance are permitted
• Each stage has a fixed time limit. Work is auto-submitted when time runs out, whether complete or not.

It's time to put the prepared data to work. Across two final stages, your team will polish what you've prepared, then build a simple model using ready-made, easy-to-use tools. This is where preparation meets performance: the sharper your groundwork from Round 1, the more room you'll have to improve here. Expect this round to move fast and reward teams who iterate smart rather than iterate often.

• Decisions and outputs from earlier stages carry forward and directly affect later stages and rounds — there is no "starting over."
• Final standings are based on model performance, evaluated against data not shown to teams during the event.
• No collusion between teams, sharing of solutions, or interference with another team's work or submissions.
• All organizer decisions on scoring and disputes are final.