You have gotten Yisak all wrong. This is a real time forecasting challenge.

This is the rule from Zindi:

A rolling leaderboard is used for this challenge. It will be updated every week on Tuesday at 15:00 GMT, as the actual maize prices in the selected counties become available.

To begin, during the week of 17- 22 November 2025 (Week 47), you will forecast prices for the weeks of 24–29 November 2025 (Week 48) and 1–6 December 2025 (Week 50). The actual prices for these forecast weeks will be released after they occur, with agriBORA providing the truth data on the Tuesday following each week (for example, prices for 24–29 November will be shared on Tuesday 2 December 2025, and prices for 1–6 December on Tuesday 9 December 2025). Your predictions will be scored against these true values as soon as the corresponding data becomes available, and the leaderboard will then be updated.

After each leaderboard update, you will have the opportunity to retrain or refine your model using all data released up to that point, then submit forecasts for the next two upcoming weeks. For example, in Week 49 you will forecast Week 50 and Week 51, in Week 50 you will forecast Week 51 and Week 52, and so on. This two-week-ahead rolling cycle will continue throughout the competition.

At the close of the challenge, your final submission in Week 52 (22–27 December 2025) will contain forecasts for Week 1 (29 December 2025–3 January 2026) and Week 2 (5–10 January 2026). Once agriBORA provides the actual maize prices for these weeks, these final forecasts will be evaluated and the leaderboard will be updated to produce the final rankings.

I stand to be corrected but from what I read, only two weeks, Week 1 (29 December 2025–3 January 2026) and Week 2 (5–10 January 2026), will be used for the final leaderboard. As for the 70% 30% split, I am sure it will be conducted to create the public leaderboard and the private leaderboard. The question is "Will it be on the final two weeks or the average of all the weeks?" I think waiting for @Zindi is the best option to clarify everything.

Good Luck!

Hello @CodeJoe, as you clearly stated. What matters most at the end is the ability of your model to accurately forecast 2 weeks into the future. During the comp, you will be able to update your model as more data becomes available. However for the last week of comp you will forecast the first 2 weeks of January and the competition will close. Hence, your score is not an average of all the weeks but just the scored obtained on the first 2 weeks of January.

Thank you @AJoel