We have looked into quantifying the delay between when a user requests an image with LMI, and when the shutter actually opens. Details can be found in the write up under the LMI link about the shutter delay. The bottom line is that:
The exposure times are as recorded to within a few hundredths of a second, based on the star streaks.
The shutter throw time in each direction is between roughly 0.1 and 0.2 seconds, meaning that there is also a temporal gradient across all the images.
Formal uncertainty on the measured time offsets are an underestimate of the true variation. The shutter throw time alone means the mid-time varies across the frame systematically by at least 0.1second (added as the systematic uncertainty below).
The shutter opens 2.05 +/- 0.06 (ran) +/- 0.1 (sys) seconds later than the UTCSTART in the image header.
The shutter closes 0.19 +/- 0.06 (ran) +/- 0.1 (sys) seconds earlier than the UTCEND in the image header.
Exposure times should be computed as:
Start time = UTCSTART + 2.05sec
End time = UTCEND - 0.19sec
Mid-time = UTCSTART + 2.05 + EXPTIME/2 or
Mid-time = [(UTCSTART + 2.05) + (UTCEND-0.19)] / 2