The rules say:
For each building and test period, it is allowed for the algorithm to use any form of model learned from previous data. On the other hand, participants are not supposed to attempt to get additional external information about the building sites. Also, when forecasting values for a given time period, participants are also not allowed to use given data from the following period (no interpolation is allowed).
Only algorithms that can be fully replicated and shown to only use past data to create the forecasts will be considered eligible for the prize money.
Are we able to mark a submission as eligible/ineligible? For instance my first submission used future data but that has been superseded by a past-data-only submission. A sensible hypothetical scenario is if one’s best submission used future data and then was corrected to only use past data (and hence probably get a lower score) - how could that lower score be correctly counted? It would also be good if the leaderboard was only populated with likely to be valid entries.