User submissions are evaluated by comparing their Submission CSV to the ground truth Solution CSV with respect to a chosen Scoring Metric, for example the Root-mean-squared Error (RMSE) metric.

Each Scoring Metric will expect the Solution and Submission CSVs to have certain columns, defined in a header row. For example, RMSE expects an Id column identifying the row and an Expected column specifying the true value which competitors are attempting to predict. Your Solution CSV's header row does not need to use the exact column names, you will be able to map your column names onto the Metric's expected columns.

To ensure that competitors cannot game the system, the rows in the solution file are sampled into Public and Private rows. Public scores are shown on the public leaderboard, while only the private scores are used to determine the competition winner. The sampling can be done manually by including a Usage column with values of Public, Private, or Ignored. Otherwise, rows will automatically be sampled into these buckets.