The empty cell in the panel/panellist performance is usually because there is no variation in your data. This happens when an assessor gives the same average score to all samples, where the average score is taken across all replicate measurements given by an assessor to one sample. So, it is not necessary that all the individual replicate scores are the same, just that the average product scores are the same. Mathematically the problem is a division by zero error, because the assessor panel performance measure depends on the standard deviation of the assessor's average scores, a calculation that is not possible if that standard deviation is zero.
You are more likely to experience empty cells when there are only two samples, since the fewer samples there are, the greater chance that all means will be the same. At the panellist level, agreement is measured through the correlation coefficient, and note that it is not recommended to compute correlation coefficients with only two samples since the coefficients are always -1, +1, or undefined owing to no variation. We recommend that at least 5 samples are necessary to get robust diagnostics based on the correlation coefficient.
More generally, when running panel/panellist performance, the more samples there are, the more valuable the results.