How do values from model report relate to the prediction profiler trends?

[#1]

When analyzing the results from a DoE, I have noticed that the values from the model report for a given factor do not make intuitive sense with the trends in the prediction profiler. For example, some factors have a value of "0" in the Model Report > Total Sensitivity, but then the prediction profiler shows a trend (bell curve, for example) implying that there is a clear impact on the response value as the factor is changed. Why would the line shape in the prediction profiler not be a flat line for a factor that has a "0" total sensitivity?

Please let me know where the gap in my understanding is. Thank you. 

[#2]

Which modeling platform are you using? Is it the Gaussian Process platform? Can you share a picture of the "values from the model report?" Which options in the Profiler did you use?

[#3]

It is the Gaussian process model from a fast flexible filling design. This is a mixture, so all the factors add to 1 for a given experiment. The profiler is set to "maximize desirability".

In the attached picture the names of the factors are redacted to numbers. For example, for factor "3" the total sensitivity is "0", but the profiler shows an interesting shape indicating that there is a clear ideal value to maximize the response. 

[#4]

Hi Mark,

Any thoughts on my response with the attached picture?

Thanks

[#5]

I do not understand how the parameters for all but the first two components are zero and yet the profiler shows profiles that are not flat. Did you request the nugget parameter estimation in the launch dialog's lower left corner?