Mars Rover Competition Exposes Three Hidden Flaws in Data Reasoning

A competitor in a Kaggle-style challenge to reconstruct missing Mars surface pressure readings from NASA's Perseverance rover documented three critical reasoning errors discovered just before the competition closed. The contest required predicting data from sols 201–300, a period outside the training range of sols 1–100, making standard cross-validation structurally blind to the target regime. Over six weeks, the competitor relied heavily on the public leaderboard to tune key model parameters, a method that later proved problematic. A self-designed diagnostic test to identify which rows the public leaderboard scored returned a null result, but the result was null by design — every perturbed row had already fallen in the private set. The episode illustrates how self-review can fail when the errors lie in the underlying assumptions rather than in the code itself.