All of the data. I wasn't sure if the color of a boat would be a lurking variable so alternated use of
each color so that I could plot by color if the data supported and effect by color.

Picking the right interval of marbles allowed for some multi-tasking which was important as some of these boats were afloat for quite a while.

So again, I removed boats one and two, here are the scatter plot's with that data removed.
Y=-6.0072x + 916.32 , Rsquared = 0.4123

Y = 1031.5e(-0.0016x), Rsquared = 0.7721

You can see that the exponential regression is a better fit (square of the correlation 0.77 vs 0.41 of the linear regression)

Here are the plots with the two outliers removed.
Which helps to better fit the linear regression line (Rsquared moves from 0.41 to 0.65) but does little to help the exponential regression (Rsquared moves from 0.77 to 0.78)

Andys Analysis