A second fit also looks good within the plot region - but as David
points out: these models go below zero at times outside the longest
time data:
Model 2:
rate = 10.85* exp( -0.0079*time) - 0.0275*time.
The data contains a saw tooth profile due to the unequal arms of the
spinner. I accumulated all beat times, and discarded every second value
of beat before inverting to a rate measure.
Either way, the new data from bc seems to show a much bigger proportion
of linear decay, and a smaller portion of exponential decay. Possibly a
too strongly clamped spinner, or a greasy bearing that could use a
solvent wash? :-)