I believe that Drum has fooled himself about the goodness of fit of his graph by choosing to display raw values in the Y-axis instead of log values. This makes the early part of the fit look excellent because the absolute differences are tiny. But those are not the differences that matter, for fitting an exponential; it is the difference in logs. A good metric of fit would make a misfit in the early part of the curve just as consequential as one at the end.
I believe that Drum has fooled himself about the goodness of fit of his graph by choosing to display raw values in the Y-axis instead of log values. This makes the early part of the fit look excellent because the absolute differences are tiny. But those are not the differences that matter, for fitting an exponential; it is the difference in logs. A good metric of fit would make a misfit in the early part of the curve just as consequential as one at the end.