This comment is not intended to help, sorry.

Do we know how long calibration exposures will be?
Which axis is wavelength?
Have you tested several shutter mechanisms, or just one?

The choice only really matters at the limit. To the extent that we care, we would want to know the shutter motion time.

The wavelength axis was horizontal.
If I reorient the graph according to the orientation we currently have (wavelength => vertical), it would be like this( not sure about the top/bottom orientation)

This graph is totally theorical, it's only based on shutters opening/closing time I've measured.

I put the previous study for normal exposure, which includes

• shutter cuts in collimated beam, but not directly cuts on focused plane. so difference among pixels are small. (e.g. beam size for one pixel is ~300mm, detector size is ~60mm)
• shutter moves in horizontal direction, which is wavelength direction

so, I think difference on light coming time range from transient period is not large as total shutter moving time, and we can say real exposure time is 'commanded exposure' plus average of shutter open/close transient time.

As Arnaud says this is probably academic. However, it would be good to respond to Shimono-san's comment about the shutter being in a collimated beam.

The exposure time is probably most important for dark current correction and I hope that there isn't enough dark current for this to matter. If this isn't true we probably have to think about a map for the spatial structure of the exposure time.

However, I think we need to distinguish a bias (exposure time really == 0) from a 0s exposure, so I think we should go with Arnaud's proposal (once we've sorted out the real effective exposure time).

Do you need anything more on this, Arnaud? I assume you are done, but for the record I agree with you and RHL: 0s should be reserved for when no light is allowed through.

At the end, exposure time is calculated as I proposed :
exptime = (transientTime)/2 + time_shutters_fully_open with transientTime = shutter_opening_time+shutter_closing_time
For each exposure: dateobs, transientTime, exptime keywords are generated.

I'm not sure how we can measure that exptime gradient accurately, it's true that in our 1 sec flat we do see that fluxes decrease along with wavelength.
But it's could be also related to halogen spectrum itself.

For the one channel, I think the solution implemented is consistent.
An action which we could take internally is to check on Zemax if this gradient is real and to quantify it.
Kjetil and Sandrine would be a good help to investigate that.

We agreed on the 2017-05-31 ICS/SPS phonecon that the current choice – fully_open_time + transit_time/2 – is what we want..